Log

Append-only. Each entry: ## [YYYY-MM-DD] <operation> | <title> followed by a short body. operation is one of ingest, query, lint.

[2026-08-05] setup | Wiki initialized

Repo and wiki scaffold created via /add-karpathy-llm-wiki. No sources ingested yet.

[2026-08-05] ingest | Digital Product Passports & the Data Commons (SINE/ECDF/Systemiq, 2023)

Ingested sources/whitepapers/2023-04-12-sine-whitepaper-final.pdf (shared by Martin Pompéry). Converted via Marker with --disable_ocr (this container’s Marker install lacks the llama-server binary its OCR backend needs; fine here since the PDF is digital/text-bearing, not scanned).

New pages: papers/sine-dpp-data-commons-2023.md (summary), concepts/data-commons.md, concepts/digital-product-passport.md, organization/sine-foundation.md, people-partners/einstein-center-digital-future.md, people-partners/systemiq.md. Index updated.

Flagged gap: this paper builds on an earlier SINE Data Commons whitepaper (with Deutsche Bank, Systemiq, TLGG) present as sources/whitepapers/2022-04-25-data-commons-final.pdf but not yet ingested — concepts/data-commons.md currently only reflects the framework’s application to DPPs, not its own foundational content.

[2026-08-05] ingest | Governing the Commons (Ostrom, 1990)

Ingested sources/whitepapers/2026-08-05-ostrom1990-governing-the-commons.pdf (shared by Martin Pompéry). Converted via marker_single (chunked by page range; see prior session notes for the font/SSL/OOM workarounds). Read directly: Ch. 1 (lines 1–413, the commons-dilemma games and the “two options” critique) and Ch. 3’s design-principles section (lines 905–1,064); read via three parallel subagent extractions: Ch. 2 (framework: CPR situation, rule typology, three puzzles — lines 414–659), Ch. 4 (institutional change, lines 1,064–1,379), Ch. 5 (failures/fragilities, lines 1,380–~1,699), Ch. 6 (framework-vs-model epistemology, situational variables, fullest “two options” statement — lines 1,731–2,007). Notes and references (~lines 2,008–3,484) not read.

Per Martin’s explicit redirection during this ingest (“stärker auf Details des Frameworks… wissenschaftliche Methodik… ‘zwei Optionen’”), this ingest emphasizes the analytical framework and methodology over case narratives, which are kept as pointers rather than full write-ups.

New pages: papers/ostrom1990-governing-the-commons.md (source summary), people-partners/elinor-ostrom.md (author entity page). Substantially rewritten: concepts/ostrom-commons-governance.md (now grounded in the primary text — CPR situation definition, 3-level rule typology, the three puzzles of supply/commitment/monitoring, the 8 design principles, the framework-vs-model epistemology, the comparative case-study method, the Table 5.2 validity test, and a full treatment of the “two options” critique with the Games 1–5 formalization and the Sugden/Rolph examples). Updated concepts/data-commons.md: replaced the ⟨Inference⟩ tag on “clear group boundaries” with a direct citation to design principle 1.

Note: this ingest ran concurrently with another in-progress ingest of the 2022 Data Commons whitepaper in this same wiki (new papers/, people-partners/, and concepts/ pages for Deutsche Bank, TLGG, WBCSD, PACT, ESG reporting) — cross-references above assume those pages exist.

[2026-08-05] ingest | Data Commons whitepaper (SINE/Deutsche Bank/Systemiq/TLGG, 2022)

Ingested sources/whitepapers/2022-04-25-data-commons-final.pdf (shared by Martin Pompéry; URL initially arrived Slack-truncated, resolved via sine.foundation/datacommons). Converted via marker_single --disable_ocr (same font-path/SSL/llama-server workarounds as the other 2026-08-05 ingests this wiki has hit; this PDF is digital/text-bearing).

Closed the gap flagged in the 2023-ingest log entry above: this is the original Data Commons whitepaper the DPP paper builds on.

New pages: papers/sine-data-commons-2022.md (summary), concepts/pact-pathfinder-network.md, concepts/esg-reporting-financial-sector.md, people-partners/deutsche-bank.md, people-partners/tlgg-consulting.md, people-partners/wbcsd.md. Updated concepts/data-commons.md (added the three core values — Reciprocity, Security, Openness — and the centralized/blockchain/distributed architecture choice, previously missing), concepts/digital-product-passport.md (added the earlier three-value-levels framing and the EU Battery Passport regulatory anchor), organization/sine-foundation.md (added Tell Münzing, the PACT involvement), people-partners/systemiq.md (added named 2022 authors). Index updated.

Discussed with Martin before writing: he redirected away from treating Ostrom’s principles as background “Unterbau” for Data Commons, toward Ostrom’s school of thought as what SINE’s standardization work is a first derivation of, and the frame SINE itself governs its outputs by — captured in concepts/ostrom-commons-governance.md, which by the time this entry was written had already been substantially rewritten by the concurrent Ostrom-primary-source ingest above; this entry’s contribution to that page was superseded there rather than lost.

Note: ran concurrently with the Ostrom Governing the Commons ingest logged above, in the same wiki, without prior coordination between the two sessions. Both landed cleanly (mutually additive; one direct edit overlap on concepts/data-commons.md’s “clear group boundaries” bullet, resolved by the other session in our favor — its version citing design principle 1 directly is kept). Flagging so a lint pass double-checks cross-references across both ingests’ pages before this is trusted as fully consistent.

[2026-08-05] query | Mozilla Foundation strategic comparison

Ad hoc web research (not a sources/-file ingest) requested for an internal strategic comparison document: Mozilla Foundation’s mission split from Mozilla Corporation, org size (~120 Foundation staff vs. ~3,000 Corporation staff), revenue concentration risk (75-86% of Mozilla’s group-wide revenue from Google search royalties, vs. the Foundation’s own more-diversified donor base), public-visibility tactics (MozFest, Privacy Not Included, small-grants/fellowship ecosystem-building), governance/board composition, and criticisms (2020 Corporation layoffs ~250 staff, 2024 Foundation layoffs ~30% dissolving its advocacy division, mission-drift criticism).

New page: market-competitors/mozilla-foundation.md (new “Market & competitors” category entry — first page in that section). Every claim cited inline to its source URL rather than to a sources/ file, since this was live web research rather than an ingested document; flagged on the page itself so future readers don’t mistake it for the usual ingest-from-sources/ provenance. Index updated.

Note: if this comparison work continues, consider snapshotting the cited URLs into sources/ (e.g. sources/comparisons/) so citations don’t rot, and/or ingesting SINE’s own financials the same way for a like-for-like comparison.

[2026-08-05] ingest | Hypothesenpapier: SINE strategische Standortbestimmung

Requested by Martin Pompéry: a deep-crawl of sine.foundation plus competitor research across two clusters — (1) data-exchange standards in the broad sense, (2) Tech-for-Good NGOs / comparable governance-and-tech NGOs — synthesized into an internal Hypothesenpapier for SINE’s leadership (to provoke discussion/further research, not to state conclusions), and delivered as a PDF.

New raw source files (live web/browser research by subagents, not downloaded documents — flagged as such, same provenance caveat as the Mozilla Foundation query entry above): sources/research/2026-08-05-website-crawl-sine-foundation.md, sources/research/2026-08-05-competitor-data-exchange-standards.md, sources/research/2026-08-05-competitor-tech-for-good-ngos.md.

New “Market & competitors” pages (15): market-competitors/gaia-x.md, idsa.md, catena-x.md, global-battery-alliance.md, gs1.md, ocean-protocol.md, commercial-clean-room-vendors.md, linux-foundation-openssf.md, mydata-global.md, wikimedia-creative-commons.md, stiftung-neue-verantwortung.md, algorithmwatch.md, bertelsmann-stiftung.md, ada-lovelace-institute.md, open-data-institute.md.

New “Strategy & positioning” category (first page in it): strategy/hypothesenpapier-strategische-standortbestimmung.md — ten hypotheses (H1-H10), a competitor comparison table for both clusters, and open questions for leadership. Cross-links into concepts/data-commons.md, concepts/ostrom-commons-governance.md, and the existing market-competitors/mozilla-foundation.md page. Index updated (new Strategy & positioning entry; Market & competitors split into two sub-groups, data-exchange standards and Tech-for-Good NGOs).

Data-quality notes carried into the Hypothesenpapier itself: several figures conflict across sources and are presented as ranges rather than resolved (Gaia-X and IDSA membership counts in particular); the “€200 million” figure commonly attached to Catena-X actually belongs to a separate Gaia-X federal funding line — corrected on the Catena-X and Gaia-X pages. The Ocean Protocol page describes an active, unresolved federal lawsuit (SDNY, filed Nov 2025) — flagged as time-sensitive.

PDF export and delivery to Martin via sine-strategy-works still pending as of this entry; a wiki commit is due once that’s done.

[2026-08-08] setup | Coaching stack: Workflow gestartet

Startet den Aufbau eines Coaching-Wissens-Layers (“Karl”-Stack) für einen später zu entwerfenden Coaching-Agenten.

• Bewusste Abweichung vom wiki-Skill: dessen Ingest-Schritt 2 verlangt, die Kernaussagen vor dem Schreiben mit dem Nutzer zu besprechen. In einem Hintergrund-Workflow ist das nicht möglich. Ersatz: die Takeaways erscheinen live im Workflow-Log und in den log.md-Einträgen der einzelnen Ingest-Schritte; der Operator kann an jeder Schrittgrenze pausieren, um zu diskutieren.
• Datumskonvention für das neue Verzeichnis sources/coaching/: Publikationsdatum des Dokuments, wo bekannt — sonst Beschaffungsdatum.
• Sprachregel (provisorisch, s. wiki/coaching-stack/open-questions.md Punkt a): englische Ordner-/Dateinamen, deutscher Fließtext. Tell entscheidet final.

[2026-08-08] ingest | Cynefin (Dave Snowden)

Vorab-Reparatur: die drei am 2026-08-08 neu angelegten/geänderten Dateien (wiki/index.md, wiki/log.md, wiki/coaching-stack/open-questions.md) hatten Umlaute/Eszett transliteriert geschrieben (fuer, spaeter, Fliesstext u. a.) statt mit echten Sonderzeichen. Nur die Schreibweise korrigiert, keine inhaltliche Änderung.

Ingestiert sources/research/2026-08-08-cynefin-question-logic.md — eine Live-Web-Recherche (Subagent, mehrere Suchen/Abrufe), keine heruntergeladene Publikation, entsprechend auf der neuen Seite gekennzeichnet.

Zusätzlich beschafft (frei zugänglich, echte PDFs per file-Check verifiziert): Kurtz & Snowden (2003) „The new dynamics of strategy: Sense-making in a complex and complicated world” (IBM Systems Journal) unter sources/coaching/2003-06-01-kurtz-snowden-new-dynamics-of-strategy.pdf, und Snowden & Boone (2007) „A Leader’s Framework for Decision Making” (HBR) unter sources/coaching/2007-11-01-snowden-boone-leaders-framework-decision-making.pdf. marker_single-Konvertierung beider PDFs zu Markdown-Zwillingen wurde gestartet, lief zum Zeitpunkt dieses Eintrags aber noch (erster echter Marker-Lauf auf diesem Host, inkl. Modellgewichte-Download — deutlich länger als die üblichen 1–3 Minuten). Die neue Seite stützt sich inhaltlich auf die Recherchenotiz; ein Abgleich/Vertiefungs-Pass gegen die beiden Primärtexte, sobald die Markdown-Zwillinge vorliegen, ist als offener Punkt auf der Seite selbst vermerkt.

Neue Seite: wiki/coaching-stack/cynefin.md — fünf Domänen nach Constraint-Typ, Fragelogik je Verbsequenz (mit der Kernthese, warum analytische Fragen im komplexen/chaotischen Bereich schädlich sind), Signature Moves (Ritual Dissent, SenseMaker, safe-to-fail probes, Liminalität/Aporetik, Estuarine Mapping), Annahmen (epistemisch/ ontologisch/axiologisch), Kritik & Grenzen (Michael C. Jackson/Critical Systems Thinking als tiefste Kritik, plus fünf weitere Einwände), Bezug zum Karl-Stack (Framework-vs-Modell-Parallele zu concepts/ostrom-commons-governance.md), Quellenlage, Siehe auch. Index aktualisiert (Kategorie „Coaching stack”).

[2026-08-08] ingest | GROW (Whitmore) und Inner Game (Gallwey)

Neue Seite wiki/coaching-stack/grow-inner-game.md (eine gemeinsame Seite fuer beide Schulen, da genealogisch durchgaengig verknuepft: Gallweys Self1/Self2- Dualismus und die Formel P = p - i als Haltungsfundament; Whitmores vierphasige GROW-Session-Choreographie (Goal/Reality/Options/Will) als Struktur darauf). Verarbeitete Quellen: die bereits vorhandene Recherchenotiz sources/research/2026-08-08-grow-inner-game.md sowie die Marker-Zwillinge von Panchal & Riddell (2020, GROWS-Modell) und Müller & Kotte (2020, “Of SMART, GROW and goals gone wild” — Dateiname irrefuehrend, s. Notiz im Wiki-Eintrag). Ergaenzt um GROWS als Will-Phasen-Erweiterung (HAPA-basierte Selbstwirksamkeitsfragen), die duenne/widerspruechliche Evidenzbasis (Mueller & Kotte: kein signifikanter Effekt in allen vier quasi-experimentellen Studien; Scoular & Linleys “sacred cow”-Befund), Grants REGROW-Kritik und den in Mueller & Kotte explizit benannten blinden Fleck zum organisationalen Kontext. Der ungeloeste Widerspruch zur GROW-Genealogie (McKinsey-Akronym vs. est/Graham-Alexander) wird dargestellt, nicht aufgeloest. wiki/index.md unter “Coaching stack” ergaenzt.

[2026-08-08] ingest | Co-Active Coaching (Whitworth, Kimsey-House, Kimsey-House, Sandahl / CTI)

Neue Seite wiki/coaching-stack/co-active-coaching.md. Verarbeitete Quellen: die Recherchenotiz sources/research/2026-08-08-co-active-coaching.md sowie drei Marker-Zwillinge, jeweils vollständig aus der .md-Fassung gelesen: der 34-seitige freie Auszug aus Co-Active Coaching, 3. Aufl. 2011 (Kapitel „Fundamentals” + „Relationship” — vier Cornerstones inkl. NCRW-Postulat, drei Prinzipien Fulfillment/Balance/Process, fünf Kontexte, Designed Alliance, wörtliche Eröffnungsfragen), das Original-CTI-Schulungsmerkblatt „The Co-Active® Model” (aktuelle Cornerstones-Reihenfolge), und das akademische Konzeptpapier „Coaching as a social process” (Brookes, nennt Co-Active explizit als Beispielfall für Individualisierungskritik, Machtdynamik-Typologie nach Fatien Diochon & Lovelace 2015). Herausgearbeitet: die fünf Kontexte als gleichzeitig aktive „constellation” statt Phasenfolge (Kontrast zu GROW), Curiosity als eigentliche Fragelogik („nicht führend, nicht an die Antwort geknüpft”), eine im Buch selbst nicht benannte Spannung zwischen NCRW-Autorität der Coachee-Person und der „Speaking the Truth”- Passage (Coach sieht manchmal klarer als die Coachee-Person), sowie vier Kritiksträngen (Machtasymmetrie trotz dyadischer Gleichheitsrhetorik vs. dreiseitigem Organisationskontrakt; Individualisierung struktureller Probleme/Neoliberalismuskritik; dünne empirische Basis des NCRW-Postulats, nur als Blogquelle mit Vorbehalt; Kommerzialisierung/Zertifizierungsökonomie, nur als ungeprüfte Suchsynthese). ICF-Kernkompetenzen und Crowe-Associates- Zitate wurden aus der Recherchenotiz übernommen, nicht separat aus PDF/HTML gegengeprüft (beide ohne Markdown-Zwilling). wiki/index.md unter „Coaching stack” ergänzt.

[2026-08-08] ingest | Cynefin-Seite gegen Primärtexte vertieft (Kurtz & Snowden 2003, Snowden & Boone 2007)

Vertiefungs-Pass gegen die beiden lokal vorliegenden Markdown-Zwillinge: Kurtz, C. F. & Snowden, D. J. (2003), „The new dynamics of strategy”, IBM Systems Journal 42(3) (sources/coaching/2003-06-01-kurtz-snowden-new-dynamics-of-strategy/) und Snowden, D. J. & Boone, M. E. (2007), „A Leader’s Framework for Decision Making”, HBR 85(11) (sources/coaching/2007-11-01-snowden-boone-leaders-framework-decision-making/). Reine Lokalarbeit, keine Websuche.

Hauptbefund: Terminologie-Drift über drei Fassungen. 2003 benennt die Domänen known/knowable/complex/chaos/disorder und lehnt die spätere simple/complicated-Beschriftung im Abschnitt “The apple-orange problem” explizit ab (zwei ungleiche Achsen: Erkennbarkeit vs. Interaktion). 2007 (vier Jahre später, gleicher Erstautor) verwendet genau diese zurückgewiesene Terminologie (simple/complicated/complex/chaotic/disorder) ohne die Zwei-Achsen-Begründung zu wiederholen. Die heutige Fassung (clear/complicated/complex/chaotic + aporetisches Zentrum) ist damit die dritte Namensgebung in derselben Autorenlinie.

Zweiter Befund (Widerspruch, dargestellt statt aufgelöst): die zentrale Domäne heißt 2003/2007 “disorder” und wird durchweg negativ gefasst (Symptom von Gruppenkonflikt, Ziel ist Verkleinerung/Verlassen). Die heutige Fassung (Recherchenotiz) wertet dasselbe Zentrum als “aporetisches Zentrum” um — eine bewusst gehaltene, produktive Konfusion. Der Zerlegungsmechanismus selbst (“in Teile zerlegen, dann in Domänen verschieben”) ist in 2007 bereits wörtlich vorhanden; nur die Bewertung hat sich seither umgekehrt.

Weitere Ergänzungen aus den Primärtexten: die vier Sense/Categorise/ Analyse/Probe/Act-Sequenzen sind in beiden Papers wortnah bestätigt (kein Drift); die bislang fehlende Sense-Making-vs-Kategorisierungs- Unterscheidung (2003) als Begründung, warum keine Cynefin-Domäne als “Zieldomäne” gelten darf; primärquellenbasierte Begründung, warum analytische Fragen im Complex-Bereich schädlich sind (retrospective coherence, analysis paralysis, “hindsight does not lead to foresight”); 2003s zehnteilige Bewegungstypologie (Imposition, Exploration, Swarming u. a.) als Vorläufer der heutigen Loosen/Tighten-Constraints-Regel; die 2003-Kontextualisierungsübung (Narrative Database, Alternative History, Hexagon-Clustering) als Vorläufer von SenseMaker. Die vierteilige Constraint-Typologie (fix/regulierend/ermöglichend/abwesend) findet in keinem der beiden Texte eine volle Entsprechung — 2007 liefert nur eine Vorstufe (System zwingt Akteure / keine Zwänge / wechselseitiger Zwang).

Abschnitt “Quellenlage” korrigiert: beschrieb bislang den Abgleich noch als offenen Punkt und behauptete an anderer Stelle bereits eine abgeschlossene Auswertung — beides jetzt auf den tatsächlichen Belegstand gebracht (Primärtexte tragen jetzt den Großteil der Seite, Recherchenotiz bleibt einzige Quelle für post-2010er-Vokabular und den Kritik-Abschnitt).

index.md-Eintrag (Kategorie “Coaching stack”) entsprechend nachgezogen.

[2026-08-08] ingest | Heidelberger Schule und zirkuläres Fragen (Mailänder Tradition, Karl Tomm)

Ingest von fünf Quellen aus sources/coaching/ (Tomm Interventive Interviewing Teil I + II, Ryan & Carr 2000, PSDP-Praxistool 2019, von Foerster 1991) plus der Recherchenotiz sources/research/2026-08-08-systemic-questioning-heidelberg.md. Neue Seite: wiki/coaching-stack/systemic-questioning-heidelberg.md.

Tomm I und II wurden vollständig im Original gelesen und gegen die Recherchenotiz abgeglichen (Auftrag: Original hat Vorrang bei Abweichung). Befund: die Notiz ist über weite Strecken präzise, aber die Übertragung von Tomms acht benannten Kategorien reflexiver Fragen ist über die Kette Tomm (1987, 8 Kategorien) → PSDP-Tool (2019, 6 in der eigenen Tabelle / 7 im Fließtext, Kategorie „normative-comparison” fällt weg) → Recherchenotiz (2026, behauptet „sechs”, zählt aber sieben auf) verlustig geworden — auf der neuen Seite mit der korrekten Acht-Kategorien-Liste aus dem Original korrigiert. Zusätzlich aus dem Original ergänzt, was die Notiz ausließ: der CMM-Hierarchie-Reversal-Mechanismus (strange loop/charmed loop) als eigentlicher Wirkmechanismus reflexiver Fragen, sowie Tomms vier Neutralitäts-Varianten und die caring/obligatory-Circularity-Unterscheidung.

Anker-Kontingent (bis zu 4 Zusatzsuchen/6 Zusatzabrufe) für de Shazers Clues (1988, Wunderfrage/Skalierungsfrage) ausgeschöpft: kein freier Volltext gefunden (Internet Archive access-restricted, kein HathiTrust Full View); dabei eine Wortlaut-Variante zwischen zwei Sekundärquellen entdeckt (unterschiedliche Schlusssätze der Wunderfrage kursieren als „das” Zitat). Passage bleibt auf der Seite ausdrücklich als schwach/sekundär belegt gekennzeichnet, nicht stillschweigend übernommen.

Zwei Entscheidungen des Beschaffungsschritts umgesetzt: (1) Bibliothekszugang zu Selvini Palazzoli et al. 1980 und Cecchin 1987/1993 als offene Frage in open-questions.md (e) eingetragen, nicht weiter recherchiert; (2) feministische/gewaltschutzorientierte Kritik an der Tradition bewusst nicht recherchiert, als benannte Lücke in „Kritik & Grenzen” der neuen Seite und als offene Frage in open-questions.md (f) vermerkt.

wiki/index.md unter „Coaching stack” aktualisiert.

[2026-08-08] ingest | Angrenzende Frageschulen (Übersicht)

Neue Seite wiki/coaching-stack/adjacent-schools.md aus sources/research/2026-08-08-adjacent-schools.md (Live-Web-Recherche, kein Einzeldokument) plus zwei bereits beschafften und mit marker_single konvertierten Primärdokumenten: Aboodi Shabi, „Ontological Coaching” (2015, sources/coaching/2013-10-01-shabi-ontological-coaching/…md) und das offizielle MINT-Handout „Four Fundamental Processes in MI” (Miller/Rollnick 2010, sources/coaching/2010-01-01-miller-rollnick-four-fundamental-processes-mi/…md). Beide Markdown-Zwillinge waren bereits vorhanden (Zwillingspflicht geprüft, kein Nachkonvertieren nötig) und wurden vollständig gelesen.

Eine Übersichtsseite für zehn Schulen (SFBT, Clean Language, Immunity to Change, Ontologisches Coaching, Motivational Interviewing, Appreciative Inquiry, Theory U plus fünf Kurzprofile), bewusst breit und flach statt als Einzelseiten. Kernbefund: durchgängiger Grundkonflikt Zukunfts- vs. Gegenwartsorientierung, den Cynefin/Snowden explizit gegen SFBT, AI und Theory U wendet („happy-clappy”, „easy to fake and easy to game”); ein zweiter Grundkonflikt zwischen individuumszentrierten Schulen (Ontologisches Coaching: Transformation „at the level of the soul”) und Cynefins interaktionszentrierter Systemsicht. Lücken (Cooperrider-Original, Scharmer-Originalformulierungen, Kegan/Lahey-Wortlaut, Groves eigene Schriften, de Shazer/Berg-Originalbücher) wurden aus der Recherchenotiz übernommen und auf der neuen Seite unter „Quellenlage” ausgewiesen statt verschwiegen.

wiki/index.md unter „Coaching stack” aktualisiert.

[2026-08-08] ingest | Warm Data / Bateson und Dark Matter Labs / Bioregioning

Zwei neue Seiten aus sources/research/2026-08-08-warm-data-dark-matter.md sowie allen dort genannten Primärdokumenten (Volltext gelesen, keine Zusammenfassung): wiki/coaching-stack/warm-data-bateson.md und wiki/coaching-stack/dark-matter-labs-bioregioning.md.

Strang 1 (Warm Data/Bateson): Nora Batesons Begriffskette Warm Data (2017, Blogpost + IBI-Booklet, Volltext) → Symmathesy (2015, ISSS-Proceedings, Volltext) → Aphanipoiesis (2021, ISSS-Proceedings, Volltext), plus Gregory Batesons Lernebenen 0–III und „difference that makes a difference” aus zwei gezielt konvertierten Essays aus Steps to an Ecology of Mind (1972, Seitenbereiche 297–328 und 468–485). Zwillingspflicht geprüft: alle Markdown-Zwillinge lagen bereits vollständig vor, keine Nachkonvertierung nötig. Provenienz-Korrektur aus der Recherchenotiz übernommen: „the pattern which connects” ist NICHT in Steps to an Ecology of Mind auffindbar (Volltextsuche negativ) — stammt laut Sekundärquellen aus Batesons späterem Mind and Nature (1979), nicht beschafft, auf der Seite als Lücke ausgewiesen.

Strang 2 (Dark Matter Labs/Bioregioning): fünf HTML-Quellen vollständig gelesen und ausgewertet (darkmatterlabs.org-Startseite, bioregions.darkmatterlabs.org, Indy Johars Substack-Essay „Existential Multipolarity”, das GitHub-publizierte xCO-Konzeptpapier, die Future-Observatory-Profilseite). Kernbefund: Meta-Target Validity Matrix (zehn Fragen) als institutionelle Entsprechung der Lernen-II-Frage nach dem Rahmen; „viability hourglass”; Bioregioning als nicht-lineare, multigenerationale (20–100+ Jahre) Prozessarchitektur. Astrophysikalische Namensherleitung von „Dark Matter Labs” blieb unverifizierbar (leere /about-Seite), als offene Lücke markiert statt übernommen.

Beide Seiten mit vollständiger Struktur (Einordnung, Fragelogik, Signature Questions, Annahmen dreiteilig, Kritik & Grenzen, Bezug zum Karl-Stack, Quellenlage, Siehe auch); wechselseitig verlinkt. wiki/index.md unter „Coaching stack” um beide Einträge ergänzt. profiles.json im Workflow-Tool um zwei schoolProfile-Einträge ergänzt (Warm Data/Bateson, Dark Matter Labs/Bioregioning).

[2026-08-08] ingest | Syntheseseite: Fragelogiken im Vergleich

Reine Lokalarbeit, keine neue Quelle beschafft. Aus den sieben bereits ingestierten Coaching-Stack-Seiten (cynefin, grow-inner-game, co-active-coaching, systemic-questioning-heidelberg, adjacent-schools, warm-data-bateson, dark-matter-labs-bioregioning) eine Syntheseseite wiki/coaching-stack/question-logic-comparison.md destilliert. Grundlage waren die Schulprofile in tools/coaching-stack/profiles.json statt eines erneuten Vollständig-Lesens der Einzelseiten; wörtliche Zitate gezielt per grep aus den Einzelseiten geholt. Vollständig gelesen wurde nur wiki/concepts/ostrom-commons-governance.md als Vorbild für die Framework-vs-Modell-Argumentation.

Inhalt: Vergleichstabelle mit 18 Zeilen (Epistemologie / Ontologie / Axiologie / charakteristischer Fragezug / blinder Fleck), dazu eine eigene Belegstärke-Tabelle in fünf Stufen — Cynefin und Heidelberger Schule stehen auf ausgewerteten Primärtexten, Warm Data und Dark Matter Labs auf Primärtexten ohne auffindbare externe Kritik, GROW überwiegend auf Sekundärliteratur (Gallwey/Whitmore nur als Bibliotheksausleihe, Grant 2011 und Scoular & Linley 2006 hinter Paywalls), Co-Active auf einem Kapitelauszug, die zwölf adjacent-schools-Zeilen bewusst breit und flach. Diese Ungleichheit ist ausdrücklich als eigener Abschnitt dargestellt, weil eine Vergleichstabelle sonst eine Gleichrangigkeit suggeriert, die die Quellenlage nicht deckt.

Acht Divergenzen dargestellt, nicht geglättet: (1) Co-Actives NCRW-Postulat gegen das Beobachterkonstrukt der Heidelberger Schule und Batesons „irrelevantes” Selbst auf Lernen-III-Ebene; (2) GROWs Zielphase und SFBT/ Theory U gegen Cynefins retrospektive Kohärenz und Warm Datas Ablehnung der Plan-Strategie-Lösung-Logik; (3) Clean Languages erreichbare Absichtslosigkeit gegen von Foerster/Cecchin und Kurtz & Snowdens „every intervention is also a diagnostic”; (4) MIs righting-reflex-Verzicht gegen Batesons Konfrontationspflicht bei self-validating Lernen-II-Prämissen; (5) Appreciative Inquiry gegen Snowden und Dark Matter Labs zur Frage, ob das Problem benannt werden darf; (6) Focusings vorsprachlicher felt sense gegen die sprachkonstitutiven Schulen; (7) Analyseeinheit Individuum/Dyade gegen Symmathesy ohne Agency gegen Bioregion; (8) Cynefins „no implied value axes” gegen Dark Matter Labs’ definitorisch eingebaute Wertsetzung.

Argumentativer Kern ist der Abschnitt „Schicht 0 und die Reflexivitätsfalle”: Das Fragehandwerk (Schicht 1) macht Schicht 0 zugänglich und bringt dabei eigene Annahmen ins Fundament ein — die Wahl des Fragesets ist damit schon eine Festlegung, bevor die erste Frage gestellt ist. Analogie zu Ostroms Framework-vs-Modell-Unterscheidung („setting a variable equal to a constant usually narrows… the range of applicability”) und zu ihrer Symmetrieannahme (dieselbe begrenzte Rationalität für Nutzer und Analysten), sowie zu Cynefins Selbstbeschreibung als „a decision support framework, not a method or model” mit bounded applicability. Ergebnis: die Reflexivität ist nicht auflösbar, nur dokumentierbar — Ort dafür ist das Annahmenregister des Karl-Stacks.

wiki/index.md unter „Coaching stack” um den Eintrag ergänzt (vor open-questions.md eingeordnet). Keine Änderung an sources/.

[2026-08-08] setup | Coaching-Agent — Entwurf und Annahmenregister-Vorlage

Reine Lokalarbeit, keine neue externe Quelle recherchiert. Grundlage waren wiki/coaching-stack/question-logic-comparison.md (vollständig gelesen) und die 18 Schulprofile in tools/coaching-stack/profiles.json.

Neue Quelldatei: sources/decisions/2026-08-08-coaching-agent-structure-decision.md — Tell Münzings Beschluss vom 2026-08-08 zur Struktur A+C+D (eigener fragender Agent, absichtliche Perspektivendiversität, Annahmenregister), zur erweiterten Fragen-Trias und zum Reasoning-Stil, als datierte Entscheidungsnotiz festgehalten, damit die Entwurfsseiten rückführbar sind. Neuer Unterordner sources/decisions/.

Zwei neue Wiki-Seiten, beide ausdrücklich als ENTWURF gekennzeichnet — es wurde KEIN Agent angelegt, keine Container-Konfiguration verändert:

• wiki/coaching-stack/coaching-agent-draft.md: Rolle (fragen statt antworten; ausdrücklich nicht beraten, nicht Fragesets unbemerkt mischen, keine Neutralität beanspruchen, keine Therapie ersetzen); Pflicht-Repertoire aus vier Schulen mit Begründung je Schule plus Adjacency-Feld; Adjacency-Prinzip (Ansage vor der Frage, markierter Linsenwechsel, stehengelassener Widerspruch), begründet über Cynefins Selbstdeklaration als „framework, not a method or model” und Ostroms Symmetrieannahme; erweiterte Fragen-Trias in fünf Fragen mit je 2–3 Beispielfragen im Wortlaut; Reasoning-Stil (Pfadabhängigkeit, Startbedingungen, Systeme zweiter/dritter Ordnung) inklusive Begründung, warum eine ergodische oder rein stochastische Lesart hier in die Irre führt — als ⟨Inference⟩ markiert, weil keine ingestierte Quelle die Ergodizität unter diesem Namen kritisiert; Pflicht zum Annahmenregister; acht offene Entwurfsfragen.
• wiki/coaching-stack/assumption-register-template.md: ausfüllbare Vorlage mit genau den sechs beschlossenen Feldern (Datum; Thema; Schule(n) und tatsächlich gestellte Fragen im Wortlaut; Freigelegtes; Bias des Frageformats selbst; Ungeprüftes) plus ein vollständig ausgefülltes, klar als fiktiv gekennzeichnetes Beispiel.

Dargestellter, nicht aufgelöster Widerspruch auf der Entwurfsseite: Warm Data bestreitet Handlungsfähigkeit einzelner Teile grundsätzlich, während der Zuschnitt eines Coaching-Agenten Handlungsfähigkeit voraussetzt.

Lücke: Die vom Beschluss genannte Neurowissenschaft ist im Stack nicht ingestiert — keine Quelle, keine Wiki-Seite; auf der Seite als erste offene Entwurfsfrage markiert. index.md (Kategorie „Coaching stack”, beide Einträge vor open-questions.md) aktualisiert. profiles.json unverändert (keine neue Schule ingestiert).

[2026-08-08] lint | Coaching stack: Abschlußprüfung

Mechanische Validierungsprüfung nach Abschluss aller Ingest-, Synthesis- und Entwurfschritte. Überprüft:

1. Index-Vollständigkeit: Alle Seiten unter wiki/coaching-stack/ haben einen Eintrag in wiki/index.md. ✓ Keine Seiten ohne Index.
2. Leere Seiten: Keine Datei unter wiki/coaching-stack/ ist unter 1 kB groß. ✓
3. PDF-Markdown-Twins: Alle PDFs unter sources/ haben entsprechende Markdown-Zwillinge. ✓
4. Kaputte relative Links: Links auf ../concepts/ostrom-commons-governance.md existieren und sind erreichbar. ✓
5. open-questions.md: Alle sechs Fragen (a–f) sind vorhanden, sauber formatiert und noch aktuell. Punkt c) zu ARIA ist als echte offene Rückfrage formuliert, ohne Vermutung. ✓
6. Git-Status: 3 modifizierte Wiki-Dateien (Index, Log, open-questions), 28 neue Dateien in sources/.

Ergebnis: Alle Prüfungen bestanden. Der Coaching-Stack-Layer ist strukturell vollständig und konsistent. Alle acht Coaching-Schulen sind mit kanonischen Seiten ingestiert, die Synthese läuft, die offenen Fragen sind dokumentiert, die Entwurfsskizze für den Coaching-Agenten liegt vor.

[2026-08-10] lint | Wiki health check across both layers (market-competitors + coaching-stack)

Full sweep of all ~40 pages under wiki/ against sources/, checking for contradictions, stale claims, orphan pages, concepts without pages, missing cross-references, and unresolved ⟨Inference⟩ claims.

Fixed directly (mechanical, low-risk):

• Added reciprocal cross-references: market-competitors/global-battery-alliance.md ↔ concepts/digital-product-passport.md, concepts/pact-pathfinder-network.md, people-partners/systemiq.md, market-competitors/gaia-x.md.
• concepts/pact-pathfinder-network.md now links to catena-x.md, global-battery-alliance.md, gs1.md.
• concepts/digital-product-passport.md’s landscape list now links to the market-competitors/ pages that didn’t exist yet when it was first written (WBCSD, GBA, Catena-X) and notes the ⟨Note⟩ is partly outdated (those three are now profiled, Battery Pass/CIRPASS still aren’t).
• strategy/hypothesenpapier-strategische-standortbestimmung.md: linked its three bare-text WBCSD/Catena-X/GS1 mentions to their wiki pages.
• Coaching-stack: added missing bidirectional links between coaching-stack/co-active-coaching.md and coaching-stack/systemic-questioning-heidelberg.md (the two poles of question-logic-comparison.md’s sharpest divergence), and from coaching-stack/cynefin.md to question-logic-comparison.md and coaching-agent-draft.md. Added inline links in open-questions.md points b/d.

Reported, not fixed (need a judgment call or new research):

• No factual contradictions or stale-superseded claims found — prior corrections (Gaia-X/Catena-X funding conflation, Ocean Protocol lawsuit, membership ranges) are already correctly reflected in light of the newer 2026-08-05/08 sources.
• people-partners/wbcsd.md is a thin placeholder despite WBCSD being named as a substantive comparator in 5+ pages — candidate for a fuller page if a dedicated WBCSD source is ever ingested.
• Concepts named repeatedly but with no page: Fraunhofer-Gesellschaft/ISST (11+ mentions across gaia-x/idsa/catena-x as the literal governance-overlap hub), Boris Otto (named individual evidencing that overlap), SFBT/de Shazer/Insoo Kim Berg (strongest candidate — three coaching pages lean on it), Werner Erhard/est (appears as two unconnected genealogy threads in grow-inner-game.md and co-active-coaching.md, possible common origin never flagged), Michael C. Jackson/Critical Systems Thinking (central critique source for Cynefin).
• coaching-stack/coaching-agent-draft.md’s ergodicity/non-ergodicity framing is the most exposed inference in the wiki — imports vocabulary from economics/physics (Ole Peters) not present in any ingested coaching source; self-flagged but worth checking against the actual ergodicity literature before it reaches production.
• Two citation-year footnotes worth a human glance (not contradictions): adjacent-schools.md flags de Shazer’s Keys to Solution in Brief Therapy (1985) as next step while systemic-questioning-heidelberg.md already cites de Shazer’s Clues (1988) for the miracle question — same author, different books, never cross-referenced on this point.

Full findings shared with Martin/Tell in Slack; no destructive changes made, all edits were additive cross-references or link insertions.

[2026-08-10] ingest | PACT Methodology v3.0 (WBCSD, 2025)

Source: sources/whitepapers/2026-08-10-pact-methodology-v3.0/pathfinder-methodology-v3.0.md (Marker-converted; delivered pre-converted by Martin Pompéry as pathfinder-methodology-v3.0.zip, no original PDF retained). Requested explicitly “im weiteren Kontext von WBCSD” — a WBCSD-authored primary source (Foreword by Dominic Waughray, EVP WBCSD), superseding the secondary/Data-Commons-only view of PACT.

New pages:

• papers/wbcsd-pact-methodology-v3-2025.md — source summary
• concepts/pact-pcf-calculation-methodology.md — cradle-to-gate scope, dual biogenic-CO2 PCF values, allocation/PCR hierarchies, biogenic/land emissions (2027 mandatory), PCF validity (3yr/10% variance), exemption cap (<3%), GWP/IPCC-AR rules
• concepts/pact-data-quality-verification.md — Primary Data Share (PDS) formula and mandatory-now status, Data Quality Ratings (DQR, mandatory from 2027), two-horizon verification roadmap, SME 2-year grace period

Updated pages:

• concepts/pact-pathfinder-network.md — expanded beyond the 2022 Data Commons framing with current scale (2,500+ companies, 40+ PACT Conformant solution providers/14 countries), split methodology/data- quality detail out to the two new concept pages
• people-partners/wbcsd.md — replaced the “landscape pointer, not yet profiled” stub with real content from the Foreword (Scope 3/CBAM framing, v3 improvements, scale claims)

Corrected during discussion with Martin: he flagged PDS as a “perspektivische Pflichtangabe” (forward-looking disclosure requirement) — re-checked against source text (Sec. 4.1, 4.2.2 ×2) and confirmed PDS carries unqualified “shall” language with no future date, unlike DQR which is explicitly and consistently deferred to “2027 onwards” (footnotes 10, 11, 73, 74, 78 all attach the same 2027 caveat to DQR/biogenic-land categories, never to PDS). Documented as a nuance on pact-data-quality-verification.md: the reporting obligation for PDS is immediate; the meaningfulness of the score is ecosystem-dependent (upstream PDS only >0% if suppliers also report) and expected to improve prospectively — both readings now live in the page rather than silently picking one.

Not ingested (flagged to Martin as reference/lookup material, not transcribed): Appendix B (existing standards table), Appendix C (GHGP LSRS alignment), Appendix D (elementary flow mapping), Appendix E (Scope 3 upstream accounting methods), Bibliography, full Appendix A glossary (spot-checked only), revision log (v2→v3 diff).

Per Martin: the wiki’s documentation task extends beyond Data Commons — PACT/WBCSD content is now maintained as a standalone technical/strategy reference, not only as a Data Commons validating case.

[2026-08-13] ingest | GIZ Tender 10036847 — ToR + Cover Letter (Data Governance in Agricultural Value Chains)

Martin shared 6 tender documents for a GIZ tender SINE wants to bid on (GIZ 10036847, “Data Governance in Agricultural Value Chains”, SAFE project). Saved all 6 originals to sources/tenders/ and converted to Markdown via Marker. Flagged to Martin first: cover letter states a submission deadline of 4 Aug 2026, 9 days before today — Martin confirmed the deadline was extended (GIZ tender portal link shared, exact new date not yet in the wiki).

Ingested the ToR and Cover Letter together (read both while checking the deadline). Wrote:

• giz-10036847-data-governance-agvc.md — full tender profile: context, 4 work packages, milestones, personnel concept + qualifications, budget/fee days, option, format requirements, award mechanics (70/30 weighting, 500-point technical cutoff), fit assessment
• giz.md — new entity page
• safe-project.md — new entity page (SAFE/DPI ecosystem)
• Cross-referenced from data-commons.md (new “Applied to: smallholder farmer data” section)

Marker note: --TableProcessor_disable_ocr needed on top of FONT_PATH/REQUESTS_CA_BUNDLE workarounds (this container’s Marker still lacks llama-server for the table-OCR fallback — same known gap as prior ingests). The Assessment Grid’s criterion-text cells came out empty/ truncated (numeric weights survived fine — used those). The List of Questions v2 PDF is genuinely table-heavy with real prose in every cell; Marker’s table extraction there was unusably garbled even with the OCR workaround, so it will be read natively (PDF, not the .md twin) when ingested next.

Fit assessment (mine, marked as inference in the tender page): strong overlap with Data Commons / Ostrom governance-design experience for WP1/ WP3; WP2 (build-and-ship an offline-first data wallet app) is a genuine capability gap — flagged for Martin/Tell to consider for staffing.

Remaining sources to ingest, in order: list of questions v2, draft contract, eligibility declaration.

[2026-08-13] ingest | GIZ Tender 10036847 — List of questions v2

Read sources/tenders/2026-08-13-giz-10036847-list-of-questions-v2.pdf natively (not via its Marker .md twin — the table extraction there dropped/garbled most cell content for this densely-tabular 4-row Q&A document; native PDF read via the Read tool got full, clean text). Added a “Questions & answers (v2, 29.07.2026) — binding” section to giz-10036847-data-governance-agvc.md: tandem Team Leader arrangement allowed, WP2b depends on both WP1 and WP3 results (not WP1 alone — ToR wording confirmed correct), no fixed travel pattern for the 2 pilot-country trips, and the WP4 capacity-building expert (NKE1) is scoped to WP4 only. GIZ states these answers are binding / integral to the tender documents.

Remaining sources to ingest, in order: draft contract, eligibility declaration.

[2026-08-13] ingest | GIZ Tender 10036847 — DRAFT Contract

Read sources/tenders/2026-08-13-giz-10036847-draft-contract/2026-08-13-giz-10036847-draft-contract.md in full (clean Marker conversion, no garbling issue this time). Added a “Draft contract — key terms” section to giz-10036847-data-governance-agvc.md: annexes order (ToR, price schedule, technical bid, current GTC, consortial declaration), contract term matching the ToR (15.10.2026–15.10.2027), settlement-by-time-statements payment mechanics, advance-payment mechanics (offset in 2 instalments against the penultimate + final interim invoices; work must start within 30 days, advance-payment request within 15 days of month-of-commencement end), quarterly interim invoicing capped at 3 interims + 1 final, VAT place-of-supply = Germany (§3a(2) UStG), and an EU-level publicity/visibility clause layered on top of GIZ’s own T&Cs (consistent with SAFE sitting under a Team Europe Initiative). Remuneration total and advance-payment amount are blank in the template, to be filled from SINE’s own financial bid — no separate cap stated beyond the ToR’s fee-day/budget structure.

Remaining source to ingest: eligibility declaration (last of the 6).

[2026-08-13] ingest | GIZ Tender 10036847 — Eligibility declaration (final of 6)

Read sources/tenders/2026-08-13-giz-10036847-eligibility-declaration/2026-08-13-giz-10036847-eligibility-declaration.md in full (clean conversion). This is a self-declaration form SINE must fill in and submit, not GIZ contract terms — added a new “Eligibility declaration — self-declaration form to submit” section to giz-10036847-data-governance-agvc.md capturing the eligibility thresholds SINE must clear: average annual turnover ≥ EUR 190,400 and ≥5 employees/managers (last 3 years); ≥3 reference projects (min. EUR 60,000 contract value each) in the relevant technical field, and ≥3 in the region “World”; standard GWB §123/124 exclusion-grounds and LkSG fine declarations; SME self-classification question; the 5/6-income rule for any individual-consultant team members; and attachment requirements (commercial register extract, consortium declaration if bidding jointly).

Flagged as an open item, not yet checked: whether SINE/partners can actually name 3+ qualifying reference projects in the required field and region — needs confirming before submission, marked as ⟨Inference⟩ on the page since this wiki has no record of SINE’s past project portfolio to verify against.

All 6 tender documents are now ingested. wiki/index.md’s Tenders entry updated to reflect complete ingest. Next step, if requested, would be drafting the actual Angebot (technical + financial bid) — not yet started.

[2026-08-13] ingest | GIZ Tender 10036847 — official Bekanntmachung (public notice)

Ingested CSX 21 - Bekanntmachung.pdf, the official UVgO public-notice publication for GIZ Tender 10036847 (Data Governance in Agricultural Value Chains) — same tender as the 6 documents already in the wiki, this is the formal legal notice rather than a tender-portal document.

Key new/confirming fact: officially confirms in writing the extended submission deadline the wiki had previously flagged as unconfirmed (per Martin Pompéry) — Angebotsfrist 21.08.2026, 12:00 Uhr. Also confirms the Aufklärungsfragen (clarification-question) deadline 18.08.2026 and the Zuschlags-/Bindefrist (award/binding period) 13.11.2026. Added contact (isabel.paul@giz.de), submission mechanics (electronic only via Vergabemarktplatz GIZ, no postal, no lots, no alternative bids), and confirmed award method (economically most favourable offer, consistent with the known 70/30 weighting).

Updated wiki/tenders/giz-10036847-data-governance-agvc.md (Status line + new “Official public notice (Bekanntmachung)” section + Sources entry) and wiki/index.md’s Tenders entry.

Process note: Marker (PDF→Markdown) failed in this environment — its font and OCR-error-detection model-server downloads hit SSL cert / spawn-timeout errors unrelated to the document. Worked around the font download via a manual curl + FONT_PATH env var, but the OCR-error-detection model server still failed to spawn in time. Since the PDF is short (2 pages) and digitally text-bearing, fell back to reading it natively rather than sinking more time into Marker — same fallback the schema reserves for scanned PDFs, applied here for an environment/infra reason instead. Noted inline in the tender page’s Sources section.

[2026-08-13] correct | GIZ Tender 10036847 — two factual errors fixed

Found while cross-checking the tender page against the sources for the dossier dossiers/2026-08-13-giz-tender-10036847-anforderungen-und-sine-fit.md:

1. Assessment-grid weighting. Project management 1.6 was recorded as 8%; the grid has 1.6.1–1.6.3 at 2% each = 6%. Only with 6% does concept block 1 sum to the 33% the grid itself states. Corrected in the “Assessment & award mechanics” section.
2. WP2b. Recorded as “documenting the data-wallet prototype”; ToR ch. 2.2.2 task b is developing the prototype (documenting is task c). Corrected in the Q&A section, where the error had blurred what GIZ’s answer 2 actually confirms.

Added an ⟨Inference⟩ note under Q&A 2: with the corrected reading, the confirmed WP2b→WP3 dependency collides with the milestone table (prototype due March 2027; the WP3 cooperative workshops that produce farmer-side results are July 2027). Basis: ToR ch. 2.2.2, ch. 2.2.4 milestone table, Q&A v2 answer 2.

No source files touched. Index unchanged (page link and summary still accurate).

[2026-08-14] ingest | LinkedIn profiles: Stefan Dierks, Michael Opitz, Tell Münzing

Three LinkedIn profile exports handed over by Martin Pompéry in Slack. Saved as immutable sources under a new sources/people/ subdirectory (2026-08-14-linkedin-<name>.pdf) and converted to Markdown twins with marker_single; ingested one at a time from the twins, with takeaways discussed in Slack per person before writing.

Marker fix (relevant for all future PDF ingests): marker_single previously failed in this container — first PermissionError on /opt/uv-tools/.../static (font download into a read-only site-packages dir), then SSLError against models.datalab.to (OneCLI proxy MITM certificate). Both are fixable from the environment; FONT_DIR alone does not work because FONT_PATH is computed at class-definition time:

export FONT_PATH=/workspace/agent/tools/marker-fonts/GoNotoCurrent-Regular.ttf \
       FONT_DIR=/workspace/agent/tools/marker-fonts \
       REQUESTS_CA_BUNDLE=/tmp/onecli-combined-ca.pem \
       CURL_CA_BUNDLE=/tmp/onecli-combined-ca.pem
marker_single <pdf> --output_format markdown --output_dir <dir>

This supersedes the note on the Tender 10036847 page that Marker’s model downloads “failed in this environment” — they fail only without these vars.

New pages: people-partners/stefan-dierks.md, people-partners/michael-opitz.md, people-partners/tell-muenzing.md.

Updated pages:

• organization/sine-foundation.md — new Chronology section from Tell Münzing’s profile (founding 10/2020; MPC carbon-benchmarking tool with Cozero 03/2021; Data Commons 04/2022; PACT for WBCSD 2023; iLEAP with the Smart Freight Centre 2025), his full role, and SINE’s own “Digital Public Goods / Collective Action Problems / digital sovereignty” framing.
• coaching-stack/grow-inner-game.md — new section “Provenienz im Auftrag”: the principal who made GROW part of the agent’s mandatory repertoire has been in leadership roles at Performance Consultants International (Whitmore’s own firm) since 2006. Recorded next to the page’s existing thin-evidence findings, not resolved against them.
• coaching-stack/open-questions.md — new (g) role of the imported people and whether HRNS is in play as an organization; (h) how to handle the GROW-provenance finding in the agent itself.
• tenders/giz-10036847-data-governance-agvc.md — new subsection on the two imported profiles that sit on this tender’s subject matter (Opitz on the farmer/implementer side incl. GIZ-internal years, Dierks on the corporate buyer side), with the explicit note that neither closes the WP2 software gap.
• index.md — People & partners split into “People” and “Partner organizations”; three new entries.

Open: Martin has not stated in which role these three are imported (advisory, partnership, tender staffing, or plain contact profiles) — the pages are written neutrally until he answers. iLEAP still has no own page; the only evidence is this profile plus a passing Hypothesenpapier mention, a Smart Freight Centre primary source is needed. Michael Opitz’s profile is unusually terse and his current status (retired / interim / still executive at HRNS) is ambiguous in the source itself.

[2026-08-14] correct | Marker: two environment defects fixed, OCR limit clarified

The Marker note in the LinkedIn-profiles entry above is superseded. It said the two failures were “fixable from the environment” and gave an export-prefixed invocation. Both defects have since been fixed in NanoClaw itself, and the note was also too optimistic about what the fix covers.

What changed:

• The font PermissionError and the models.datalab.to SSLError are gone. The image now sets FONT_DIR + FONT_PATH (both are needed — Marker builds FONT_PATH from FONT_DIR at class-definition time), and the container runner sets REQUESTS_CA_BUNDLE / CURL_CA_BUNDLE / GIT_SSL_CAINFO from the gateway’s CA bundle, because Python’s requests and git ignore the SSL_CERT_FILE the gateway already set. No export prefix anymore.
• /home/node/.cache is mounted per group from the host, so Marker’s weights and font download once instead of on every container start.

What did not change — this resolves the contradiction between the entries of 2026-08-05 (--disable_ocr needed, llama-server missing), 2026-08-13 (--TableProcessor_disable_ocr needed for tables) and 2026-08-14 (“no longer necessary”):

• --disable_ocr is still required and belongs in every call. Marker 2.0 runs OCR through a local inference server (llama-server) that this image does not ship. The LinkedIn exports converted without it only because clean digital PDFs never trigger the OCR path — that was luck of the source type, not the env fix. The wiki skill now carries the flag in its canonical command.
• Scans and OCR-fallback table extraction remain out of scope. The Tender 10036847 page’s note that Marker’s table output was unusably garbled stands; the ⟨Update 2026-08-14⟩ appended to that page overstates the fix and is corrected by this entry.

Canonical invocation:

marker_single <file>.pdf --output_format markdown --output_dir <dir>/ --disable_ocr

No source files touched. No wiki content pages changed beyond the tender-page note referenced above.

[2026-08-15] ingest | ARIA CCTI proposal (Schedule 3, draft 2, 2026-06-17)

Source saved as sources/proposals/2026-06-17-aria-ccti-proposal-draft-2.md (new sources/proposals/ subdirectory; Markdown handed over directly in chat by Martin Pompéry, no conversion needed).

New pages:

• projects/ccti.md (new wiki/projects/ section) — canonical project page: thesis, six work packages, six milestones incl. three Go/No-Go gates, facilitator role, consortium, open questions, strategic reading for SINE
• organization/aria.md — funder; Scaling Trust programme, cohort of creators, Arena. Marked ⟨Background⟩ throughout — no primary ARIA source ingested yet, this is the biggest gap left by this ingest
• concepts/epistemic-integrity-certificate.md — SINE’s WP-C primitive
• concepts/cryptographic-trust-infrastructure.md — the four-layer stack
• concepts/ai-advantage.md — the three-arm empirical comparison
• people-partners/martin-kleppmann.md — Co-PI, Cambridge
• people-partners/ink-and-switch.md, people-partners/light-squares.md — stubs, sub-contractors only

Updated: organization/sine-foundation.md (CCTI under Known involvement + source list), index.md (new Projects section + 6 entries), strategy/hypothesenpapier-strategische-standortbestimmung.md (dated Nachtrag: H3/H6 on funding concentration and the competence description are overtaken by CCTI; the paper itself left unedited).

Substantive note for follow-up: this source establishes a second centre of gravity for SINE (agentic-AI trust infrastructure, UK/ARIA-anchored) alongside the sustainability-data line (Data Commons, PACT, iLEAP, DPP). The wiki’s market/competitor coverage maps only the second. Bridge named in the source itself: outcome 4 targets supply chains, manufacturing, logistics. Open: the WP-F integration map (due 30/9/2026) requires knowing the Scaling Trust cohort, which nothing in the wiki currently does.

[2026-08-15] synthesis | Coordination failure vs. bargaining failure

Filed concepts/coordination-vs-bargaining-failure.md out of the strategy discussion with Martin Pompéry (same day, no source file — analytical page, ⟨Inference⟩-marked throughout). Content: the distinction (aligned interests + missing common knowledge vs. opposed interests + missing credible commitment), the one-question diagnostic, the classification of SINE’s own work — PACT and iLEAP as coordination successes in which the cryptography was not load-bearing, the MPC line and CCTI as bargaining — and the proposed core description “SINE makes claims believable to others without those claims having to be disclosed”, which replaces an earlier “SINE is a verification institution” framing rejected in the same conversation.

Three objections recorded on the page rather than resolved: cases are usually both shapes in sequence; “aligned interests” often disguises a distributive fight, which would undercut the flattering reading of PACT; and cryptographic monitoring may satisfy Ostrom’s fourth design principle technically while destroying its social function.

Index updated. Discussion ongoing — the open thread is whether this distinction can serve as the objective function for the thematic-field search that started this conversation.

[2026-08-15] transcript | Strategiegespräch Themenfelder (Martin Pompéry × Elinor)

Full verbatim record of the 2026-08-15 Slack strategy conversation filed as a raw source: sources/discussions/2026-08-15-strategiegespraech-themenfelder.md (new sources/discussions/ subdirectory). Requested by Martin so the conversation can be referred back to from Slack later.

Covers 10:54–13:48: the redirect from competitor research to Themenfelder; the correction of the relevance filter (transition acceleration as the north star, “Positionierung” dropped); the CCTI ingest; sectors and meta-levers, insurance as the missed entry and the re-sorting of meta-levers by function; the five objections to the externality frame; coordination vs. bargaining failure; the rate decomposition (window × hit rate × lead time); and the final pass using economic traditions as rival testable hypotheses.

Navigation aid added at the top of the file (marked as such); everything else is message text as sent. Wiki-writing tool actions are summarised as bracketed interstitials rather than reproduced.

Indexed under a new ## Discussions & transcripts section in index.md.

Open at the point the transcript ends: the four socratic questions, the proposed shift of unit of analysis from Themenfeld/sector to decision type, and whether/how the Themenfeld research gets cut. No research agents launched.

[2026-08-17] lint | Second scheduled health check — 63 pages, ARIA status stale, GIZ deadline under-flagged

Scope: all 63 wiki pages. Mechanical checks (link resolution, index coverage, inbound-link counts) run deterministically with a script; content checks (contradictions, supersession, unsupported claims, ⟨Inference⟩ review) run as three scoped subagents over the clusters ingested since the 2026-08-10 lint — CCTI/ARIA, GIZ tender + people, PACT v3.0.

Mechanical, before/after: 7 broken links → 0; 2 orphan pages → 0; index coverage complete both times (63/63).

Fixed:

• coaching-stack/open-questions.md (c) — said ARIA was “nicht öffentlich verifizierbar” and deliberately un-researched. Superseded by the 2026-08-15 CCTI ingest; rewritten as partially resolved, with the genuinely-still-open part (no primary ARIA source) kept and pointed at organization/aria.md.
• log.md — 7 links in the 2026-08-13 GIZ entries had one ../ too many (log.md sits inside wiki/, so they resolved outside the repo). Link targets corrected; no log text changed.
• strategy/hypothesenpapier-…md — Mozilla Foundation was the only row in the Tech-for-Good table written as bare text (“siehe eigene Seite”) instead of a link; that was why mozilla-foundation.md was an orphan.
• concepts/coordination-vs-bargaining-failure.md was the second orphan — now linked from projects/ccti.md, concepts/epistemic-integrity-certificate.md and the Hypothesenpapier’s 2026-08-15 Nachtrag.
• projects/ccti.md — “one of its co-founders is a named Co-PI” carried a fact the CCTI source does not state (it lists Martin Pompéry only as “Co-PI, SINE Foundation”). Reworded and the provenance marked.
• tenders/giz-10036847-…md — added a dated deadline banner; linked the fit assessment to organization/sine-foundation.md.
• concepts/safe-project.md — had no cross-reference section at all; added.
• people-partners/wbcsd.md — added the v3.0 self-description (240+ member businesses) and direct links to the two PACT sub-concept pages.
• organization/sine-foundation.md — the chronology’s “2023 — established the PACT framework” reads as PACT’s founding date; noted that PACT itself launched 2020 / COP26 2021 per v3.0, with the LinkedIn wording kept.

Represented rather than resolved: epistemic-integrity-certificate.md says PACT and the EIC share “the same structure”; coordination-vs-bargaining-failure.md puts them on opposite sides of the coordination/bargaining split. Both readings are now on the page with the tension named.

Correction to an earlier entry in this log: the 2026-08-13 Marker note recommends FONT_PATH/REQUESTS_CA_BUNDLE exports. The wiki skill now states these are configured in the image and the recipe is stale — use marker_single … --disable_ocr and nothing else.

Checked and found sound (no action): the 60 → 2,500+ company supersession from PACT v3.0 is already carried side by side with both sources on pact-pathfinder-network.md; the WBCSD “landscape pointer only” placeholder flagged by the last lint was resolved on 2026-08-10; the DPP ⟨Note⟩ is still accurate; index.md line 11’s “no primary ARIA source ingested yet” is correct as written. No contradictions or unsupported load-bearing figures found in the GIZ tender cluster — both 2026-08-13 corrections still hold.

Not fixed, needs a human: the tender’s reference-project eligibility (⟨Inference⟩, never checked against SINE’s actual project history) gates the whole bid four days before submission. Escalated to the group.

[2026-08-17] ingest | GIZ 10036847 — Konsortialkonstellation: Dossiers ins Repository, Befunde ins Wiki

Two changes at once, on Martin Pompéry’s instruction: the dossiers now live in this repository, and their findings are integrated into the wiki as pages of their own.

New repository layer — dossiers/. A fourth layer next to sources/, wiki/ and the schema, holding self-authored analysis products so they can be versioned alongside what they were derived from. Moved in under their original filenames (an earlier attempt had mis-filed them into sources/dossiers/ with renamed prefixes; that directory was removed — sources are immutable and these are not sources):

• 2026-08-13-giz-tender-10036847-anforderungen-und-sine-fit.md (+ PDF, 14 pp)
• 2026-08-14-giz-tender-10036847-konsortialszenarien.md (+ PDF, 23 pp)
• 2026-08-15-cooperation-failure-taxonomies.md
• 2026-08-06-sine-foundation-executive-summary.md (+ PDF)

Relative links inside the dossiers rewritten from ../sine-strategy/… to ../…; both GIZ PDFs re-rendered in place. ⟨Note: the dossiers carry no independent authority in this wiki. Wiki pages cite the underlying source files, not the dossier that first drew the conclusion.⟩

New pages.

• tenders/giz-10036847-konsortium.md — the bid-side companion to the tender page: the seven-party constellation, Martin’s four decisions of 2026-08-14 (client relationship yes / tandem Team Leader no / Tell as Key Expert 1 yes / KE2 later), the Team Leader candidate comparison against the ToR ch. 4 profile, the four scenarios A′/A/B/C for splitting the work packages, five formalia, the pilot-country question, the conflict-of-interest ⟨Inference⟩, and eight open/blocking items.
• organization/hanns-r-neumann-stiftung.md — HRNS as the hypothetical consortium lead: key figures, the two programmes that read as Key Expert 2 leads (ICP, coffee&climate incl. the offline-capable toolbox; aCLIMAtar with Bioversity/CIAT), the leadership handover, and an #ede-consulting section.

Updated.

• The three person pages imported on 2026-08-14 all carried “⟨Open: the role SINE intends for him was not stated at import time.⟩” — they were created by a parallel session without tender context. All three now resolved: Opitz = proposed Team Leader (availability unasked), Münzing = Key Expert 1, Dierks = origin of the constellation and fallback Team Leader. Each gained a section checking the profile against the ToR.
• tenders/giz-10036847-data-governance-agvc.md — the people section of 2026-08-14 marked superseded, with a pointer to the companion page. The old text stays as written; it is the step before the roles existed.
• index.md — HRNS entry, a new ## Dossiers section, the consortium page, and the three person entries refreshed with their now-decided roles.

Contradictions represented, not resolved: the HRNS endowment (€150 m per Daily Coffee News vs €30 m per de.wikipedia); Dierks’ start at Melitta (May 2019 per the HNEE lecturer portrait vs July 2019 per his own profile — for a CV in GIZ format the self-declaration governs).

Weakest evidence in the cluster, marked as such on both pages that use it: E.D.E. Consulting’s presence in Vietnam rests on search results only (Devex, VOV World, a business directory, one LinkedIn profile). It is load-bearing for the second-pilot-country argument and needs a human answer from HRNS.

[2026-08-17] refactor | Person page schema — Münzing as reference, Opitz/Dierks aligned

Not an ingest: no new source. A structural pass on the three person pages, following a discussion with Martin Pompéry in Slack and two decisions he made in it.

The finding that prompted it: the three person pages of 2026-08-14 have no schema of their own — they inherited LinkedIn’s gliederung (Career / Skills / Education / Publications), because all three came from one source class on one day. Consequences: “Top skills as listed” was carried as a finding when it is a LinkedIn ranking; the pages had no place for a second source class; dated decisions sat scattered in the prose with ad-hoc ⟨added …⟩ markers; and no page carried the CV fields a GIZ bid actually consumes.

Martin Pompéry’s two decisions (Slack, 2026-08-17):

1. Critical judgements and open points do belong on person pages.
2. Redundancies and tender role assessments belong in the tender context, with a link from the person page.

Applied:

• people-partners/tell-muenzing.md — rewritten as the reference implementation. New ## Status ⟨date⟩ block after the intro (principal role, tender role, what is open). SINE chronology reduced to a pointer at organization/sine-foundation.md, which holds the richer version including the PACT-date caveat — the duplication is gone, the caveat survives. Career table now strictly reverse-chronological by start date (it had LinkedIn’s order). Three marketing-copy sections condensed into one ### The other stations. New ## Formal record — what the source does and does not establish and ## Open questions. The GROW-provenance finding kept and sharpened per decision 1. Internal inconsistency fixed: the intro said PCI “since February 2009” against a section heading of “2006 – present” — he joined the firm 01/2006, took the Director role 02/2009.
• tenders/giz-10036847-konsortium.md — new section ## Key Expert 1 against the profile, receiving the assessment moved off the person page, plus the CV fields missing for the KE1 CV.
• people-partners/michael-opitz.md, people-partners/stefan-dierks.md — same ## Status block; the duplicated ## Fit against the Team Leader profile sections reduced to a summary line plus a pointer, after checking sentence by sentence that the tender page already carried the content. Dierks gained a ## Open questions section, which he did not have.

The one sentence that was not already on the tender page and therefore survived the move: the Dierks Melitta start-date contradiction (HNEE portrait May 2019 vs his own profile July 2019). It is now an open question on his page rather than a line inside a role assessment.

Deliberately not done: people-partners/ still mixes persons, organizations and intellectual sources (Ostrom, Kleppmann). Splitting it is proposed, not decided.

Open, and cheap to close: languages with levels, nationality, availability in fee days, and the four artefacts in GIZ reference format for Münzing. Mandatory CV fields, absent from every source in the wiki, four days before the 21.08.2026 deadline.

[2026-08-17] query | Gap-Analyse GIZ-Angebot 10036847 — Abgabe-Checkliste

Martin Pompéry asked for a very short gap analysis of everything still needed for the GIZ bid, each item referenced to its originating document. Answered from the wiki and re-verified against the sources; filed back as ## Abgabe-Checkliste — Gap-Analyse 2026-08-17 on wiki/tenders/giz-10036847-konsortium.md.

New findings, not previously on any page:

• Five of the nine annexes named in the cover letter (p. 3/3) are not in sources/ at all. Two of them are production-blocking: the Application requirements (which contains the mandatory GIZ CV form that ToR ch. 7 binds the CVs to, plus section 5.2 on register extracts) and the Price schedule (which the financial bid must follow exactly). Also missing: the consortium declaration form, the AVB (backstopping is priced into fee rates per AVB § 3.1), and the data-protection information. The wiki previously recorded “all 6 tender documents + Bekanntmachung ingested”, which was true and also incomplete — the six are the ones GIZ published as reading matter, not the forms to be filled in.
• Three of five CV slots are unfilled (Key Expert 2 and both expert-pool members). The pool’s per-member criteria (2.6.1–2.6.5) were on the tender page in prose but never stated as two separate CVs to be produced.
• The eligibility check ran dry for all three candidate members: a subagent sweep of wiki + sources found no turnover figure, no qualifying three-year headcount average and not a single reference project in the form the declaration’s table requires, for SINE, HRNS or E.D.E.
• Criterion 2.2.8 requires specialist EUDR knowledge in addition to the smallholder experience — not evidenced for Münzing and previously not called out as a substantive (not merely formal) KE1 gap.

Ranking recorded: reference projects → HRNS commitment → the two missing forms → KE2 + pool CVs → the 10-page concept. Commit due.

[2026-08-17] ingest | The Carbon Accounting Landscape Review (ICC, 2026)

Requested by Martin Pompéry from a truncated URL; the exact path was recovered by fetching the ICC Carbon Expert Panel page rather than guessing. Source: sources/whitepapers/2026-08-17-icc-carbon-accounting-landscape-review.pdf (863 KB), with its Markdown twin in the sibling directory of the same name. The FAQ PDF surfaced during the same lookup and is downloaded and converted but not yet ingested — it is a separate source and stays queued.

Two source-handling exceptions, both worth recording.

1. Marker garbled every multi-column table. Table 1 was dropped entirely; Tables 1b, 2, 3, 4 and A1 collapsed into unreadable runs. The scanned-PDF exception in the wiki skill applies by analogy — Marker’s table processor falls back to OCR, which this image cannot run. Recovered natively with pdfplumber (uv run --with pdfplumber python3 … — note that pip is absent and uv pip install --system is silently ineffective here) into sources/whitepapers/…/tables-recovered-from-pdf.md. The wiki cites that file for Tables 1b, 2, 3 and 4.
2. The Markdown twin was replaced mid-ingest. Martin supplied a conversion he had produced locally; it is now the twin the wiki cites against. A word-normalised diff against my own conversion showed nine differing lines, all Table 1 rows — his version carries Table 1 intact, mine had dropped it. Body text identical, so the full read already done stayed valid and was not repeated. Tables 1b/2/3/4 are collapsed in both, so the pdfplumber aid remains in use; its header now records the replacement.

What the document is. A landscape and gap analysis by the Secretariat of the ICC Technical Expert Panel on Carbon Accounting, commissioned on the openly stated premise that a ledger-based architecture is the answer. The Review repeats that qualification four times; every “open item” it records is distance from that target, not an independently established defect. Read it as an interested party’s brief, which is how the wiki now frames it.

Six pages created: papers/icc-carbon-accounting-landscape-review-2026.md, people-partners/icc-carbon-measures.md, concepts/ledger-based-carbon-accounting.md, concepts/decision-grade-carbon-data-criteria.md, concepts/product-level-carbon-regulation.md, concepts/confidentiality-gap-in-carbon-accounting.md, plus the new wiki-root open-tasks.md.

Eight pages updated: concepts/pact-pathfinder-network.md (the largest edit — a full “external assessment” section), concepts/coordination-vs-bargaining-failure.md, concepts/data-commons.md, concepts/pact-pcf-calculation-methodology.md, concepts/pact-data-quality-verification.md, concepts/digital-product-passport.md, people-partners/wbcsd.md, market-competitors/catena-x.md. The five mechanical cross-referencing edits went to a Sonnet subagent; the judgement pages were written on the main thread.

The finding that matters most. This is the first assessment of PACT in the wiki that comes neither from PACT, from WBCSD, nor from a SINE whitepaper. It is a promotion and a demotion at once: PACT is named the leading interoperability and data-exchange framework, and placed below the accounting layer where the next standards fight will be fought, by a coalition that intends to define that layer. Table 4, verbatim: “Exchanges data rather than accountable quantities; period-average static figures rather than transaction-level records; does not provide conservation, transfer, or reconciliation mechanisms; inherits the quantification latitude of its methodological basis.” The wiki holds this against the 2022 Data Commons whitepaper’s reading of the same facts as a validating success case — contradiction represented, not resolved.

Two further findings, both now on their own pages: confidentiality is stated as a design requirement of Criterion B and carried into none of the document’s later instruments (concepts/confidentiality-gap-in-carbon-accounting.md, ⟨Inference⟩ throughout, at Martin’s instruction to give it its own page); and allocation is framed by the Review itself as market design rather than method, with “methodology arbitrage” named — which is external support for the standing objection that “aligned interests” can be a polite name for a distributive fight.

Martin’s objection to the criteria grid (A Accuracy / C Comparability / D Representational faithfulness / G Completeness of scope are near-synonyms) is recorded on concepts/decision-grade-carbon-data-criteria.md#objections-on-record with my assessment: upheld for A, D and G — the Review lists FASB’s faithful representation alongside two of its own components as siblings — but resisted for C, which is a property of a set of figures rather than of one, and for D’s neutrality half, which catches the bias A cannot. Whether the redundancy is the Secretariat’s or inherited from the Panel’s Mission, Opportunity, and Principles statement cannot be settled until that statement is in sources/; it is not, and that is now an open task.

wiki/open-tasks.md is new and holds four items, headed by Martin’s instruction that SINE must formulate critical feedback on the architecture candidate. Index updated. Commit due.

---

[2026-08-17] ingest | Verifiable Carbon Accounting in Supply Chains (Heiss, Oegel, Shakeri, Tai, 2023)

Source: sources/whitepapers/2026-08-17-verifiable-carbon-accounting-in-supply-chains.md, supplied by Martin Pompéry as a Markdown file. IEEE Transactions on Services Computing 17(4), 1861–1874, doi:10.1109/TSC.2023.3332831.

The paper supplies exactly the mechanism the ICC Review states as a requirement of Criterion B and then never provides: zk-SNARK proofs (ZoKrates) that a supplier’s reported PCF was computed correctly from signed sensor readings under an auditor-approved MRV schema, disclosing neither the readings nor the intermediate figures. Implemented, benchmarked and costed — 214 s setup, ~40 s per operational cycle, 602,461 Gas per on-chain verification — with the authors’ own limits recorded rather than smoothed: recurring audit cost per instance, no amortisation across suppliers, the auditor as a double trust concentration (schema approval and proof-key generation), and sensors simply assumed trustworthy.

New page papers/heiss-verifiable-carbon-accounting-2023.md; new person page people-partners/jonathan-heiss.md. Martin states Heiss is part of SINE inside the CCTI programme; he is not named in the ingested CCTI proposal draft, which is now open task 4.

[2026-08-17] ingest | Emission Impossible (Man & Kleppmann, 2026)

Source: sources/whitepapers/2026-08-17-emission-impossible-verifiable-cloud-emissions.pdf plus the Markdown twin supplied by Martin. E-Energy ‘26, Banff, doi:10.1145/3744255.3811720, CC BY 4.0. The wiki’s first primary source on the research output of CCTI’s Cambridge Co-PI.

Three proofs — total emissions from a signed electricity meter, per-customer shares, and a recursive per-customer proof — in o1js, with no trusted setup, requiring only one public hash commitment per accounting period: “a task that can also be fulfilled by a simpler system such as a transparency log.” Two moves matter beyond the cryptography. First, the emissions figures are proven against the company’s audited financial records, so “in order to lie about its emissions, the company would then also have to lie about its finances, which raises the stakes by making misreporting a crime” — an enforcement mechanism ICC argues by analogy to but never builds. Second, the authors concede that the single-meter assumption “allowed us to side-step the problem of ensuring that all sources of emissions have been accounted for.”

New page papers/man-kleppmann-emission-impossible-2026.md.

Method note. Martin instructed: “Du musst zwingend beide Papiere (Kleppmann, Heiss) zusammen betrachten.” Each paper was read individually and in full and has its own summary page and source attribution — no batch reading — while the joint analysis lives on one shared concept page, concepts/verifiable-carbon-accounting.md. Analysis joint, sourcing separate.

The finding the joint reading produced, and the wiki’s own claim: the completeness of an emissions report is provable wherever an independently attested closing aggregate exists, and cryptography can prove reconciliation against such a constraint but cannot supply one. The ledger’s conservation rule and the data centre’s electricity meter are the same device built out of different materials — one institutional, one physical. So ICC’s architecture and SINE’s cryptography are complements, not rivals, and neither side has said so. This corrects an earlier claim of mine, caught by Martin, that completeness (G) is structurally unprovable in zero knowledge; it is not.

Pages updated: concepts/confidentiality-gap-in-carbon-accounting.md (Status rewritten — thesis upheld and sharpened, per-criterion breakdown), concepts/ledger-based-carbon-accounting.md (new section on the closing constraint), concepts/decision-grade-carbon-data-criteria.md (Operability decomposed into four demands, at Martin’s request), concepts/cryptographic-trust-infrastructure.md (tension kept visible: both papers rely on a trusted signing meter, which is what CCTI’s M5 gate undertakes to remove), concepts/data-commons.md (the blockchain-vs-P2P contradiction partly defused — conservation needs one public append-only commitment, not a chain; still live for the Heiss design, which verifies in an Ethereum contract), projects/ccti.md, people-partners/martin-kleppmann.md, wiki/open-tasks.md (item 1 unblocked and extended to six points; both ingests moved to Closed). Index updated. Commit due — this one and the ICC ingest before it.

[2026-08-17] synthesis | Two corrections and one new page, from discussion with Martin Pompéry

Not an ingest — a revision of what was written earlier the same day, driven by three challenges from Martin, each of which held.

1. Completeness (G) was stated too generously. The earlier wording, “F and G hold only inside a closing constraint”, read as if completeness were achieved under a condition. Corrected: a proof is a statement about a computation, whereas “nothing was omitted” is a statement about the world and is unreachable from inside the circuit. It becomes provable only against an aggregate that is (a) independently attested, (b) outside the reporter’s control, and (c) demonstrably covering the boundary. Man & Kleppmann have (a) and (b) and concede they lack (c) — twice, once in the conclusion and once in §5, where an operator with multiple meters could simply omit some. Their own proposed repair is to cross-check against the electricity supplier’s bills, i.e. to import a second independent attestation of the same aggregate. So G is reduced, not solved: to does such an aggregate exist, and does the report reconcile to it? Corrected on concepts/verifiable-carbon-accounting.md, concepts/confidentiality-gap-in-carbon-accounting.md and concepts/ledger-based-carbon-accounting.md. Also recorded there: the trusted meter carries the load twice — as integrity for D and as boundary for G — which makes it one continuum with the M5 tension, not two findings.

2. Proof cost was over-weighted. Martin: verification cost is not a fundamental limit and does not bear on the ICC document; Heiss’s 602,461 Gas is a fact about Ethereum in 2023. Accepted. The figures stay on the two paper pages and are out of the criterion assessment on concepts/decision-grade-carbon-data-criteria.md and concepts/verifiable-carbon-accounting.md. What survives is narrower: Man & Kleppmann as a first indication of what Operability costs in an operationalised environment.

3. New page concepts/temporal-resolution-and-the-service-case.md, from two objections Martin raised and asked to have verified against the original text before answering. Both upheld:

• H demands what C forbids. Criterion H — the one criterion the Secretariat wrote itself rather than adopting from the Panel — requires records “aligned to financial periods and tending towards near-real-time”. But PCF calculation is ex post and aggregated because averaging removes temporary variance and is what makes figures comparable, which is Criterion C. The Review treats period-averaging only as a defect (Table 4 debits PACT for “period-average static figures”; §5.2 praises E-Ledgers for being “event-driven”) and never states what averaging is for. Its one near-answer, that ledger numbers “can be aggregated in any reporting window”, relocates the averaging rather than avoiding it — and moves the choice of window from the Product Category Rule to whoever queries, which is a comparability problem added.
• Services are absent. The word “service” occurs once in the entire Review, in the ICC boilerplate about dispute resolution. No logistics, no freight-as-a-service, no IT. The vocabulary is physical throughout — “physical product flows”, tanks, pipelines, terminals, transfer of legal title — while most of Scope 3 is services, which have no product for embedded emissions to attach to and no custody event at which title transfers. §6.2’s list of deferred hard cases does not name services either: the finding is the silence, not a stated exclusion.

⟨Inference recorded on the page: these two are a different kind of finding from the confidentiality gap. They sell nothing, cost nothing to concede, and are checkable against the document in minutes — which is what makes them the credible part of a response.⟩ Added as point 7 of open task 1.

Index updated. Commit due.

Addendum, same day. At Martin Pompéry’s instruction, the meta-claim that the criteria grid was built backwards from the preferred architecture is now carried as point 8 of open task 1, with the defence on record (seven criteria adopted from the Panel’s statement “in that statement’s order and wording” — a document not in sources/, so part of the redundancy may be inherited; the defence does not cover H) and an explicit handling note: it is the one point that cannot be conceded gracefully by its recipient, so the dossier must decide deliberately between the sharp version, the constructive version (propose a non-redundant grid) and omission. Cross-referenced from concepts/decision-grade-carbon-data-criteria.md.

[2026-08-18] build | Coffee — Minimum Viable Context (MVC) v0.1

New artifact type, at Tell Münzing’s request. He named the problem in the conversation itself: “the lack of a common context. My world: I need to get you into the context and relationships. So how could we experiment with an MVC, minimum viable context, for this task?” Test case: the GIZ/WHH approach in coffee. Not an ingest — no source was handed over; six web-research inventories were commissioned and filed first, then the map was built from them plus the existing wiki.

New sources (immutable, research-grade — commissioned web research, not primary documents; three recovered verbatim from the session transcript after a context compaction):

• sources/research/2026-08-18-global-coffee-platform.md
• sources/research/2026-08-18-welthungerhilfe.md
• sources/research/2026-08-18-giz-leadership-private-finance.md
• sources/research/2026-08-18-coffee-value-chain-data.md
• sources/research/2026-08-18-agri-coffee-finance-instruments.md
• sources/research/2026-08-18-coffee-asset-owners-principals.md

New pages: context-maps/coffee.md (the MVC), organization/global-coffee-platform.md, organization/welthungerhilfe.md, people-partners/mathias-mogge.md, people-partners/ingrid-gabriela-hoven.md, people-partners/anna-sophie-herken.md.

Updated: index.md — new “Context maps (MVC)” section ahead of Organization; two organizations and three people added.

Structure per Tell’s two corrections. First correction: the MVC “misses at minimum the economics i.e. how value propositions and business case are shifting,” and must “keep the core offering of SINE at the center of your value chain perspective.” So each of nine chain stages (nature → farmer → co-op → wet mill → dry mill/exporter → trader → roaster → retail → consumer → cycle closure) carries five columns: primary data held and missing; the claim that must be made believable and to whom; SINE’s five instruments applied (governance approach, data collaboration, primary data, standards, open-source code); the economics; the finance touchpoint. Second correction: people “provide adjacency that enables new alliances and innovation” — so the people register carries an adjacency field, and the actor register carries mandate / money-and-horizon / relationships / principal-or-agent-and-distance-to-asset.

Load-bearing findings:

• Principals sit at both ends — smallholder (owns land, trees, harvest; no agency) and pension funds/insurers/roaster owners (large assets, remote sight). Everyone between, SINE today included, is an agent. Tell’s “shift closer to actual asset owners and principals” is therefore two moves in opposite directions, connected by a verified primary-data stream.
• Coffee’s buy side is family- and state-controlled, not capital-market-facing (Melitta, Lavazza, Tchibo, ECOM, LDC/ADQ, Olam Agri/SALIC, NKG, JDE Peet’s/JAB; Nestlé the exception). A “capital markets” strategy mostly mis-addresses this sector.
• Warrantage exists because land cannot be mortgaged — Ethiopia’s land is constitutionally state-owned and unmortgageable, Vietnam’s LURCs are mortgageable at 70–85% LTV. Its verification technology is a dual padlock with keys split between bank and farmer group, which caps the instrument at walking distance. The sharpest available formulation of SINE’s proposition in this domain: a verified data stream can do what the padlock does without the padlock’s radius. Caveat recorded: that substitutes fraud risk for theft risk, and no source says the trade is favourable.
• No instrument anywhere prices a roaster’s capital against verified farm data. Confirmed sustainability-linked facilities (Olam $500m/50 ESG criteria via Sustainalytics; ECOM $600m) sit at trader level and are gated on an ESG rating, not primary data. The clearest white space, at the end of the chain where margin exists (roaster €0.89/kg vs. trader €0.26/kg).
• GCP already holds a farm-level data standard (15 indicators, Delta Project, aiming at government endorsement) with no located position on data ownership, governance, or the EUDR. SINE need not build a coffee data standard; the gap is exactly its layer. Counter-consequence stated: GCP is also best placed to build or buy that layer itself.
• The coffee PCF median of 2.18 ± 2.04 kg CO2e/kg with nitrogen fertiliser at 65–100% of emissions is the economics of primary data: it does not lower the number, it narrows it, and only a narrow number can carry a financial instrument.
• BIS/IAIS names data infrastructure as a systemic constraint on reinsurance pricing and underwriting for climate risk — the strongest external corroboration on file for SINE’s premise, and support for Martin Pompéry’s 2026-08-15 insurance-first argument. Parametric insurance’s adoption barrier is basis risk, which is a primary-data resolution problem.
• Timing: EUDR now due 30.12.2026 (medium/large) and 30.6.2027 (micro/small) after two postponements; GIZ Tender 10036847 runs 15.10.2026–15.10.2027, across that date. A third-delay scenario is noted as necessary.

Two contradictions to Tell’s account held open rather than resolved:

1. No source attributes warrantage implementation to WHH — all located sourcing credits FAO (Niger 1999, extended regionally), Oxfam and the World Bank. Three readings left open on the WHH and Mogge pages; resolves only from Mogge’s own account.
2. GIZ’s chair is Thorsten Schäfer-Gümbel since November 2022 with no public reporting of any succession — Tell’s “Ingrid probably becoming the next Vorstandsvorsitzender” is recorded as his non-public information, not as a sourced fact.

Three holes only Tell can close, restated at the end of the map: the FAO warrantage document (the URL arrives truncated on my side both times — three candidate bitstream UUIDs are listed on the page), the Mogge conversation notes, and the finance-side names. The map has a populated development and industry layer and an entirely empty finance layer: not one named individual at a bank, insurer, asset owner or finance association.

Design principle recorded on the page itself: the MVC’s job is to be wrong in visible places. The register of holes is the most valuable part.

A commit is due.

[2026-08-21] dossier | Adtech als regulatorischer Präzedenzfall

Martin Pompéry asked for research papers on online advertising from two angles — (a) the actor taxonomy of the broader market, (b) market structure / market failure plus remedies — as input for informing policymakers on AI market regulation. Framing agreed in advance: adtech as a precedent for regulating a multi-layered, vertically integrated stack, not “what AI does to ad markets.” Output: a German dossier with quotations in the original language, 10–15 papers “mit echtem Tiefgang.”

Delivered: dossiers/2026-08-21-adtech-als-regulatorischer-praezedenzfall.md (+ 17 pp. PDF, sent to Slack). 18 texts in six mechanism families — auction design and opacity; header bidding and the first-price transition; verifiable auctions; data as a production factor; privacy regulation as a concentration mechanism; separation and interoperability remedies. Angle (a) became the opening taxonomy section from agency documents (CMA 2020 App. M, ACCC 2021 ch. 2, Bundeskartellamt 2023) rather than consuming paper slots.

Sources: 19 verified PDFs in sources/adtech/. Two were paywalled and Martin supplied them himself — Witte & Krämer (Telecommunications Policy 2025) and Hovenkamp, Structural Antitrust Relief Against Digital Platforms (JLI 7(1), 2024). The latter corrected a catalogue error: SSRN 4616175 is a distinct paper from Hovenkamp’s 2023 Columbia interoperability piece, not a preprint of it, and it does not treat adtech at all; the PDF was renamed and the dossier carries two separate Hovenkamp entries.

Reading Witte & Krämer in full also corrected my abstract-based reading of their proposal: not “divest the buy side,” but divest DV360 and confine Google Ads to first-party inventory — they explicitly reject the broader cut as needlessly costly. The narrow cut is chosen for enforceability alone (“verifying that no requests are sent should be a straightforward technical check,” p. 18), which is the construction rule the dossier builds on.

New wiki pages:

• concepts/adtech-as-regulatory-precedent.md
• concepts/binary-verifiable-remedies.md

Cross-reference added to concepts/verifiable-carbon-accounting.md: the completeness limit found in Man & Kleppmann’s single-electricity-meter side-step recurs independently in the verifiable-auction literature (VEX 2013, Addax 2023) — both prove correct rule application to the committed inputs, neither proves the committed set is complete. ⟨Inference: a domain-independent boundary of cryptographic verification.⟩

Two things worth carrying forward as findings about the field rather than about any paper. First, provenance: the technically sharpest mechanism literature is written inside the firm under study (the autobidding survey is 25 of 25 Google authors) and the sharpest structural critique comes from complainants’ economists — and all three empirical papers depend on a commercial data provider or one firm’s cooperation, because the data is not public. The same constellation is forming around AI evaluation data. Second, the two most sophisticated economic claims about data as a production factor — Acemoglu et al. (submodular externalities, prices too low, competition does not help) and Farboodi & Veldkamp (data has the diminishing returns of capital) — reach opposite conclusions and neither cites the other.

Method disclosed in the dossier’s falsification section: 5 of 18 texts read directly in the main thread, the rest via schema-driven extraction agents. An extraction schema finds what it looks for.

A commit is due.

[2026-08-22] dossier | Zwei Inversionen und eine Leerstelle — KI-Marktstruktur vs. AdTech

Second dossier for Martin Pompéry, commissioned as the comparison piece to Dossier 1: market structure across the whole AI value chain (compute → foundation models → inference API/MaaS → orchestration/routing/agents → distribution) against the adtech sector. dossiers/2026-08-22-ki-marktstruktur-im-vergleich-zu-adtech.md, 12 pp. PDF, 16 primary sources newly filed under sources/ai-market/.

Explicit instruction: routers integrated as actors and evidence, not as the spine; the falsifying version wanted; Dossier 1 left untouched but corrected out loud where the two disagree.

Method: property-by-property, not layer-by-layer — eleven structural properties, each with one of three verdicts (same mechanism / analogous but different magnitude / structurally absent). Rejecting the layer mapping was the design decision that made the comparison say anything; mapping layers always finds a fit somewhere.

Three headline findings, all on the new page concepts/ai-stack-market-structure.md: the stickiness inversion (in adtech the rule-setting layer was also the stickiest, which is what let it hold a rent — in the AI stack compute is sticky but sets no third-party rules, routing sets rules but is a changed base URL, so the pathology has no single home); the opacity inversion (adtech buyers knew what they bought but not what they paid; here they know what they pay but not what they got); and the missing auction (no auction anywhere in the AI stack, so roughly half the transferable adtech findings — auction findings — have no target).

Also: four agencies locate the chokepoint on four different layers, where in adtech they converged on the ad server, and none of them names the routing layer. Network effects are denied outright by the leading AI-economics paper. The entry barrier is capital and geography rather than information — and unlike adtech’s, it is measurable.

Three corrections to Dossier 1, stated in the dossier rather than applied silently: Transfer 4 moves to the compute layer (France’s SREN Law Art. 26–28 is already a binary-verifiable remedy, including a statutory reporting duty on self-preferencing — concepts/binary-verifiable-remedies.md updated, its “Transfer to the AI stack” inference was wrong on both availability and layer); the intermediation-rent claim becomes precise (0–5.5 %, payment-fee-shaped not spread-shaped — the original claim was wrong and so was the full retraction of it); and “circular spending” is the FTC’s own term, not our finding.

Third literature gap of the project, same shape as Acemoglu vs. Farboodi & Veldkamp: Korinek & Vipra recommend data siloing as a remedy; Witte & Krämer identify data siloing as unverifiable. Neither cites the other. Plus an instrument asymmetry — natural-monopoly rate/service regulation appears only in the AI literature, binary verifiability only in the adtech literature.

The load-bearing non-finding, reported as such: no documented accusation, complaint, audit, investigation or benchmark alleges router self-preferencing, and no vendor publishes an auditable routing policy. Absence of evidence, not evidence of absence.

Provenance caveat carried into the dossier: this one leans harder on agency documents than Dossier 1, and agency documents are party documents. The routing evidence is almost entirely vendor self-documentation. An adversarial verification pass killed or downgraded nine load-bearing numbers from the first research wave, including a widely quoted take rate with no traceable origin.

Wiki footprint closed on adtech-as-regulatory-precedent.md: transfer 1 marked corrected (stickiness inversion), transfer 2 marked sharpened-and-inverted, the first Open question (“is AI allocation auction-like enough?”) answered negatively with the CME compute-futures exception noted, and a ## Related section added (the page had none).

Phase 2 offered; a commit is due (both dossiers, both source folders, five wiki files: index.md, log.md, concepts/ai-stack-market-structure.md, concepts/binary-verifiable-remedies.md, concepts/adtech-as-regulatory-precedent.md).

---

[2026-08-31] lint | Third scheduled health check — 84 pages structurally clean, the tender went stale

Scope: all 84 wiki pages plus the source tree. Mechanical checks (link resolution, orphans, index coverage, inbound-link counts, source-reference integrity) and two content inventories (⟨Inference⟩ markers, time-sensitive claims) run as four scoped subagents; contradictions and the fixes decided here.

Mechanical result: clean. 0 broken internal links (anchors included), 0 orphan pages, 0 weakly-connected pages, index coverage 84/84. The wiki’s structure has held across three lints — the 2026-08-17 lint’s 7 broken links and 2 orphans have not recurred. Inbound-link counts remain healthy: concepts/data-commons.md at 32, concepts/pact-pathfinder-network.md at 24, nothing below 2.

The finding: the GIZ tender pages describe a future that has happened. The submission deadline (21.08.2026, 12:00) passed ten days ago. Both tender pages still read as live task lists — “morgen”, “in vier Tagen”, “four days”, Status: open, eight blocking items, a five-CV gap analysis, “to be defined in the week of 17.08.” — and the wiki records no outcome: not whether SINE submitted, not with which consortium, not whether HRNS ever committed. This is the largest stale surface in the wiki and the one most likely to mislead a reader who arrives via the index.

Fixed:

• tenders/giz-10036847-data-governance-agvc.md — deadline banner rewritten from a countdown to an expiry notice; the pre-deadline text kept verbatim underneath for reconstructibility. Status: now states that the deadline passed and the outcome is unrecorded.
• tenders/giz-10036847-konsortium.md — snapshot banner added at the head; ## Open and blocking (as of 2026-08-17) retitled to mark it historical. No item in the list was edited: the record of what was blocking on 08-17 is worth keeping intact.
• organization/hanns-r-neumann-stiftung.md × people-partners/michael-opitz.md — a real contradiction, now represented rather than resolved. The org page read Opitz’s departure as settled (successors reported by Daily Coffee News, 01.07.2026); his own page read the same LinkedIn profile as transitional (“Managing Director 2005 – Present” and “Former Managing Director … July 2026 – Present” both present). Each page now carries the other’s reading and a link to it. Only Opitz closes this.
• concepts/adtech-as-regulatory-precedent.md — provenance repair. The page carried 18 papers’ worth of claims against a single bulk pointer (“Sources in sources/adtech/”) plus author-year prose, so no claim resolved to a file — while its sibling ai-stack-market-structure.md, written a day later, links every source by path. Ten author-year citations converted to file links (Witte & Krämer, Despotakis, Zeithammer, Srinivasan, Decarolis & Rovigatti, Prat & Valletti, Farboodi & Veldkamp, Acemoglu, Gilbert, Bichler). Agency reports (CMA, ACCC, Bundeskartellamt) and Latham et al. left unlinked — no file exists for them; the two Hovenkamp papers left unlinked because the prose cites both collectively. No wording, number or page reference changed. All ten targets verified to resolve.
• index.md — Kleppmann described as “WP-A/WP-B lead”; the CCTI proposal assigns those work packages to Cambridge, not to him personally. Corrected to match projects/ccti.md and his own page.
• concepts/pact-pathfinder-network.md — source citation ended in a literal ellipsis; completed to the Markdown twin’s real filename.

Raised as open tasks (open-tasks.md, now seven open):

• #5 the tender outcome — see above. The Zuschlags-/Bindefrist runs to 13.11.2026, so this is a live question, not a post-mortem.
• #6 Dafoe et al. (2020), Open Problems in Cooperative AI — downloaded and converted on 2026-08-22, then never ingested. No page, no index entry, no log entry: the only source in the repository in that state. It sits directly under the AI Advantage claim.
• #7 FNG-Marktbericht 2026 — appeared today under a new sources/sustainable-finance/, unannounced: no Markdown twin, no page, no record of who filed it. First sustainable-finance source in the repository, so it opens a subject area rather than extending one.

Carried, unchanged since 2026-08-17: open tasks #1 (SINE’s critical response to the ledger architecture candidate), #2 (ICC FAQ ingest, still queued), #3 (the Panel’s Mission, Opportunity, and Principles, still missing), #4 (Heiss’s role inside CCTI, still undocumented). Two weeks without movement on four items is itself worth naming.

Not fixed, reported instead:

• 12 pages carry no date stamp at all (10 of 17 in people-partners/, plus organization/aria.md and organization/giz.md). Mostly stable organizational profiles where this matters little, but aria.md and giz.md are load-bearing and undated.
• ~272 ⟨Inference⟩/⟨Background⟩ markers across the wiki, 37 of them in context-maps/coffee.md alone. Nine are checkable rather than interpretive — six of those resolve by asking one person (Mogge on warrantage, Opitz on availability and on E.D.E. Vietnam), which is the cheapest verification backlog in the wiki.
• EUDR at 30.12.2026 (context-maps/coffee.md) and CCTI M1 at 30.09.2026 (projects/ccti.md, evaluation use cases “not yet selected”) both fall inside the next four months. Not stale yet; stale soon.

A commit is due — this lint’s seven files, plus everything still uncommitted from 2026-08-22 (both dossiers, sources/ai-market/, the Dafoe source, and concepts/ai-stack-market-structure.md).

[2026-09-07] lint | Fourth health check — a frozen week, so the lint went after content: one canonical page created, 23 cross-references repaired

Scope: all 86 pages plus the source tree. Two content inventories dispatched as sonnet subagents (concepts without pages, missing cross-references) — the two checks the 2026-08-31 lint covered least well; mechanical verification and all judgement here.

The wiki did not change this week. Nothing under wiki/ was modified since the last lint, no log entry was appended, and the page count held at 86. The operator’s commit landed (55709a3, “Dossier 2 … + wiki lint 2026-09-05”), so the overdue commit from last week is resolved and the tree is clean. Only sources/ moved, and only for one file. ⟨Inference: a frozen week means nothing new can have superseded anything, which is why this lint spent its budget on the structural-content checks rather than on staleness — staleness this week is just last week’s findings, one week worse.⟩

Mechanical result: clean, for the fourth consecutive lint. 0 broken internal links, 0 broken heading anchors, 0 orphan pages, index coverage complete. The 2026-08-17 lint’s 7 broken links and 2 orphans have still not recurred.

⟨Method note: the anchor checker in the 2026-08-31 run collapsed runs of whitespace to a single hyphen. GitHub replaces each space individually, so headings containing an em-dash — ## Contradiction held open — warrantage resolves to #contradiction-held-open--warrantage, with two hyphens — produced seven false positives. Corrected; those seven links are fine.⟩

The finding: warrantage had no page

The coffee map calls warrantage “the hinge of the whole map” and the index lists it among the map’s load-bearing findings. It is discussed on four pages. It had no canonical page — the mechanism, the Niger/FAO origin, the preconditions, the evaluation figures and the land-tenure argument all lived inside §3 of a context map, reachable only through a 90-character heading anchor, while two other pages linked to that anchor to explain themselves.

That is the one-canonical-page-per-concept convention failing on the wiki’s own stated load-bearing finding. Created concepts/warrantage.md: mechanism and dual-lock, origin and spread, the three preconditions, the Niger evaluation figures (flagged as single-evaluation with very wide ranges), the Ethiopia-Art.-40-vs-Vietnam-LURC tenure argument, the ⟨Inference⟩ that a verified data stream is a claim-verification technology without the padlock’s radius, the limit stated against that inference (substituting verification for custody moves risk from theft to fraud, and no source says the trade is favourable), the WHH attribution contradiction, and the four open items. Every claim carries its sources/research/2026-08-18-* citation; nothing was invented for the page.

The coffee map keeps §3 — it is a strategic argument, not an explanation of an instrument — and now opens with a pointer to the concept page. Mogge now links the concept page instead of the context-map anchor; WHH’s contradiction heading links it too.

Five more concepts without pages — reported, not fixed

EUDR, Neumann Kaffee Gruppe, FAO, IDH and BMZ all clear the same bar warrantage did: discussed substantively across 4–7 pages, no page of their own. EUDR is the strongest — it is the regulatory anchor of the entire SAFE/tender/coffee thread, and the coffee map already flags “whether GCP has any EUDR position at all” as a hole. IDH is flagged in the coffee map’s own register of holes (“IDH as an institution … no wiki page”), so the wiki had already caught that one itself.

⟨Inference: these five are not lint fixes. Warrantage could be written because every claim was already ingested and traceable; these five would need either a source that is not in sources/ (the EUDR text itself) or a research pass. Writing them from search results inside a lint would put unsourced pages into the wiki, which is the more expensive mistake.⟩ Left for the operator to prioritise.

23 missing cross-references repaired

Applied as a mechanical pass by a sonnet subagent under an explicit no-rewording rule — existing words turned into links, no prose touched, every relative path verified against disk. The notable ones:

• The coffee map’s whole Global Coffee Platform actor-register subsection had no link to the GCP page anywhere in the file.
• Tell Münzing is quoted as the source of the central framing claim on both the GCP page and the WHH page, and was linked from neither.
• concepts/ledger-based-carbon-accounting.md cites Man & Kleppmann’s specific findings without linking the paper page — every other page that discusses that paper links it.
• One systematic gap: all six market-competitor pages cite the Hypothesenpapier by hypothesis number in their strategic-relevance section and none linked it. Fixed as one batch.
• The German coaching-stack assumption-register-template.md names five schools in worked blocks (Cynefin, Heidelberger Schule, Dark Matter Labs, Warm Data, GROW) and linked none of the five pages.
• Two places in coaching-stack/open-questions.md cited the target as a bare backticked path rather than a link.

⟨Inference: the pattern across all 23 is that pages written as an argument link well, and pages written as registers, templates or comparison tables link badly — the cross-reference decays exactly where a page enumerates rather than argues.⟩

Staleness: last week’s items, one week worse

• GIZ Tender 10036847 — seventeen days past the deadline, outcome still unrecorded. Both banners rewritten to drop the relative-time phrasing (“seit zehn Tagen”) that would rot again; the status line now records that the item has stood open across two lints.
• CCTI Milestone 1 (30.09.2026) is three weeks out and the wiki’s only source on it is the 2026-06-17 proposal draft, which lists the evaluation use cases as “not yet selected”. Noted on the CCTI page and raised as open task #8 — ⟨Inference: the defect is not that the gate may be missed, it is that the wiki cannot tell a missed gate from an unreported one. The wiki holds no source on CCTI newer than the draft proposal while leaning on CCTI for SINE’s answer to Criterion B.⟩
• EUDR 30.12.2026 — unchanged, still ahead.
• No countdown language (“morgen”, “in vier Tagen”) survives anywhere outside the log; last week’s fix held.

The FNG source: a twin appeared, an ingest did not

sources/sustainable-finance/ was the only part of the repository that moved this week. The Marktbericht now has two Markdown renditions: a proper marker_single conversion (1,427 lines) and a flat text extraction (4,608 lines) that interleaves layout fragments into the running text. Open task #7 now names the marker file as the one to ingest from, and records the two cosmetic convention divergences (twin sits flat rather than in a <slug>/ folder; carries a -marker suffix instead of the PDF’s stem) as recorded, not corrected — sources/ is immutable.

Still no wiki page, no log entry, and no record of who filed it. ⟨Inference: someone converted it, which means someone intends it to be read. That the conversion happened without a word to me is the thing to fix, not the file.⟩

Not fixed, deliberately

• The five concept pages above.
• ~272 ⟨Inference⟩ markers. Unchanged from last week; the same nine are checkable and six of those resolve by asking one person.
• The eight sources/ai-market/ PDFs are still cited nowhere by file path, the same provenance asymmetry the 2026-08-31 lint repaired on the adtech page. Left because the mapping from claim to paper on ai-stack-market-structure.md is not mechanical the way the adtech one was — several claims are attributed collectively, and a subagent guessing which of eight papers carries which figure is exactly the failure mode worth avoiding. It needs one focused pass with the papers open.
• 12 undated pages. Cosmetic.

Open questions to the user

Three, all repeats, which is itself the finding: was the tender submitted and in which constellation; who filed the FNG-Marktbericht and should it be ingested; and have the CCTI evaluation use cases been selected. Open tasks #1–#4 have now stood unmoved for three weeks.

A commit is due — 25 files: one new page, the index, the log, and 22 existing pages carrying the 23 repaired links plus the staleness edits.

[2026-09-10] ingest | Evaluating the ability of LLMs to quantify consequences (Ojala / Upright, 2025)

First of three Upright Project texts, requested by Martin Pompéry on 2026-09-09 with an explicit CCTI reading and two added lenses (ML approaches to coordination; training data and feedback loops, read against Ostrom’s commons and cybernetics). Processed in publication order, one at a time.

New topical subdirectory sources/impact-measurement/ — the first source class in this wiki that is neither a whitepaper, a tender, nor carbon/ag material. Filing-date prefix (2026-09-10), publication date recorded in each file’s provenance header. Full HTML retrieved with curl and converted locally, per the no-summarising-fetch rule.

• Source: sources/impact-measurement/2026-09-10-upright-evaluating-llms.md
• New page: Evaluating LLMs (Ojala, 2025) — CoCoCo defined, the full 11-model table, the greenwashing-in-corpus finding
• Updated: ESG reporting in the financial sector — new section on the reporting→corpus→assessment loop; AI Advantage — CoCoCo as a precedent for the M4 “publishable either way” gate, with the warning about self-authored benchmarks

The finding worth carrying forward: LLMs systematically flatter companies and minimise environmental impacts, attributed by Upright to corporate sustainability reports in the training corpora. ⟨Inference⟩ that closes a loop Upright names the bias of but not the shape of, and it is the strongest argument in this material that attested primary data and inferred data are not interchangeable.

[2026-09-10] ingest | When prompt engineering just isn’t enough — Leena-2 (Ojala / Upright, 2026)

• Source: sources/impact-measurement/2026-09-10-upright-leena-2.md
• New pages: Leena-2 (2026); Coherence as a truth substitute — the concept page the material earns
• Updated: Epistemic Integrity Certificate — new section on the independent statement of the same problem, incl. the commitment rhyme ⟨per Martin Pompéry, 2026-09-09⟩ and the WP-C use-case argument

The substantive claim: when there is no ground truth, train and solve for coherence. ⟨Inference⟩ the loop then has no exogenous error signal and converges to a fixed point of its own priors — which is exactly the half WP-A’s attested-data layer supplies. Recorded on both the paper page and the concept page rather than as a passing remark, because it is the cleanest statement of CCTI’s complementarity to an outside system found so far.

Two ⟨Inference⟩ markers left deliberately unresolved: whether the requisite-variety reading of the ~800-topic Universal Impact Structure is Upright’s implicit argument or only a better one available to them, and whether the Fermi-benchmark generalisation survives replication. Neither is answerable from the text.

[2026-09-10] ingest | AI’s shortcomings in understanding consequences are a practical alignment problem (Ojala / Upright, 2026)

• Source: sources/impact-measurement/2026-09-10-upright-consequences-alignment-problem.md
• New pages: Consequences as an alignment problem (2026); The Upright Project — the entity page, filed under a new **Impact & ESG data vendors** sub-group in the index
• Updated: Cryptographic trust infrastructure — the outside argument for WP-A; Ostrom’s commons design principles — Upright as a live enclosure counter-case to the commons prescription; Coordination vs. bargaining failure — a worked classification of an outside organisation; Data Commons — cross-reference to the counter-case
• Index: three paper entries, one concept entry, one market entry, one new sub-group

Load-bearing finding: models are hardened against being asked and remain vulnerable to being told, so the exploitable surface is facts rather than instructions. ⟨Inference⟩ instruction-filtering therefore cannot be the defence and provenance is the only discriminator — an independent, empirically-backed argument for the attested-data layer, made by people with no knowledge of CCTI. Best available external pitch material for WP-A.

Two things recorded rather than smoothed over

• A false positive on our own side. Two subagents reading uprightproject.com reported a prompt injection. Checked against the raw HTML: the flagged strings are the article quoting its own illustrative examples. A post about prompt injection pattern-matching as one. Noted on the paper page because it is a live data point on that post’s thesis.
• The commons reading is a clean negative result. A terminology sweep of 41 pages found zero hits for commons, Ostrom, collective action, free rider, externality, data commons — and zero for Ashby, Beer, Meadows, feedback loop, requisite variety, viable system, despite the structure being thoroughly cybernetic. Recorded as a finding, not as a gap in the search.

Evidentiary status

None of the three texts is peer-reviewed; two carry self-assigned report numbers; the benchmark is Upright’s own, the harness closed, the model whitepaper behind a lead form whose v2 PDF 404s. A search for independent academic evaluation or critical journalism on the net impact model returned nothing — all located third-party coverage is descriptive or promotional. The company facts on the entity page are live web research, marked ⟨Background⟩ and explicitly not in sources/.

A commit is due — 18 files from this ingest: 5 new pages (3 papers, 1 concept, 1 market-competitor), 3 new source files under the new sources/impact-measurement/, 7 updated concept pages, the CCTI project page, the index and this log. The 2026-09-07 lint’s files are still uncommitted underneath, so the working tree now carries two sessions’ work.

---

[2026-09-14] lint | Fifth health check — the first lint run against a fresh ingest: caveats were dropping in transit

Structure: clean for the fifth consecutive lint. 92 pages, 0 broken links, 0 broken anchors, 0 weakly-linked pages, index complete. One orphan turned up — log.md itself, which nothing linked to. Fixed in index.md. (Earlier lints reported 0 orphans; this run’s checker no longer treats the log as exempt, which is right — the persona names it one of the wiki’s two key files, so it should be reachable from the catalog like anything else.)

The finding: the Upright caveat was not travelling with the claims.

The 2026-09-10 ingest was well executed and closed with an unusually honest “Evidentiary status” note — none of the three texts peer-reviewed, two with self-assigned report numbers, the benchmark Upright’s own, the harness closed, the whitepaper gated, no independent evaluation found anywhere. The entity page carries that as a full Evidentiary health warning section.

The caveat did not survive the cross-references. Eight pages now lean on this material, and the warning lived on two of them. What travelled instead were bare vendor numbers, presented in the register the wiki uses for established fact:

• concepts/ostrom-commons-governance.md — restated Upright’s corpus size (200M+ articles), database count, taxonomy size and a 2018 blog quote as plain fact. The entity page marks every one of those as ⟨Background⟩ and says explicitly that they are not in sources/. So a claim the wiki had correctly refused to treat as sourced became a sourced-looking claim by being cited. That is not a style problem; it is the traceability rule failing at a link boundary. ⟨Background⟩ restored.
• concepts/cryptographic-trust-infrastructure.md — the red-team figures (45 breaks / 25 LLMs / 12 scenarios) stated flat, and the inference on them called the result “Empirically-backed, independent”. Both words were wrong in the way that matters: the study is a single self-published vendor red-team, and “independent” meant arm’s-length from SINE, not independently verified — a reader reaching for pitch material would not have made that distinction. Rewritten, with the substitution noted in place so the change is visible rather than silent.
• projects/ccti.md — the same figures again, feeding WP-A, WP-C and M4. Pointer to the health warning added at the head of the section, before anything leaves the wiki in a proposal.
• concepts/esg-reporting-financial-sector.md — “An empirical finding worth carrying on this page” for what is one unreplicated in-house study. The causal half was already hedged (“a suspicion, not a demonstration”); the evidentiary weight of the finding itself was not. Qualified.
• concepts/ai-advantage.md — carried the caveat correctly, in the same breath as the claim. Left alone, and worth naming as the positive control: the difference is that this page states the caveat inline rather than delegating it to a link.

⟨Inference, and the generalisable part: a caveat parked in its own section does not propagate. A link is not a carrier — the citing page inherits the claim and leaves the warning behind. Pages that hedge in the sentence that makes the claim keep the hedge when they are quoted; pages that hedge in a closing section do not. This is worth applying at ingest time, not at lint time: when a source is weak, the hedge belongs in the clause, not in a footer. The failure mode is specific — it was vendor facts and vendor numbers that lost their status, not inference markers. Every ⟨Inference⟩ in this material kept its bracket everywhere it recurred, which is the convention working.⟩

One contradiction, now held open. The ESG page’s inference reads that “attested primary data and inferred data are not interchangeable, however good the inference gets”. Criterion H of decision-grade-carbon-data-criteria.md holds the opposite in the general case — a decision-grade system grades measured / supplier-specific / modelled / default data and discloses the quality of each — and pact-data-quality-verification.md scores estimated data rather than excluding it. The two are reconcilable only if “inferred” means LLM judgement from a corpus of self-reports and not modelled data with a disclosed rating, a distinction the wiki draws nowhere. Neither page knew about the other. Stated on the ESG page as a tension held open rather than resolved, because resolving it needs a source, not an editor.

Two one-directional citations repaired. upright-project.md invoked Ostrom by name without linking the Ostrom page (which links to it twice); the two earlier Ojala paper pages were cited by ccti.md without linking back. Three links added.

Relative time removed again. Three phrasings written by the 2026-09-07 lint had themselves rotted in a week: “M1 is three weeks out” (16 days), “seventeen days after the deadline” (24). Replaced with absolute dates and lint-count framing, which is the form that does not decay. Two German lines in giz-10036847-konsortium.md still read as live gates against dates in August (“Blocking for the KE1 CV before 21.08.”, “Unchanged if HRNS does not commit by 18.08.”) — both converted to past tense. The index’s “Seven open as of the 2026-08-31 lint” was two weeks and one item out of date; now eight.

Not fixed, deliberately. The ~15 status claims resting on sources older than three weeks (HRNS “under examination”, Opitz “not yet asked”, E.D.E. “unverified”, PACT adoption figures from 2025) are not lint material — they are accurate records of what the sources say, and correcting them needs new sources, not editing. Three pages still have no outbound links at all (ada-lovelace-institute.md, gs1.md, mozilla-foundation.md); they are reachable and correct, just terminal. Noted for a third consecutive lint.

Open, unchanged, and now conspicuous: all eight open tasks. Nothing has closed since 2026-08-17, four weeks ago. CCTI M1 falls 30.09.2026 and the wiki still cannot distinguish a gate about to be missed from one met without being reported. The GIZ tender outcome has been unrecorded through three lints.

A commit is overdue by two sessions — 39 files: the 2026-09-07 lint, the 2026-09-10 Upright ingest, and this lint’s 11 edited pages. The working tree now holds three sessions’ work and HEAD has not moved since 2026-09-05.

[2026-09-21] lint | Sixth health check — the dossier→wiki boundary leaks citations, the same way the entity→citing-page boundary leaked caveats

Scope: all 92 pages plus the source tree and both dossiers. Two inventories dispatched as sonnet subagents (mechanical link/orphan/index check, source-provenance map) and one comparison pass (contradiction sweep); all verification of their claims, and all prose, done here. Three subagent findings were checked and corrected before use — the mechanical run initially reported nine broken anchors that were explicit {#custom-id} overrides, and the provenance run twice mis-sorted bare-name citations as absent.

A second frozen week. Nothing under wiki/ or sources/ has changed since the 2026-09-14 lint. Page count holds at 92, HEAD still at 55709a3 (2026-09-05). ⟨Inference: with no new material, staleness can only be last week’s items aged by seven days — so this lint again spent its budget on the provenance and contradiction checks, which do not need new sources to find anything.⟩

Mechanical: clean for the sixth consecutive lint. 0 broken internal links, 0 broken anchors, 0 orphans, 0 weakly-linked pages, index coverage complete in both directions.

The finding: findings reach a dossier and stop there

The 2026-09-07 lint deferred the sources/ai-market/ provenance gap with a stated reason: the claim-to-paper mapping “is not mechanical … a subagent guessing which of eight papers carries which figure is exactly the failure mode worth avoiding.” That premise was wrong, and worth recording as wrong. The mapping was never a guess — Dossier 2 cites every one of those papers by sources/ path with page and paragraph numbers. The work was not a research pass; it was transcription from a document already in the repository. Three lints deferred a one-hour job by mis-describing it.

Doing it exposed the real defect, which is larger than the missing paths. Provenance degrades when a claim crosses from a dossier onto a wiki page. Three distinct losses, all in the same direction:

• Paths dropped. The dossiers cite by file path; the wiki pages distilled from them cite by case number or author name. ai-stack-market-structure.md carried the Autorité (24-A-05), the Bundeskartellamt (B6-26/23), the EC/CMA/DOJ/FTC joint statement, Widder/Whittaker/West, Witte & Krämer and Campbell/Goldfarb/Tucker with no path to a file that exists in sources/. Six paths added. On adtech-as-regulatory-precedent.md the 2026-08-31 lint converted ten author-year citations and missed four that are cited by instrument name rather than author — VEX, Addax, “the autobidding survey” — plus both Hovenkamp papers, left unlinked then because the prose cites them collectively; “in either paper” names both, so both are now linked. binary-verifiable-remedies.md names Hovenkamp’s JLI 7(1) 2024 unambiguously and was covered by that same blanket deferral; pathed.
• Citations dropped entirely. The Nvidia circular-spending figures ($2 bn into CoreWeave at $87.20, $2 bn into Nebius at $94.94) sat on the wiki as bare assertions with no citation and no marker. Both carry Nvidia press-release citations in Dossier 2. Restored.
• Findings dropped entirely — the serious one, now open task #9. Five papers were acquired, read, and used to carry an argument in a dossier, and never reached any wiki page: Akbarpour & Li’s credible-auction trilemma (Dossier 1, 2×), D’Annunzio/Russo/Shekhar on the adtech tax (4×), Göke & Weintraub on auction format change (2×), the CMA Update Paper 2024’s “over 90 partnerships involving GAMMAN firms” (Dossier 2), and Jacobides on AI ecosystem dynamics. The persona’s rule is explicit that a dossier is never the citable source; these five are that rule failing.

⟨Inference, and the part that generalises: this is the 2026-09-14 finding one layer up. There, a caveat did not survive the hop from an entity page to the pages citing it. Here, a citation does not survive the hop from a dossier to the wiki page distilled from it. Both are losses at a boundary crossing, and in both cases the receiving page looks complete on its own terms — which is why neither surfaces as a broken link, an orphan or a missing page. No mechanical check can catch this class; it needs a reader holding both documents. That is an argument for making it a standing step of the dossier workflow rather than a thing lints rediscover.⟩

Acquired and never used anywhere, dossier included: Bommasani’s Ecosystem Graphs, Chen’s FrugalGPT, Ong’s RouteLLM, the CMA Initial Report 2023, and the Bundeskartellamt’s Microsoft/OpenAI case summary (B6-34/23). Reported, not fixed. ⟨Inference worth one person’s attention: FrugalGPT and RouteLLM are both routing papers, and the routing layer is where ai-stack-market-structure.md locates its sharpest claims — “no agency document contains [a router taxonomy]; this is our own” and “no router publishes an auditable routing policy”. Those claims are about commercial disclosure, so an academic routing method probably does not refute them — but the page asserts a gap at exactly the layer whose academic literature sits unread in sources/, and I cannot tell from here whether the taxonomy survives contact with it.⟩

Two unacknowledged contradictions, both fixed

• The Upright caveat repair was incomplete in a systematic way. papers/ojala-consequences-alignment-problem-2026.md still called the red-team result “an independent, empirically-backed argument” — the exact phrase the 2026-09-14 lint retracted from cryptographic-trust-infrastructure.md as an overstatement. That lint swept the pages citing this material and skipped the paper pages the material originates on. Corrected, with the correction marked in place. The other two Ojala pages were checked and are properly hedged (Status of the evidence, Do not cite) — so it was one page, not a pattern across three. ⟨Inference: a repair that works outward from a claim’s citing pages will systematically miss the claim’s origin, because the origin is not in its own citation graph.⟩
• coaching-stack/coaching-agent-draft.md misread the table it cites as its authority. It claimed three of the four repertoire schools are primärquellenbelegt and only GROW is not; question-logic-comparison.md’s evidence-tier table puts only Cynefin and the Heidelberger Schule in the primary-text tier and places Co-Active in its weakest tier (a 34-page chapter excerpt, no dedicated academic critique, one blog source). Corrected against the table. This one matters beyond bookkeeping: the page is the specification for a coaching agent, so the error would have taught the agent to present Co-Active’s claims with a confidence its sourcing does not support.

Ten already-held-open contradictions were checked and correctly left alone (WHH/warrantage attribution, GIZ chair succession, Opitz’s departure status, Criterion H vs. C, the Heiss/Kleppmann architecture split, and others). The carbon-accounting and coffee/tender clusters came back clean.

Staleness: one gate now inside the window

• CCTI Milestone 1 falls 30.09.2026 and this is the last scheduled lint before that date. The wiki still holds no source on CCTI newer than the 2026-06-17 proposal draft, which records the evaluation use cases as “not yet selected”. Open task #8 updated to say so. ⟨Inference unchanged from two lints ago, and now load-bearing rather than tidy: the wiki cannot distinguish a gate about to be missed from one met without being reported.⟩
• GIZ Tender 10036847 — outcome unrecorded through four consecutive lints since the 21.08.2026 deadline. Bindefrist still runs to 13.11.2026. Status line updated; no prose rewritten.
• No relative-time language has rotted. The 2026-09-14 conversion to absolute dates held across the whole wiki — a scan turned up only false positives (“availability in fee days”, a quoted Wunderfrage). Second consecutive lint where this check costs nothing, which is the point of it.

Not fixed, deliberately

• The five concept pages still missing — EUDR, Neumann Kaffee Gruppe, FAO, IDH, BMZ. Unchanged since 2026-09-07; each needs a source not in sources/.
• Latham et al. 2021 still carries an ⟨Inference⟩ on adtech-as-regulatory-precedent.md with no file behind it. Recorded since 2026-08-31; the fix is to acquire the paper or drop the leg.
• External-URL citations on ai-stack-market-structure.md (LiteLLM, SEC, CME, OpenRouter, AWS and Microsoft docs, and now the two Nvidia releases) point outside sources/ and are not mirrored into it. The convention says every claim traces to a file in sources/; these trace to a live web page that can change or vanish. Not a lint fix — it is a schema question for the operator, raised below.
• Four terminal pages with no outbound links (ada-lovelace-institute, gs1, mozilla-foundation, wikimedia-creative-commons). Reachable and correct, just leaves. Fourth consecutive lint.
• ~272 ⟨Inference⟩ markers, nine checkable, six resolvable by asking one person. Unchanged for four lints.
• 12 undated pages. Cosmetic.

Open questions to the user

1. Was CCTI M1 met? Nine days out at this lint and nothing in the wiki.
2. The GIZ tender outcome — fourth time of asking.
3. Mirror web citations into sources/, or amend the convention? The AI market page depends on roughly a dozen live URLs. Either is defensible; the current state — a stated rule the wiki’s most recently written page does not follow — is not.
4. Who filed the FNG-Marktbericht, and should it be ingested? Third time of asking.

A commit is overdue by three sessions — 43 files: the 2026-09-07 lint, the 2026-09-10 Upright ingest, the 2026-09-14 lint, and this lint’s 9 edited pages. HEAD has not moved since 2026-09-05.

---

[2026-09-24] ingest | From Data to Decision: Leveraging Essential Variables in Standardizing Biodiversity and Ecosystem Services Monitoring and Reporting (Kim et al., 2026)

Handed over by Martin Pompéry with “import this scientific report”. Filed as sources/whitepapers/2026-09-24-kim-essential-variables-biodiversity-standardization.pdf. Conservation Letters, 2026 — the first source in this wiki on biodiversity and natural capital, and the first on a subject area SINE has no work in.

Three new pages, two updated, plus index and open tasks:

• papers/kim-essential-variables-2026.md — the summary
• concepts/essential-variables.md — the mechanics, with Tables 1 and 2 transcribed in full
• concepts/nature-related-reporting-frameworks.md — the demand side (KM-GBF, SEEA EA, TNFD, NPI, NGFS), built as the counterpart to product-level-carbon-regulation.md
• organization/geo-bon.md — deliberately a stub; see below
• updated: concepts/coordination-vs-bargaining-failure.md, concepts/data-commons.md, people-partners/stefan-dierks.md
• open-tasks.md item 10 (new), index.md

How it was read

As a position paper with a review’s apparatus. Nominally a synthesis of the state of essential-variable standardisation; actually the EBV community writing its own interface to two UN reporting regimes, published by the people who would operate it. The distinction is on the paper page and is carried through every claim taken from it.

What it is for, in a wiki about carbon

Not obvious on arrival, and worth recording because the answer was not the expected one. Four connections held up:

1. A third instance of the intermediate-layer architecture, next to PACT’s PCF and the ICC’s accountable quantities — reached from a fourth direction, biodiversity monitoring, by people who cite none of the others.
2. The near-pure coordination control case. 23 pages in which commercial confidentiality is never named and cryptography never reached for, because the reporting entities are states with no rent to divide. The diagnostic on coordination-vs-bargaining-failure.md had only mixed cases until now; this isolates one half and shows the instrument set behaves exactly as the table predicts. ⟨The finding that cost the most to accept: it also bounds SINE’s addressable market. In a pure-coordination domain, the distinctive technology is irrelevant and only the non-distinctive convening half transfers.⟩
3. Federation without governance. The paper’s §5.3 answer to data sovereignty is the Data Commons architecture, arrived at independently — with the entire Ostrom apparatus reduced to “data licensing, and user agreements”. Read on the Data Commons page as evidence about when that apparatus is load-bearing, not as an oversight.
4. The corporate demand side. TNFD / Nature Positive Initiative / NGFS, named in passing. Central banks entering on nature-related risk is the BIS/IAIS finding from the coffee map, generalised.

This also closes the white space flagged on stefan-dierks.md since 2026-08-17 — “biodiversity & natural capital is not yet a topic in this wiki”. It is now, and his guess that nature-related disclosure has carbon’s primary-data problem turns out to be half right: contradicted on the state side, unverified on the corporate side, which is where he sits.

One inference deliberately left unsupported

That confidentiality becomes a binding constraint the moment the reporting entity is a firm rather than a state. It is the sharpest thing the paper opens and it rests on no TNFD or NGFS primary source — Kim et al. mention them in one paragraph and nothing else is in sources/. Marked unverified in place and registered as open task 10 rather than written as a finding.

⚠ Method exception — Marker destroyed the landscape tables

Recorded per the wiki skill’s requirement. marker_single reported {'tables_pdftext': 1, 'tables_total': 5}: four of five tables fell back to the OCR path, which is disabled in this image. Tables 1 and 2 — the data- dimensions matrix and the SEEA EA crosswalk, i.e. the two most information- dense objects in the paper — came out as space-stripped word salad with their bodies missing entirely. Table 3 was partially legible.

Both are landscape-rotated pages. Recovered by rotating pages 5, 6, 9 and 10 with pypdf and re-extracting with pdfplumber (unrotated extraction returns character-reversed text). Table 3 never extracted cleanly in either rotation and its three rows were reconstructed from Marker’s partial pipe table, which was legible — this is the weakest transcription on the pages and is flagged where it appears.

Consequence for citation: Tables 1–3 on essential-variables.md and the paper page are cited against the PDF, not the Markdown twin, which does not contain them. Both pages say so. Everything else comes from the twin as normal.

Pending

A commit is now overdue by four sessions. HEAD has not moved since 2026-09-05; this ingest adds 1 source PDF + its twin folder and touches 9 wiki pages, on top of the 43 files already uncommitted.

---

[2026-09-28] lint | Seventh health check — the wiki is acquiring faster than it ingests, and the lint’s own bookkeeping was the stalest thing in it

First lint in three weeks to run against a wiki that moved. Four source bundles were filed 22–25 September; one was ingested. Mechanical layer clean for the seventh consecutive check: 0 broken links in wiki/, 0 broken heading anchors, 0 pages missing from index.md, 0 index links to nonexistent files, no orphans other than log.md itself (reachable from the index since 2026-09-14).

The finding: three bundles in the state item 6 calls exceptional

sources/standards/ (GHG Protocol Actions and Market Instruments Phase 1 RFI — 437 responses), sources/climate/ (IPCC AR6, seven files) and sources/foresight/ (CIFS Megatrends 2026) arrived and were never ingested: no page, no index entry, no log entry, no citation anywhere. Two of the three open subject areas the wiki has never covered.

Registered as open task 11, with the GHGP RFI first. The reason it goes first is not that it is newest — it is that it is the only one of the three that is not a new subject area. It is the foundation under everything the wiki already says about carbon, and it proposes splitting corporate GHG reporting into four statements, one of them consequential rather than attributional. That is the seam the ICC Review works, arrived at from the incumbent side, with Phase 2 live. A SINE response to ICC written without it risks arguing against a position the field is already leaving.

Ingesting it would also close the wiki’s largest page gap: the GHG Protocol is cited on nine pages and has no canonical page of its own. It is the most-cited external standard here and the only one with no entity page. Six lints missed it, and the reason is worth recording — a missing page is invisible to every mechanical check, because nothing can link to a page that does not exist. Orphan-detection finds pages nothing points to; there is no check for the concept everything points at in prose and nothing points to as a page.

The contradiction: one silence, two valuations, same day

The Kim ingest read the paper correctly as a position paper with a review’s apparatus, and then spent that caveat unevenly. data-commons.md discounts the paper’s silence — “its silence is weak evidence either way”. coordination-vs-bargaining-failure.md, where the silence does more work than anywhere else in the wiki, carried no hedge at all, and drew from it a conclusion that bounds SINE’s addressable market.

Both pages written 2026-09-24. Side by side they read as a contradiction. They are not, and the resolution is worth more than the fix: the two silences are differently motivated. Data Commons reads a silence about the authors’ own federation — a self-serving omission, correctly discounted. The coordination page reads a silence about an instrument set the authors have no stake in either way, and an unmotivated silence is better evidence than a motivated one. Written onto both pages, with the asymmetry argued rather than assumed, plus a scope bound the section did not have: one article by one community is not the biodiversity literature, so the market-bounding consequence is a hypothesis with one observation behind it.

Three lints running, the defect has been at a boundary crossing — a caveat lost from entity page to citing page (09-14), a citation lost from dossier to wiki page (09-21), and now a caveat applied on one citing page and not another. This one is the hardest of the three to catch, because nothing is missing: both pages are internally coherent and only disagree when read together.

Self-inflicted staleness, three instances

The most out-of-date claims in the wiki this week were ones the lint wrote.

1. Open task 8 said “the 2026-09-21 check is the last one that falls before the milestone date.” Wrong on its own arithmetic — the lint is weekly, so today also precedes 30.09.2026. The error retired the deadline a week early, and cost the one remaining ask. CCTI Milestone 1 is now two days out and the wiki still holds no source on it newer than the 2026-06-17 proposal draft. Corrected, and asked again.
2. Open task 6 called Dafoe “the only source in the repository in that state.” Three more joined it this week. Dafoe is now the oldest, at five weeks.
3. Open task 2 has said “next in the queue” for six weeks while four ingests went ahead of it. The real ordering rule is most recently handed over by Martin Pompéry wins. Named rather than rewritten — the item is still owed, and it is the prerequisite for item 1.

Smaller fixes

• papers/kim-essential-variables-2026.md: the 2026-09-24 log entry stated that Table 3’s reconstruction was “flagged where it appears.” It was not, on either page. The warning now sits above the table.
• index.md: open-task count 10 → 11.

Reported, not fixed

dossiers/2026-08-15-cooperation-failure-taxonomies.md:206 links to ../notes/2026-08-15-recon-cooperation-failure-taxonomies.md, described as “superseded in part; see the retraction in §7”. There is no notes/ directory in this repository. Left in place deliberately: deleting the link would erase the record that a round-1 note existed and was retracted, and the dossier’s §7 retraction is argued against a document nobody can now read. Whether that note was lost or never committed is a question for the operator.

Carried, unchanged

Five concepts without pages (EUDR, Neumann Kaffee Gruppe, FAO, IDH, BMZ) — now six with the GHG Protocol, which is the one that matters. The Latham et al. 2021 inference with no file behind it. External-URL citations on ai-stack-market-structure.md not mirrored into sources/. Four terminal pages. ~272 ⟨Inference⟩ markers. Five acquired-never-used AI-market sources, two of which (FrugalGPT, RouteLLM) are routing papers sitting unread while the routing layer carries that page’s sharpest claims.

Nothing has closed on open-tasks.md since 2026-08-17. Eleven open.

A commit is overdue by five sessions. HEAD has not moved since 2026-09-05; 53 uncommitted paths, now including four source bundles.

[2026-10-01] ingest | iLEAP Technical Specifications v1.1.0-draft.2

Requested by Martin Pompéry. The first primary iLEAP source in this wiki. iLEAP had been cited on seven pages with no page of its own — a gap the 2026-09-14 lint recorded and this ingest closes.

The source, and the detour to get it

The URL given first (raw.githubusercontent.com/ileap-initiative/specs/.../specs/index.bs) is dead — 404. The GitHub API could not be used to find the moved repository either (the proxy’s injected credentials returned 401 Bad credentials), so I probed eight candidate locations by hand; only sine-fdn.github.io/ileap-extension/ answered, and that is the old PACT-extension document, not the current spec. Martin supplied the live location unprompted: https://specs.ileap.global/. Worth knowing that the GitHub path is not a stable citation surface for this document.

A new source format for this wiki. Not a PDF, so marker_single does not apply; a Bikeshed-rendered HTML specification. Path used:

1. curl -sL the rendered page → sources/standards/2026-10-01-ileap-technical-specifications-v1.1.0-draft.2.html (774 KB, immutable original).
2. pandoc 3.10.1 (the static binary in tools/, not the system 2.17.1.1), -f html -t gfm --wrap=none → the .md twin.
3. The twin was unreadable — 3442 lines carrying 30k tokens of Bikeshed <span class="secno"> / self-link / data-link-type noise, over the Read tool’s limit. Stripped in place with a short regex pass (29,153 → 22,280 words). The twin was minutes old and not yet a citable source, so it was regenerated rather than treated as immutable; the HTML original was not touched. Cite against the .md.

What the document is

A wire protocol and data model, not a methodology — it borrows its quantification wholesale from ISO 14083:2023 + GLEC v3.1 and says so. Six data types in two categories (emissions: TCE, TOC, HOC, ShipmentFootprint, AggregatedReport; activity: TAD), four business cases, four data transactions. Version 1.1.0-draft.2, status Living Document, changelog-dated 2026-06-15; last stable release v1.0.0 (2025-05-27).

Five findings that changed existing pages

1. v1.1.0 decouples iLEAP from PACT at the protocol layer. A standalone iLEAP API (/v1/ileap/shipments|tocs|hocs|tad|aed, OAuth2 client credentials, mandatory pagination) is now required for conformance; the PACT Data Exchange Protocol binding is deprecated and scheduled for removal in v2.0.0. Cause: §7.1, “Since April 1 2026, the v2 family of the PACT Technical Specifications, including [PACTDX], is sunset.” — the first date in this wiki for the PACT v2 sunset, and from a source outside the PACT/WBCSD orbit.
2. Data Transaction 4 / AggregatedReport: the confidentiality gap, answered by aggregation. §3.4 names the constraint in the open — an aggregated report is shared instead of TCE-level data “due to privacy or competition concerns” — and §4.4 confirms the loss is permanent and intended. This is the sharpest field evidence the wiki has for the confidentiality gap, and it cuts against that page’s comfortable reading: the constraint was named, and the answer reached for was the lossiest of ICC’s three. No cryptographic mechanism appears anywhere in the specification.
3. Verification is out of scope and conformance is syntactic. §8: “Currently, iLEAP Conformance ensures interoperability and syntactic correctness.” 28 test cases, bilateral testing, nothing about whether the numbers are true.
4. TAD is the SME on-ramp — raw transport activity data from a carrier too small to calculate, with the calculation done by the receiving party.
5. SINE sits at the conformance chokepoint — the automated conformance tester runs at act.sine.dev, the reference API at api.ileap.sine.dev. ⟨Inference: the document never says who operates them; the domains are the only evidence.⟩

Pages written

Created: iLEAP (canonical entity page) · iLEAP Technical Specifications (2026) (source summary, incl. the draft’s own in-line Needs revision / pending community input advisements).

Updated: Pathfinder Network (new section: the v2 sunset and iLEAP’s departure) · Confidentiality gap in carbon accounting (addendum — diagnosis confirmed, mechanism is the lossiest one, cryptography not reached for) · Coordination vs. bargaining failure (the iLEAP classification now rests on a primary source) · SINE Foundation · index · open tasks (new item 12).

What this ingest could not settle

Open task 12, both halves, for Tell Münzing: whether the “Internet of Emissions Data” converges on one network or settles into sector networks joined by mapping tables; and why a standard SINE co-authored — the same SINE whose first project was an MPC benchmarking tool — contains no cryptography where it explicitly names a confidentiality constraint. The Smart Freight Centre partnership remains undocumented: this source documents the standard, not the partnership.

A commit is overdue by six sessions. HEAD last moved 2026-09-05.

---

[2026-10-05] lint | Eighth health check — a client engagement ran a full week outside the wiki, and an inference lost its marker on the way to a named reader

Mechanically the wiki is clean for the eighth consecutive check: 0 broken links, 0 orphans across 97 non-index pages, 0 index gaps, 0 broken heading anchors, 0 broken dossier links. Every finding below came from reading, not from a checker — which is now the settled pattern rather than a remark.

The finding: sources/hitachi/

Four files appeared on 2026-10-01. One is a real raw source (the Hitachi R&D Happiness Planet FIRA news release, .html + .md). The other two are our own work products — an interview question set for Hitachi Rail’s CSO drafted for Tell Münzing, and a German positioning memo written for Katharina Roehrig — filed into the immutable raw-sources layer. Registered as open task 13, with four problems separated by severity:

1. Wrong layer. The memo’s own second line says its basis is “die SINE-Seiten im Strategie-Wiki”. A wiki derivative filed as a wiki input. This is the inverse of the dossier rule — not a dossier mistaken for a source, but a deliverable placed where that distinction stops being visible.
2. Its basis is missing. Both documents cite a “2026-10-01 pre-analysis (six research streams)” that is in neither sources/ nor dossiers/. Second week running: the 2026-09-28 lint found a dossier citing a nonexistent notes/ file. One is an accident, two is a pattern — analysis is being produced, used, and discarded without landing.
3. Unsourced figures. 76 suppliers / 14.93 % of Scope 3 Cat 1 / 864,424 tCO₂e; the HTR 412 “Blues” SPD; “Enhanced by HMAX” from 2027; Hitachi Corporate a PACT participant while Hitachi Rail is not. None has a primary document here. The one ingested Hitachi source supports exactly one claim in the memo — the Engelbart intelligence amplifier framing.
4. An inference hardened on the way out. iLEAP says in an explicit marker: “The document does not state who operates these hosts; the domain is the only evidence.” The memo tells Katharina flatly: “SINE betreibt Konformitätsgate und Referenz-API.” Same for the Smart Freight Centre partnership, which the wiki flags as resting on a LinkedIn profile alone.

(1) and (4) compose into the thing worth watching. An inference leaves the wiki, loses its marker in a client document, and that document is filed into sources/. If anything later reads that directory as raw material — which is what the directory means — the inference returns as a sourced claim, laundered through our own prose. Nothing cites these files yet. That is the whole of the remedy’s window.

I did not move or edit the files: sources/ is immutable, so this is recorded, not corrected, exactly as the FNG (item 7) and AR6 (item 11) deviations were. Relocating them to dossiers/ is an operator decision.

And the plainest part: grep -ril hitachi wiki/ returns nothing. A live engagement with named counterparties produced a diagnosis in the wiki’s own vocabulary — “Verhandlungsversagen, nicht Koordinationsversagen” — and the wiki does not know. That diagnosis is a real addition: an OEM that can measure and can compute, blocked only by a supplier’s rational refusal to disclose to its bargaining counterparty, is the cleanest corporate-side instance of the bargaining-failure class on record. It exists in one German memo for one reader.

Draft status did not travel — and this time it happened in a day

The 2026-10-01 iLEAP ingest was careful: both the concept page and the source summary state plainly that v1.1.0-draft.2 is a Living Document, below Consultation Draft and below Release, with v1.0.0 (2025-05-27) the last stable release, and they list the spec’s own in-line “needs revision” advisements.

The pages updated in that same session dropped it. PACT states the v2 sunset and iLEAP’s departure as accomplished fact across a whole section; coordination vs. bargaining failure rests a sharpened claim on the specification without noting it is unreleased. Four previous lints found provenance decaying as claims aged. This one decayed within twenty-four hours, at the moment of propagation — which means the mechanism is not memory, it is the act of summarising for a neighbouring page.

Fixed on both pages, and split rather than blanket-qualified, because the two kinds of claim differ: the sunset date is a report about someone else’s published lifecycle, which a draft can state accurately (though no PACT/WBCSD source in this wiki corroborates it); the removal in v2.0.0 and mandatory standalone API are iLEAP’s own forward intentions in an unreleased document and can still change. Likely vs. announced-not-settled.

Smaller fixes

• Two different artefacts in this wiki are called “v2” — PACT Technical Specifications v2 (Feb 2023, sunset 01.04.2026) and PACT Methodology v2 (2023, the former “Pathfinder Framework”). The sunset attaches to the first. Nothing said so; disambiguated on the PACT page, with the open question of whether the two retirements are one event left open.
• The “40+ conformant solution providers” figure counts conformance against the now-retired v2 family and still read present-tense. Bridged forward to the sunset section; how many moved to v3 is unrecorded.
• Item 8, CCTI Milestone 1, is now past due — 30.09.2026, five days ago, and the wiki still records the evaluation use cases as “not yet selected”. Five weekly asks, none answered; the longest-running unanswered question in the file. Noted alongside it that detailed CCTI content did circulate this week — in the Hitachi memo. The information exists; it does not come here.
• Bundesbank climate roadmap 2026-27 (filed 2026-10-02) registered under item 11, which is now four un-ingested sources rather than three. Worth saying that this one is filed correctly: a four-line provenance header with source URL, retrieval date, extraction method, and an explicit “no PDF offered by the publication server”. That is the convention the AR6 bundle lacked — a documented absence instead of an unexplained one.

Carried, unchanged

Six concepts without pages (EUDR, Neumann Kaffee Gruppe, FAO, IDH, BMZ, and GHG Protocol — nine citations, no page); the Latham et al. 2021 ⟨Inference⟩ with no file behind it; ai-stack-market-structure.md’s external URLs unmirrored into sources/; four terminal pages; ~272 ⟨Inference⟩ markers; five acquired-never-used AI-market sources. Nothing has closed since 2026-08-17.

A commit is overdue by seven sessions. HEAD has not moved from 55709a3 since 2026-09-05.

---

[2026-10-12] lint | Ninth health check — a contract that starts in three days, and a backlog growing faster than it is read

Mechanically clean for the ninth consecutive check: 98 pages, 0 broken links, 0 orphans, 0 index gaps, 0 broken dossier links. Every finding below came from reading. ⟨Method note: the link checker initially reported seven broken heading anchors, all false positives — the pages use explicit {#id} anchors, which a naive GitHub-slug reimplementation does not see. Recorded so the next lint does not re-discover and re-dismiss them.⟩

The finding: the tender contract starts on Thursday

GIZ Tender 10036847 carries two dates that have sat beside each other since the 2026-08-13 ingest without anyone putting them in the same sentence: the contract term is 15.10.2026 – 15.10.2027, and the Zuschlags-/Bindefrist runs to 13.11.2026. A contract due to begin four weeks before the period for awarding it expires.

Six consecutive lints have asked the same question about this tender — did SINE submit? — and item 5 has stood unchanged since 2026-08-31. The date makes it a different question. Exactly one of three things is true: the award was made some time ago and nothing reached the wiki; the start date has already slipped; or the procedure lapsed. The wiki can distinguish none of them, and in the first branch SINE people are inside a live twelve-month contract from Thursday — 56 fee days for the Team Leader alone, Key Expert 1 for Tell Münzing — with not one line here about it. The question stops being did we bid and becomes are we delivering.

This is the lint’s recurring shape in its sharpest form so far. The facts were both present, both correct, and both ingested on the same day. Nothing was missing and nothing was stale. What was absent was the one sentence that puts them together — which no mechanical check can find, because there is nothing wrong with either half.

The tracking was current; the page was a month behind

CCTI Milestone 1 fell due 30.09.2026, twelve days ago. open-tasks.md has re-asked about it weekly since 2026-09-07 and the 2026-10-05 entry called it the longest-running unanswered ask in the file.

Meanwhile the CCTI page itself still carried the milestone in the future tense — “due by M1 (30/9/2026)” — and dated its last re-check to 2026-09-14. Four lints had logged the question without the page a reader would actually open ever being touched. The 2026-09-28 lint’s own headline was that its bookkeeping had gone stale; this is the inverse and the more dangerous version — bookkeeping kept meticulously current in the file nobody reads first, while the project page drifted a month behind it. Corrected on both passages.

Same shape, smaller, on Tell Münzing: a CV gap marked “blocking before 21.08.2026”, seven weeks after that deadline. The identical correction was made on the consortium page at the 2026-09-14 lint and did not travel the one hop to the person page. Also corrected: Michael Opitz’s “treat his current status as transitional until confirmed”, resting on a LinkedIn export from 2026-08-14 and describing a handover that began over three months ago — the two readings stay open, but they are no longer equally weighted.

The backlog grew while it was being worked

Two more un-ingested bundles arrived since the last lint, taking item 11 from four to six:

• sources/trade-policy/ (2026-10-06) — the French-German non-paper of 05.10.2026 on European resilience and market-distorting practices, plus two PDFs. A subject area the wiki has never covered.
• sources/ai-market/2026-04-09-prewitt-network-sourced-ai.md (filed 2026-10-09) — Matt Prewitt (OpenMined), What Is “Network Sourced” AI?

Since the 2026-09-28 lint named this item, one source has been ingested and three have arrived. ⟨Inference: at that ratio the list is not a queue, it is an archive. Four of the six have no identified consumer among existing pages, which is the finding — not the count. The honest options are to raise the ingest rate or to mark some of them deliberately-not-to-be-read, the same disposal item 9 already applies to dossier-only findings.⟩

The Prewitt essay was promoted to the front of the queue, and it is the one piece of genuine content this lint turned up. Data Commons currently records “two independent arrivals at the same architecture” — ICC from regulation, GEO BON from biodiversity. This is a third, from the AI-training-data side: private repositories “titrating” insights outward through a protocol layer enforcing attribution, consent and control, with stewards setting the rules — naming secure enclaves, differential privacy and compute-to-data, the last being Ocean Protocol’s own vocabulary.

And unlike the previous two, it disagrees. The essay puts governance second: “the first generation of network-sourced AI will run through technical guarantees”, with standards bodies and collectives arriving later to “tame the wild west”. The whole of Ostrom’s design principles is an argument that the governance layer is not a later accretion. ⟨Inference: the wiki has carefully recorded where its independent arrivals agree and has never yet found one that contradicts it on sequencing. A disagreement from an aligned party is worth more than another confirmation, and it should not be sitting on a backlog list.⟩ Not ingested here — ingest is a separate operation and takes a conversation.

sources/ discipline: the pattern recurred, with its own antidote

A third self-authored document landed in the immutable layer this week — sources/trade-policy/2026-10-06-recherchenotizen-kontext-non-paper.md, my own research notes compiled for Martin Pompéry. The structural problem in item 13(a) is therefore unchanged and now has three instances.

But it was filed the way the 2026-10-05 lint asked for, one day after that lint: it opens by declaring itself “eigene … Rechercheergebnisse … keine Primärquelle”, citable only as a research protocol; every claim carries a URL and retrieval date; the EUR-Lex URLs that hit a bot challenge are marked “nicht fetch-verifiziert”; unfindable items are left unfilled rather than filled in. ⟨Inference: that closes the laundering loop at the source. The Hitachi memo was dangerous because nothing inside it told a future reader it was ours; this one cannot be mistaken for a primary source and pre-emptively refuses the citation it would otherwise invite. Which reframes the operator decision in item 13 — the question is not move or leave, it is what header must a self-authored file in sources/ carry, and this file is a usable template for the answer.⟩

Carried, unchanged

• Concepts without pages. The six carried (EUDR 7 pages, FAO 6, NKG 4, BMZ 4, IDH 2, and GHG Protocol at 9 — still the largest gap) plus five the inventory added: CBAM on 12 pages, TNFD 6, CSRD 5, Smart Freight Centre 5, GLEC 4, ISO 14083 3. CBAM and TNFD are arguably covered inside composite pages. Smart Freight Centre is not, and is SINE’s own iLEAP partner resting on a LinkedIn profile — the second-largest gap after GHGP.
• 292 ⟨Inference⟩ markers across the 96 content pages (353 including log.md and open-tasks.md); the Latham et al. 2021 inference still has no file behind it; ai-stack-market-structure.md’s external URLs still unmirrored into sources/.
• 27 acquired-never-used source files, concentrated in ai-market/ (8) and adtech/ (3) — the latter three being item 9’s dossier-only papers.
• Three terminal pages (ada-lovelace-institute, gs1, mozilla-foundation).
• Nothing has closed since 2026-08-17.

A commit is overdue by eight sessions. HEAD has not moved from 55709a3 since 2026-09-05.

---

[2026-10-14] ingest | Sounding the Alarm (Inside Track x (Re)Insurance, 2026)

Requested by Martin Pompéry via Slack, with the explicit instruction to save a copy of the PDF as well. Source arrived as a LinkedIn feedshare URL; the full PDF was downloaded (1.59 MB, digital, 6 pp.) to a new sources subdirectory, sources/insurance/ — the insurance line had been named as a meta-lever since 2026-08-15 and had no source ablage of its own. Marker twin produced with --disable_ocr; clean conversion, the only artifact being spurious intra-word spaces, which is recorded as a method note on the source page so that quotations can be checked without confusion.

What it is. Six pages from an anonymous group of senior UK/European non-life insurance, reinsurance and broking professionals, addressed to government, customers and investors. No data, no models, no citations, no named author, and a direct interest in the conclusion it argues for — that government should absorb risk the industry intends to stop covering. Ingested as what senior people in this industry say about their own industry under anonymity, with the explicit instruction on the source page to discount the claims that run with the authors’ interest and to weight the ones that run against it.

The reweighting that shaped the ingest. The first plan filed the industry’s reliance on “a small number of base case models” as one bullet among four under a modelling page. Martin Pompéry asked how that observation had been addressed; it had not been, and the answer was that the bullet conflated two different objects — the “haircut” (an incentive failure inside a firm) and the base-case concentration (a structural fact about a supplier layer). On his go, the concentration became the spine of the NatCat model layer and the haircut became the second mechanism riding on it. That page carries the three reasons the structural claim does more work: correlated rather than averaged errors, a rule-setting layer outside the regulated perimeter (the adtech transfer), and concentration as an affordability artifact — the structure of the risk-information layer as a function of the price of data, which is SINE’s object and not the insurers’.

Pages written

• papers/inside-track-sounding-the-alarm-2026.md — source summary, the interested-party discount, the anonymity read as a mechanism, three findings that bear on SINE.
• organization/inside-track.md — entity page. A stub by design: the group is built to be unresearchable, so this is not a gap more reading closes. Its construction (anonymous outward, anonymous inward, plus an “aggregated and anonymised industry trends” disclaimer) is read as antitrust-shaped, which makes it a worked instance of Kartellrecht als übersehene Bremse from 2026-08-15.
• concepts/natcat-model-layer.md — the spine, as above. Holds open the tension it will not resolve: a shared base case is also how comparability is produced. Third instance in the wiki of that shape (after KM-GBF vs. SEEA EA, and Criterion C vs. Criterion H) and the first where the comparability-producing artifact is a priced commercial product — so the trade-off is not being made badly, it is not being made.
• concepts/insurance-protection-gap.md — reprice → restrict → withdraw, the 12-month renewal cycle as the mechanism that lets step 2 complete without anyone registering it, the gap as a negotiating position rather than a finding, and the GFC analogy marked as the memo’s weakest analytical move (collateral-value mechanism, not a contagion mechanism).

Pages updated, and what moved on them

• concepts/coordination-vs-bargaining-failure.md — third worked case, and the first run forwards: the page’s own rule, written on 2026-09-24, says confidentiality machinery in use signals a bargaining component. The memo’s anonymity is visible before any content is read; the prediction holds. What is new is that here the instrument was reached for and is a bad one — the group bought confidentiality by destroying attribution, which is why its own source page has to discount it. Demand demonstrated by substitution. Also carried: a limit on the distinction, from the same source. The three-step withdrawal fits neither column — every contract concludes correctly and the cost lands on non-counterparties, which is an ordinary externality. ⟨That cuts against the 2026-08-15 stance where the externality frame was contested in five points; the stance survives for the cases this wiki works on, but for a narrower reason than was given then — coordination and bargaining failures are the ones with an information remedy, and the honest diagnostic therefore has a third outcome.⟩
• context-maps/coffee.md — the §4 insurers block. The BIS/IAIS line is not refuted but is no longer the whole constraint: that source says the data infrastructure is missing, this one says the models exist, vary in quality, are commercially shaded and are too few. Both held open. And the insurance-first thesis is narrowed: better information does not make an uninsurable risk insurable — this memo describes an industry whose answer to a sharper view of risk is a faster exit. What it does is move the line, pricing an individual farm apart from its region. A claim about granularity, which is the same claim basis risk already makes.
• people-partners/anna-sophie-herken.md — the adjacency now has a live controversy to be adjacent to, and a caution attached: the memo says these questions “barely make the top three” inside insurers, so an insurance background is not evidence of an insurance-climate interest. The portfolio match stays structural.
• concepts/nature-related-reporting-frameworks.md — the “two supervisory bodies, one bottleneck” parallel survives, but “one bottleneck” is now its weaker half.
• index.md, open-tasks.md.

Raised

Open task 14 — how concentrated is the NatCat model layer, actually? The memo says “a small number” and gives no number, no vendor, no share. An entire page’s spine rests on that clause. ⟨Inference: it is also the claim from this ingest most likely to travel into a SINE-authored document, because it is the one that turns an industry anecdote into a market-structure argument — which is exactly why it should not travel yet.⟩ Recorded with an explicit prohibition: the wiki holds ⟨Background⟩ knowledge of who the major catastrophe-model vendors are, that is not a source, and no vendor name appears on any page from this ingest. Bounded research, roughly an hour, with Solvency II internal-model approval as the obvious place a supervisor would already have had to look.

Standing

Nothing closed. A commit is overdue by nine sessions — HEAD still at 55709a3 since 2026-09-05, now with four new pages and a new sources subdirectory on top of it.

[2026-10-15] ingest | Hitachi Integrated Report 2026

Ingested sources/hitachi/2026-10-15-hitachi-integrated-annual-report-2026.md (Markdown twin of the PDF supplied by Tell Münzing, converted with marker_single --output_format markdown --disable_ocr; 4,929 lines). Marker wrote into a subdirectory, so the .md was moved up beside the PDF and the image directory renamed to …-2026-figures/. Supplied as the document that joins Hitachi’s sustainability, digitalisation and organisational-transformation strategies, after Tell corrected my premise that only two Hitachi documents existed.

This is the first primary source in the wiki on the live Hitachi engagement. Everything the engagement rested on until today was SINE’s own work product filed into sources/ — open task 13, which this partly closes.

New pages: papers/hitachi-integrated-report-2026.md, organization/hitachi.md, organization/hitachi-rail.md, concepts/customer-zero.md. Updated: concepts/verifiable-carbon-accounting.md (new section “Hitachi as a worked gap case”), index.md (Organization, Concepts, Papers, and the lint note in the preamble), open-tasks.md (item 13).

Findings worth carrying:

• Customer Zero is Hitachi’s own term, not a framing SINE brings to it: “leveraging Hitachi’s own diverse business sites as a testbed to deliver value to customers faster” (CEO message, p. 7), repeated in Connective Industries and DSS. It dissolves SINE’s missing-reference-customer problem — under Customer Zero, having no external reference is the expected starting condition. It is absent from the Mobility section, so “Rail as internal first customer” is an argument to be made, not a fact to cite.
• EcoSpace (HMAX Energy, p. 14) is a sustainability-data platform already shipping inside HMAX. The “circular economy as a Lumada offering” argument therefore has a live sibling in another sector. Its verbs are quantify and visualise; verifiability is the missing layer.
• The group-level gap is sharper than Rail’s. No Scope 3 figure anywhere in the ESG data; the one compensation-linked supply-chain KPI counts partners who have set plans (428 of 700, FY2027 target); and the headline environmental claim — 280 Mt/yr avoided emissions — is ~600× the assured actuals (460 kt Scope 1+2, FY2025, limited assurance by EY ShinNihon). Strong where Hitachi owns the meter, silent where it depends on counterparties. This argument needs none of the memo’s unsourced figures.
• Three corrections to the fit memo. “Enhanced by HMAX” does not appear anywhere in the report, so the 2027 sensing-in-every-vehicle commitment remains unsourced in the repo; circular economy is a Connective Industries growth area, not a Mobility one; and Mobility’s 90 agentic-AI agents (RAILA / Project Jackie+) are in engineering, bidding and code, not procurement.
• The report never mentions PACT, product carbon footprints, digital product passports or material passports. SINE’s vocabulary is absent from Hitachi’s integrated report entirely.

Commit due: the new source PDF, its Markdown twin, the figures directory, and all wiki changes above.

[2026-10-15] dossier | Signierte Sensordaten als Customer-Zero-Fall (SINE × Hitachi)

Written at Tell Münzing’s instruction: frame signed sensor data as a Customer Zero case in Hitachi’s own operations, with a roadmap of the learning transferring to Rail and the “sensing in every new vehicle” commitment as a Physical AI business case — built on (a) the commitment with (b) the report’s evidenced facts as its foundation. Two sections he added at the checkpoint come first, ahead of the fit discussion: the shared logic of circular economy and sensor data, and Hitachi’s business strategy for HMAX and Lumada 3.0.

New: dossiers/2026-10-15-hitachi-customer-zero-signierte-sensordaten.md (German, [F]/[B]/[I] markers throughout). Six analysis parts plus a falsification section and an internal block.

The three load-bearing findings:

• Circular economy and carbon are one data problem. All three of Hitachi’s CE KPIs (landfill rate 1.8 %, plastic-waste utilisation 99.1 %, eco-design 100 %, p. 28) measure its own waste handling — the same inside-the-fence boundary as the carbon reporting. One proof mechanism serves both; only the circuit changes.
• Confidentiality sits on the critical path of Lumada 80-20. Growth is explicitly via “services to third-party assets” (p. 11); Mobility already runs 2,500 trainsets of which >35 % are not Hitachi-built (p. 15). Every step toward the 80 % ratio increases dependence on counterparty data.
• The first case is internal, not a supplier pilot. The STI-linked site decarbonisation KPI (75 % target, 64 % actual, p. 28) is computed from meters at Hitachi’s own plants — the exact Heiss setup — and cannot fail on counterparty willingness, which isolates the technical and governance questions.

[2026-10-15] ingest | Hitachi Rail — “Enhanced by HMAX” press release

Second source of the day, supplied by Tell Münzing as a URL while the dossier was being written, to answer the gap the morning’s ingest had flagged. curl → sources/hitachi/2026-09-24-hitachi-rail-press-enhanced-by-hmax.pdf, marker_single → Markdown twin plus -figures/ directory.

It closes the wiki’s most load-bearing unsourced claim. Verbatim: “all new rolling stock being equipped with proprietary HMAX sensor technology from 2027”, ETR1000 included; CEO Giuseppe Marino restates it. The fit memo’s InnoTrans attribution (22.09.2026) was exactly right.

New page: wiki/papers/hitachi-rail-enhanced-by-hmax-2026.md. Updated: organization/hitachi-rail.md (new sourced section; the register entry struck through and marked closed; status, cross-references, sources), papers/hitachi-integrated-report-2026.md (finding 4 annotated — the absence is a reporting-period artefact, the report closes 31 March 2026), open-tasks.md item 13, index.md (Papers, Dossiers, Organization).

Findings worth carrying:

• “Proprietary” is the word that matters. The sensor is Hitachi’s own design; the edge compute beside it is NVIDIA’s. Signing has to happen where measurement happens — so “sign at the sensor” is an internal engineering decision, not a supplier negotiation. This corrected the dossier, which had rested that argument on Hitachi’s edge AI semiconductors; those are not the Rail stack’s compute layer.
• The commitment covers new rolling stock only. No retrofit. The >35 % third-party fleet — where the Lumada leverage is — is not covered.
• A deadline is implicit. A fleet-wide 2027 commitment means the sensor specification is being frozen now; carrying material identity alongside condition data is cheap before the freeze.
• Sustainability is absent from the release. Energy consumption only; no material, origin, carbon or supplier vocabulary. The sensing layer is being built by people measured on downtime — which is the premise of the marginal-cost argument and the reason it needs an owner outside sustainability.
• An unresolved number. The release says HMAX sensor technology is on ≈2,500 trains; the Integrated Report says 2,500 trainsets on HMAX Mobility with >35 % built by others. Whether the third-party share is retrofitted with Hitachi sensors or managed on other data is not resolved by either document — a question for Hitachi.

Commit due: the new PDF, its Markdown twin and figures directory, the new dossier, the new papers page, and the wiki updates listed above.

[2026-10-15] ingest | Hitachi Rail — HTR 412 “Blues” Social Product Declaration (trade press)

Source: sources/hitachi/2023-12-13-rollingstockworld-hitachi-rail-social-product-declaration.md (+ .html twin), ROLLINGSTOCK Agency, 13 December 2023. URL supplied by Tell Münzing, 2026-10-15, in response to the dossier’s product-side gap.

Captured via agent-browser get text "body" and hand-formatted; the HTML is archived alongside. ⟨Trade press — secondary. The SPD report itself is still not in sources/.⟩

Filing decision: folded into Hitachi Rail rather than given its own papers/ page. It is a ~2-minute trade article with roughly six facts, all of which bear on one register entry; a separate page would be an orphan that restates the entity page. Noted here so the choice is auditable.

What it closes, partly: the HTR 412 SPD entry in Hitachi Rail’s unsourced register moves from “unsourced” to partly sourced — EPD International as partner, raw-material suppliers from 23 countries, a 1–4 risk scale, validity to 2028, and the stated ambition of a platform for data from more than 20,000 suppliers are now evidenced. What it does not close: the memo’s 38 / 8 / 11-of-26 supplier breakdown and the “very high risk” ratings for steel, aluminium and silicon — those sit in a figure image in the article and in the SPD report, which we do not hold.

Two register entries gained in passing: Hamburger Hochbahn’s 94 % recycling rate requirement in its 2022 metro tender (the repo’s only evidence that a buyer prices circularity contractually) and Hitachi Rail’s PAS 2080 certification claim of September 2023.

Findings worth carrying:

• The product declaration is the artefact that crosses the company boundary. Hitachi’s three circular-economy KPIs all measure its own waste handling inside the fence; the SPD is the instrument that reaches past it — and Hitachi Rail has already built one, socially.
• It records the data gap as a risk rating, in Hitachi’s language and under Hitachi’s signature. That is a slot a proof attaches to. SINE therefore proposes no new document, only an attachment to existing ones.

Also updated: the dossier dossiers/2026-10-15-hitachi-customer-zero-signierte-sensordaten.md gained a new section 5, “Was die Unternehmensgrenze überquert: Produktpässe und Deklarationen” (sections 5 and 6 renumbered to 6 and 7), two Kernaussagen, per-stage artefact lines in the learning path, two falsification bullets and three source entries — drawn from this source plus two wiki pages already held, product-level carbon regulation and digital product passport. Prompted by Tell Münzing’s observation that framing Customer Zero as a place left the product side unlit.

Commit due: the two new source files and the wiki and dossier updates above.

[2026-10-15] work product | Consolidated Hitachi question set

sources/hitachi/2026-10-15-questions-hitachi-consolidated.md — SINE-authored, requested by Tell Münzing so the engagement can be discussed as one whole. It merges the 1 October CSO set (21 questions, deep on the sustainability data layer) with the 15 October mandate-conversation set (10 questions, written for a business leader in the N+P / PCI advisor constellation), regrouped by what each question decides rather than by origin, with [CSO] / [MD] marks and seven ★ priorities.

Structural changes made in the merge, worth recording because they are judgements, not formatting:

• “Coordination failure or bargaining failure?” became its own section. Both sets contained its questions; only together do they read as one diagnosis with a binary outcome — and that outcome decides whether SINE is the right instrument at all.
• The HTR 412 SPD question moved from “parked” to a main question, on the strength of the trade-press source ingested the same day.
• Four questions are new, from reading Hitachi’s own strategy reading against the dossier: whether Customer Zero exists in Mobility at all, whether anything has completed the Customer Zero route, when the 2027 sensor specification freezes, and how the STI-linked site KPI is assured today.
• Two framing cautions added: hold back the gross-profit denominator (verified at p. 28, but it reads as an audit without a mandate), and state the SINE core as a mechanism rather than as governance.

Context, not filed: a third-party strategic reading of the Integrated Report (ChatGPT-authored, supplied by Tell Münzing 2026-10-15) was assessed but deliberately not taken into sources/ — Tell’s decision. Two of its figures were verified against the Integrated Report markdown we hold and are recorded here because they are now citable from the primary source: HMAX FY2025 revenue ¥300bn at 22 % adjusted EBITA margin (CEO message, verbatim) and the PLEDGES product-intensity KPI’s footnote “Per gross profit” (p. 28) — the latter meaning that KPI improves as gross profit rises even at unchanged absolute emissions. A revised version of that paper was written for Tell and lives outside the repository at /workspace/agent/hitachi/.

Commit due: the new consolidated question set and this log entry, on top of the earlier commit still outstanding.

[2026-10-16] ingest | Hitachi Sustainability Report 2026 (p. 36, standardization and CVP)

Source: sources/hitachi/2026-10-15-hitachi-sustainability-report-2026.pdf (163 pp.), fetched from hitachi.com on Tell Münzing’s pointer, 2026-10-15 23:43. Scope agreed with him as M: the standardization chapter, not the whole report.

What triggered it. Tell: “wir haben eine Feinheit übersehen … es gibt den CVP Ansatz, die Standardisierung im CE Bereich voranzubringen. Der Fokus auf ökonomischen Value sticht heraus.” My first answer was wrong — I reported that no such passage existed in Hitachi’s reporting, having searched only the Integrated Report. Tell corrected the source to p. 36 of the Sustainability Report. The two documents are published the same day and the report itself says they are meant to be read together; the integrated report is the capital-markets document and carries none of the rule-making apparatus.

Pages created

• Hitachi Sustainability Report 2026
• Circular Value-added Productivity (CVP)
• Hitachi-AIST CECRL

Pages updated

• Hitachi, Ltd. — new “Standardization as an apparatus” section; Scope 3 finding corrected (see below); “What is missing” rewritten.
• Hitachi Integrated Report 2026 — findings 1 and 5 qualified as document-specific, new finding 6 (no standardization or CVP content anywhere in it).
• Digital Product Passport — the Japanese rule-making track as a parallel, convergent instrument.
• Confidentiality gap in carbon accounting — Hitachi as a worked instance, stated by the party suffering from it.
• index.md, open-tasks.md (two new items, 15 and 16).

Correction made, and it was load-bearing. The wiki asserted “No Scope 3 figure appears anywhere in the report’s ESG data” as a structural finding about Hitachi. It is true of the Integrated Report and false of the group: the Sustainability Report carries 209.18 Mt-CO₂e Scope 3 against 0.46 Mt Scope 1+2 — 99.8 % of the footprint — with Cat. 11 at 191.46 Mt and Cat. 1 at 15.09 Mt. The finding survives in a better form: Hitachi publishes the quantity but does not govern it as one. Cat. 1 is spend-based with partial primary data, by Hitachi’s own description, and the reduction targets are intensities.

The substantive answer to Tell’s question. Hitachi converts environmental quantities into productivity ratios wherever it governs them, and says why each time: circularity → CVP, because existing CE metrics “lacked a cost-efficiency perspective, limiting their ability to incentivize investment” (p. 36); Scope 3 Cat. 1+11 → reduction per unit of added value, because absolute emissions “fluctuate greatly” and improvements are “difficult to see” (pp. 51–52, 64); Cat. 11 near-term → GHG per unit of gross profit. So his observation is evidenced as a repeated design choice, not a tone. The mechanism, stated as an inference on the CVP page: a metric built to trigger investment is defined over data the investor already holds, so the system boundary lands at the company boundary. Each conversion is individually defensible; the aggregate leaves no absolute quantity governed by anything.

The positioning consequence is the opposite of a critique. CVP answers is this investment worth making; a signed-data layer answers is the input figure true without disclosing the cost base. Hitachi’s rule-making apparatus — committee under the CTO, Rule-making Promotion Group, IEC SMB seat, METI awards — is a distribution channel SINE cannot build, which argues for entering as a contributor to their standard rather than as the author of a rival one.

Method note — conversion defect. This PDF’s text stream emits each line’s words in reverse order and flips hyphenated compounds. Marker inherits the defect; the Integrated Report from the same publisher converted cleanly, so it is the document, not the tool. A de-reversed reading copy with a provenance header is filed at sources/hitachi/2026-10-15-hitachi-sustainability-report-2026-dereversed.md. All quotations were hand-checked; citations are anchored to the PDF’s printed page numbers.

Left undone, deliberately: the other 160 pages (open task 16) and the AIST research report that would define CVP formally (open task 15).

Commit due: the new source files, three new wiki pages, five updated pages, index, open-tasks and this entry — on top of the earlier commits still outstanding.

[2026-10-16] ingest | Proposal for Social Implementation of the CVP (CE Association)

Source: sources/standards/2026-10-16-ce-association-social-implementation-cvp.pdf (19 pp., Hitachi / AIST / AIST Solutions, March 2026), supplied by Tell Münzing as an attachment with three questions: (1) situate the value-network perspective against our company-boundary thesis, (2) the role of the ISO and WBCSD partnerships, (3) digest the CVP Conference report. This entry covers (1) and (2); (3) is the next source.

New pages. papers/ce-association-social-implementation-cvp-2026.md, organization/circular-economy-association.md, concepts/value-networks-iso-59000.md.

Rewritten. concepts/circular-value-added-productivity.md — wholesale. The first version was written from Hitachi’s one-paragraph summary in the Sustainability Report and got both the inputs and the system boundary wrong.

Updated. organization/hitachi-aist-cecrl.md (published CVP output, named people, the ISO 59011 / TC 323 WG3 and WBCSD GCP routes, the undescribed “data model standards” half), people-partners/wbcsd.md (GCP / CTI v4.0 / CMP and the plan to position CVP inside the GCP), concepts/coordination-vs-bargaining-failure.md (a fourth worked case — the cleanest bargaining instance in the wiki), concepts/confidentiality-gap-in-carbon-accounting.md (the gap stated outright by standard-makers), concepts/digital-product-passport.md (retraction, below), index.md, open-tasks.md.

Two self-corrections, both from the same mistake. Earlier the same day I inferred CVP’s data requirements from a single paragraph of prose. Eq. (2) refutes the first inference: the inputs are sales, intermediate-goods costs, COGS and SG&A — not material composition, origin, prior use and remaining lifetime, and so the “CVP and the DPP need the same substrate” claim on the DPP page is withdrawn. Eq. (3) refutes the second: CVP is defined over a multi-organization value network, so “the system boundary gets drawn at the company boundary” is false as a statement about the metric’s definition. The thesis survives in its sharper form — the computable and believable boundary is still the firm, and the only remedies the authors offer are trust arrangements. Both corrections are recorded on the pages themselves rather than edited away.

Conversion. marker_single produced a clean twin except for Table 1, which its table processor destroyed. Recovered by re-extracting PDF page indices 9–11 with pypdfium2 (marker’s own interpreter) into tables-recovered-from-pdf.md, alongside the three equations in Unicode-normalised form.

Also decided (carried over from the Sustainability Report ingest). The full marker_single run on the 163-page Sustainability Report failed a second time, leaving no output and no log. Abandoned rather than retried: the twin would have inherited the PDF’s word-reversal defect anyway, and the de-reversed reading copy (sources/hitachi/2026-10-15-hitachi-sustainability-report-2026-dereversed.md) is the usable ingest artifact. Recorded here because it is a standing exception to the PDF-conversion convention for that one file.

Received, filed, not yet ingested. The H-AIST Circular Value-added Productivity research report (April 2026, 12 pp.) arrived mid-session as an attachment and is now at sources/hitachi/2026-04-13-h-aist-cvp-research-report.pdf. It closes the fetch half of open task 15. Its Eqs. (1) and (2) match what the white paper records, so nothing in the wiki needs correcting on its account yet.

[2026-10-16] ingest | Circular Value-added Productivity (CVP) — H-AIST CE Lab research report (April 2026)

Source: sources/hitachi/2026-04-13-h-aist-cvp-research-report.pdf, 12 pp., supplied as an attachment by Tell Münzing the same day. Converted with marker_single (clean except Table 2 and the three figures, all noted on the page). Closes open task 15.

New page. papers/h-aist-cvp-research-report-2026.md.

Updated. concepts/circular-value-added-productivity.md (worked values, Methods 1–3, the consolidation analogy, the §3.3 third-party collector, the authors’ four limitations), concepts/value-networks-iso-59000.md (five VN topologies, ISO/TR 59032 added to the standards table, the platform case), concepts/confidentiality-gap-in-carbon-accounting.md (the remedy named as a role, and the boundary on what verifiability fixes), concepts/coordination-vs-bargaining-failure.md (the fourth case strengthened: a number on the stake, the distributive purpose stated outright, the remedy specified as an institution still to be designed), organization/hitachi-aist-cecrl.md, people-partners/wbcsd.md, index.md, open-tasks.md.

No corrections forced. Eqs. (1)–(3) match what had been recorded from the white paper. One small fix elsewhere: CMP is an indicator within CTI v4.0, not a separate WBCSD instrument — the WBCSD page implied otherwise.

The substantive finding. §3.3 specifies “a trusted, impartial third-party organization” to collect participants’ figures and compute the network CVP “without public disclosure to the entire VN”, and leaves its placement open. The role is drawn and the instrument is not — the most direct opening the Hitachi material has produced.

The counterweight, and it is ours to state honestly. §4.1.1 concedes that the CE segmentation “can easily become arbitrary”, and §2.2 that intermediate goods costs must be reconstructed per company accounting practice. Neither is a confidentiality problem; no trustee and no protocol fixes either. Recorded as open task 18 so the wiki does not let this blur into the gap SINE’s work does close.

Also answered, on Tell Münzing’s question (recorded on the CECRL page and in task 17): the proposal is formally the Common Infrastructure Standards for Circular Business and CVP is its second half, the “Growth Level Assessment Standards”. Searched all four documents on this track for “passport”, “DPP”, “declaration”, “PCDS” — zero occurrences. ISO 59040 appears once, in an appendix survey, and only in the negative.

[2026-10-16] ingest | Driving Circularity Transitions (H-AIST CE Lab., June 2025)

Source: sources/hitachi/2025-06-19-h-aist-driving-circularity-transitions-cvp.pdf (4 pp.), fetched on Tell Münzing’s instruction from unit.aist.go.jp/hitachi-cecrl/news/2025/pdf/250619.pdf — reference [1] of the CE Association white paper, unlinked from the lab’s site and not search-indexed. Marker twin clean (only two logo images).

New page: wiki/papers/h-aist-cvp-overview-2025.md. Updated: concepts/circular-value-added-productivity.md (formula lineage 2025→2026; the missing investment term), organization/hitachi-aist-cecrl.md (Yordas Group commission; document table; name forms), people-partners/wbcsd.md (CTI identified in full; WBCSD route dated to June 2025; ISO demoted-then-promoted between documents), index.md, open-tasks.md (task 17: candidate 1 eliminated).

No corrections forced. Three findings: the programme’s evidence base is an unpublished 2023 Yordas Group assessment; the WBCSD/CTI route predates the study group by a year; and CVP has never had an investment term in any generation, which the comparative argument against CTI implies it does. Fifth source with zero quality/data-model/passport vocabulary — task 17’s cheapest candidate is now closed without an answer.

[2026-10-16] ingest | Third Open Forum report (H-AIST CE Lab., July 2026)

Source: sources/hitachi/2026-07-30-h-aist-ce-lab-third-open-forum.md — rendered text capture of https://rd.hitachi.com/_ct/17854147 via agent-browser (no PDF edition exists; Tell Münzing’s pointer called it the “CVP Conference” report). ~8,100 words.

New pages: wiki/papers/h-aist-third-open-forum-2026.md, wiki/concepts/residual-value-grading.md. Updated: organization/hitachi-aist-cecrl.md (three research themes — CVP demoted to one of two outputs of one theme; a correction), concepts/circular-value-added-productivity.md (the magnet-remanufacturing case study on real data, held in tension with the missing investment term), concepts/digital-product-passport.md (the upstream/downstream division of labour; the CVP-only scope of the earlier correction made explicit), organization/circular-economy-association.md (the personnel union with ISO/TC 323), concepts/value-networks-iso-59000.md (ISO 59011 confirmed from inside the committee), concepts/confidentiality-gap-in-carbon-accounting.md (third addendum: METI and Hitachi Green DX state the gap), index.md, open-tasks.md.

Open task 17 substantially closed: the undescribed half of the H-AIST proposal is residual-value grading, aimed at the IEC Common Data Dictionary — which is why five CVP documents never mentioned it. Residue: no document exists, and nothing says who attests a grade.

One correction recorded (the lab’s shape). One tension left open rather than resolved (real-data validation vs. no capital term).

[2026-10-16] dossier | Revision: Signierte Sensordaten als Customer-Zero-Fall (Fassung 2)

Revised dossiers/2026-10-15-hitachi-customer-zero-signierte-sensordaten.md with the day’s two ingests (June 2025 CVP white paper, July 2026 Third Open Forum report). New part 5 on the H-AIST CE Lab: its three research themes, Kamigaki’s plastics-recycling study with its two information failures, the three requirements and the “uncertainty-related costs” argument (verification that pays for itself on the side performing it), Shinkai’s and Mimaki’s quotations, the CVP-vs-residual-value-grading split, the lab’s own downstream sensor work, and the association→ISO personnel union as the access route. Part 6 re-cut around Hoshino’s upstream/downstream/CDD division of labour and Nakamura’s keynote, with the attestation layer named as unoccupied in both architectures; the first edition’s scope explicitly corrected rather than overwritten. Added a parallel stage-3 track on secondary materials, three falsifiers, three sources and six next steps. 5,873 → ~8,700 words, 14 → 19 pp.; PDF re-rendered; wiki/index.md dossier entry updated.

[2026-10-19] lint | Tenth health check — the week that disproved the backlog’s excuse, and a safety net that was never there

Mechanical layer: the first breakage in ten checks. 117 pages (114 content pages plus index, log and open-tasks), 0 orphans, 0 index gaps, 3 terminal pages (ada-lovelace-institute, gs1, mozilla-foundation — unchanged for six lints) — and two broken anchors, the first this check has found since it began. Both pointed at organization/hitachi.md#ecospace-the-precedent from customer-zero and the Integrated Report page. The heading is ## EcoSpace — the precedent, and an em dash slugs to a double hyphen: #ecospace--the-precedent. Fixed in both places.

⟨Worth recording as a method note, since nine clean checks had made this layer feel decorative: the breakage arrived with the Hitachi burst, in pages written at speed, in the one construction — em dash inside a heading — where the slug does not look like the heading. Volume is when the mechanical check earns its keep, and this wiki had its heaviest week.⟩

The carried items, and what four days of arithmetic did to them

The GIZ tender’s contract term started on Thursday. The last lint flagged 15.10.2026 as three days out; it is now four days past, and nothing arrived in between. The item’s question has changed shape twice in eight days: “did we bid” → “are we delivering” → “has a twelve-month engagement been running since Thursday”. ⟨Inference: of the three branches, award made and the wiki not told now carries a weak counter-indication — an award that held the advertised start date should have left operative traces by 15.10., and none are visible. The wiki sees almost nothing of this procedure, so that is an indication, not evidence. What has changed is the cost of the gap: a recording gap about a future date costs nothing, one about a contract four days into performance is how deliverables get missed.⟩ Eight consecutive lints, one unanswered question. Banner and status line on the tender page updated from in drei Tagen to seit vier Tagen überschritten; item 5 rewritten.

CCTI Milestone 1 is nineteen days past due — seven weekly asks, none answered, now the oldest unanswered question in the wiki. Dates refreshed on the CCTI page (two passages) and in item 8.

The finding: the backlog’s excuse died this week

Between 14 and 16 October the wiki ingested eight sources in four days — the heaviest run in its history, roughly nineteen new pages. None of them was from the un-ingested backlog. The seven items there have now been passed over by a run four times their size.

⟨Inference: whatever keeps that backlog unread, it is not throughput, and every previous lint that framed it as a rate problem was framing it wrong. The distinction that actually separates the two piles is whether a source has a requester. Every document ingested since 2026-10-14 arrived from Tell Münzing with a decision hanging on it. Every document on the backlog was acquired by me against a general sense that it would matter — IPCC AR6, CIFS Megatrends, the Bundesbank roadmap, the GHGP AMI RFI, the trade-policy non-paper. That points at retirement for most of the list rather than at scheduling, and it makes the standing question — what are these for? — the only one worth putting to the operator.⟩

And the count was wrong: seven, not six. The Hitachi R&D Happiness Planet FIRA release has been in sources/ since 2026-10-01, un-ingested, and was never on the list — it landed inside sources/hitachi/ and was therefore described in item 13’s prose rather than counted in item 11. The backlog was split across two items by the accident of which directory a file fell into, while item 11 is the only page claiming to hold the whole count. Added, with the heading corrected to Seven un-ingested sources. ⟨It is also the one Hitachi source the Hitachi week did not take in — eight documents were ingested because someone sent them with a question attached, and this one sat because nobody asked anything of it. The same variable, visible inside a single directory.⟩

Prewitt stays first and is the exception that proves the rule: the only backlog item that contradicts a live wiki position — governance sequencing, against the Data Commons and Ostrom pages — rather than merely adding to one. A disagreement is a requester of a kind. One week at the front of the queue, unread.

Item 13 inverted, and three of its four problems survived

Retitled today, because “a live Hitachi engagement that exists entirely outside the wiki” stopped being true: Hitachi is now the wiki’s largest single cluster. The original complaint is answered in full. What survived:

• (c) partly closed, and the item’s own 2026-10-15 update is one ingest out of date — it records the HTR 412 SPD as “still owed” when the SPD was partly sourced later the same day via the ROLLINGSTOCK trade article. Corrected, with the genuinely still-unsourced list restated: the 38 / 8 / 11-of-26 supplier breakdown, the “very high risk” ratings, the ISO 14067 PCF, the 76-suppliers / 864,424 tCO₂e figure, and Hitachi Corporate’s PACT participation.
• (a) recurred a fourth time, three days after this lint proposed the fix. sources/hitachi/2026-10-15-questions-hitachi-consolidated.md is SINE-authored, filed into the immutable layer, logged honestly as a work product. Its header gives who, when and for whom — three of the five elements named on 2026-10-12 — and omits the two that do the work: what it may be cited as, and what in it is unverified. ⟨Inference: not carelessness, but the predictable fate of a template that exists only as a sentence inside a lint entry. The Recherchenotizen header worked because its author had just been told. The next author will be told by the file layout or not at all — which argues for a sources/README.md and turns the operator decision from move-or-leave into move-or-leave plus a required header.⟩
• (b) has its first located instance rather than a lost one: the revised strategy paper for Tell Münzing sits at /workspace/agent/hitachi/ — outside the repository, so outside its history, but findable. Three instances in three weeks is a pattern, not three accidents.
• One counter-example, recorded because it is the discipline working. A third-party (ChatGPT-authored) reading of the Integrated Report was deliberately kept out of sources/ on Tell Münzing’s decision, and the two figures worth keeping were verified against the primary document first. That is exactly what (a) and (d) ask for, done by the operator unprompted.

A safety net that was never there

The Sustainability Report page carries a section headed Conversion defect — read this before citing, which stated that “the faithful Marker twin is kept as the mechanical conversion, and a de-reversed reading copy is stored alongside it”. The faithful twin is not in sources/. The directory holds the 163-page PDF and the de-reversed copy and nothing else; no …-2026.md, no Marker output directory. The claim was about the repository’s own contents, on the page that says read this before citing, and it was false.

⟨Inference: the practical consequence is small but exactly inverted from what the page implied. The quotations were spot-checked against the PDF’s printed page numbers — the right check, and the only one available — while the page advertised a mechanical twin as the fallback. A reader following the page would have gone looking for a file that does not exist. Corrected in place, and the de-reversed copy is now named in the Sources section, which listed only the PDF: the one file needed to re-check a quotation was absent from the source list of the page built on it. Re-running marker_single would restore the twin and is cheap, but adding a file to sources/ is an ingest-shaped decision, so it is recorded rather than done.⟩

Carried, unchanged

• ⟨Inference⟩ markers: 408 on content pages, up from 292 a week ago — a 40 % rise in seven days, all of it from the Hitachi cluster. ⟨Inference: the marker discipline held under the heaviest load the wiki has seen, which is the good news; the unverified-claim stock grew by a third in a week, which is not. Nothing in the process converts a marker into a check.⟩
• 27 source bundles acquired and never cited by any page outside index/log/open-tasks. Unchanged in count: the Hitachi week consumed several and added about as many.
• Concepts without pages — the ranking changed. Scope 3 is now the largest gap by a distance: 45 mentions across 19 pages, no canonical page, and it is the quantity on which the Hitachi analysis, PACT, iLEAP and the confidentiality gap all turn. Then CBAM (27 / 12), ISO/TC 323 (17 / 9), EUDR (16 / 7), GHG Protocol (14 / 8), Smart Freight Centre (6 / 4).
• A fifth thing at the repository root. figures/ appeared on 2026-10-16 with a two-file sketch, is referenced by no page in wiki/ or dossiers/, and is untracked by git. The declared layers are sources/, wiki/, dossiers/ and the schema; this is none of them.

Commit status: the eight-session backlog was cleared. HEAD is 766fbc9 (was 55709a3 since 2026-09-05). Uncommitted after this lint: seven modified wiki pages plus this entry, and the untracked figures/.

[2026-10-26] ingest | Agentic Communication Networks (Li et al., 2026)

Source handed over by Martin Pompéry as a bare arXiv link (2026-10-25, “bitte importieren”). Downloaded in full to sources/whitepapers/2026-10-25-li-agentic-communication-networks.pdf and converted with marker_single --disable_ocr; ingested from the Markdown twin. One of six tables went through pdftext, the rest are the table of contents and figure captions — nothing load-bearing lost. Eight figures are scenario diagrams, extracted as JPEGs and not transcribed.

What it is. An arXiv white paper, eight academic authors across China, France, Germany, the USA and Italy, submitted 20.10.2026. No results, no implementation: a term being staked on a future standardisation agenda (3GPP and ETSI are in its references). The first author carries the BPP / “New IP” lineage, which the paper names as a candidate network-layer realisation and whose contested history in ITU and IETF it does not mention.

Pages written. One source summary (li-agentic-communication-networks-2026) and two concept pages (agent-identity-and-delegated-authority, agent-protocol-interoperability), each carrying one half of the source. Updated: CCTI (second external signal), cryptographic trust infrastructure (a named alternative to be measured against, with the trust-anchor comparison stated both ways), confidentiality gap (fourth addendum), plus cross-references on iLEAP and PACT. Index and open tasks updated; open task 19 raised.

Why three pages rather than one. Proposed as a scope question to Martin Pompéry before writing. It resolved itself in the discussion: both halves of the paper turned out to be arguments SINE has a position on, and neither reduces to a paragraph on a source summary.

The correction that shaped the ingest. The first reading reported to the channel claimed §7.4 names the gap the EIC fills. Martin Pompéry rejected that on 2026-10-26 as a misreading of the EIC, and he is right: the EIC is an ex-ante commitment device attacking bargaining failure — commit beliefs privately, match over them, negotiate further evidence — while §7.4 states an ex-post attribution problem across a delegation chain. The slide ran through “both concern an agent’s reasoning”, which is true and does not make them the same object. The corrected reading, now on the concept page: the CCTI objects that bear on §7.4 are WP-A’s business-data attestation and recursive proofs, the latter being structurally what a delegation chain needs.

The substantive finding, put by Martin Pompéry as a question — is accountability treated here as an identity-management problem? Yes, predominantly. §2.5 files accountability under the entity level beside naming and registration; §7.1 absorbs it into “From Device Identity to Agent Identity” and closes on an identity-shaped principle; §4.5 and §5.5 deliver it as authentication plus access control plus tamper-evident logs. The tell is a terminological slide in §7.1 between accounting (resource consumption — genuinely identity plus metering) and accountability (answerability — not). The paper escapes the frame exactly once, in §7.4 — “accountability cannot be established solely from packet logs … ACN needs an auditable relationship among authority, decisions, actions, and outcomes” — and the next paragraph returns to standardised audit records. Diagnosis beyond identity, therapy within it. Worth stating that way rather than as “it ignores accountability”, which it does not.

Second finding, elaborated at Martin Pompéry’s request: §5.4’s O(n²)→O(n) claim for a protocol-translation gateway is SINE’s own common-data-model argument in a domain where SINE has no voice. Three things it has to carry — semantic completeness of the intermediate representation, the institutional centralisation that buys the saving, and a translation path proposed as probabilistic (embedding alignment, LLM-assisted parsing) running straight through the authorisation boundary. The PACT/iLEAP case is on the page with its honest reading: the split’s stated cause was PACT’s own v2 sunset, not semantic failure — what the case shows is that semantic interoperability was never achieved or claimed in the first place, which is the premise the PIIR assumes.

Also noted, not pursued: the architecture depends throughout on heterogeneous agents converging on a shared capability vocabulary without a central registry, and never asks how. An unnamed naming game — a usable setting for the foundational-games research requested on 2026-10-23, which remains unstarted.

Commit due. Three new pages and eight modified files, uncommitted.

---

[2026-10-27] ingest | OECD, Data Spaces between Co-ordination Challenges, Financial Viability and Public Policy

Requested by Martin Pompéry in Slack with the URL. Downloaded to sources/research/2026-10-27-oecd-data-spaces-co-ordination-viability-public-policy.pdf (2.97 MB), converted with marker_single … --disable_ocr to a 951-line Markdown twin in the sibling folder; ingested and cited against the Markdown. OECD Digital Economy report, declassified by the Digital Policy Committee 10 June 2026, drafted by Christian Reimsbach-Kounatze with Shunya Muromachi and Maiko Meguro under Clarisse Girot, funded by voluntary contributions of Japan, CC BY 4.0. Takeaways were sent to Slack before writing; writing proceeded without a reply, with the message that the plan could be redirected.

Conversion defect. Marker emptied Table 3.1 (comparison of data-sharing mechanisms and enablers — all but the first column lost) and Table 3.2 (governance configurations — body lost entirely). Both are reconstructed from the body prose, which states the same content in text, and every reconstruction is marked as one on the page that carries it. Not hand-restored from the PDF: unlike the CE Association and GEO BON cases, the prose here is complete enough that re-extraction would add presentation rather than content. Recorded as a caveat on the paper page.

Five new pages, eight updated, three catalog files touched.

The spine. This is the first committee-level external assessment of the data-space field in the wiki, and it is both the strongest independent corroboration of the Data Commons thesis and the sharpest available critique of the field SINE operates in. It reaches “governance structure, not technology, is the binding constraint” by derivation rather than assertion — Arrow’s information paradox and context-dependent value kill price formation; competitive and jurisdictional relationships kill centralised co-ordination; what remains is Williamson’s hybrid — and then reports that the field built on that insight has 235 registered initiatives, 128 at implementation, 25 operational and none scaling, against EUR 1.7 bn of EU public money.

Four findings that changed wiki state.

1. A boundary on our own thesis. The four conditions (three+ parties; competitive or jurisdictional relationships precluding a trusted intermediary; non-excludable collective value; ongoing flows) are the first disciplined statement in this wiki of when the Data Commons architecture is not the right answer. New page: concepts/data-space-boundary-conditions.md, with a first pass at scoring PACT, CCTI’s reinsurance case and the GIZ tender against them — PACT scores on all four, the reinsurance case turns on condition 3, and the GIZ case satisfies condition 2 in a different register (no actor with the legitimacy to hold the data, rather than competitors who distrust one).

2. Open task 19 has its counterexample. “Done looks like” asked for a document that does consider and reject the cryptographic instrument. The OECD considers it (Box 3.2: federated learning, MPC, TEEs/Data Control Rooms, synthetic data), makes supporting it one of eight policy considerations (§5.5, “Neither alone is sufficient”), and cites the G7 Industry, Digital and Technology Ministerial Declaration of December 2025 — and does not reject it. Effect: “not known” falsified at policy level; “not deployable” standing but on a governance barrier, since the named obstacles are cost, expertise and “the need for co-ordination among participants on which technologies to deploy”; “not wanted” unsupported. A fourth reading is now best-supported — sequencing: confidentiality is the second problem, the first is who governs and who pays, and a community stuck on the first has nothing to deploy a PET into. That is uncomfortable for a cryptography-first pitch and comfortable for SINE’s actual governance-first position. Item 19 narrowed, not closed; the revised test is iLEAP, where the governance arrangement is settled.

3. A funding contradiction on IDSA, represented not resolved. §4.3 says the IDSA-RAM “originated from publicly funded research (approximately EUR 10 million from the German Federal Ministry of Education and Research)”. The wiki holds €4.876 million for BMBF 01IS15054, from a FragDenStaat FOI response — the highest-confidence funding figure in the research set. Both are on market-competitors/idsa.md with the inference that the OECD figure may be an unsourced aggregate across several grants, and the note that it cites no project number.

4. The field’s central statistics are endogenous. The EUR 1.7 bn figure comes from a Gaia-X AISBL portfolio published by Otto and Tardieu (2025) in Le Grand Continent, and the OECD states in its own endnote that “both authors hold or held board positions at Gaia-X AISBL and IDSA”; there is no published methodology. The registry behind every count is the IDSA Radar, which IDSA operates. Boris Otto is in both, and already on record here as chairing at IDSA and Catena-X. Recorded on the IDSA and Gaia-X pages as a caution about independence of evidence, not about good faith — the OECD itself treats the number as usable.

New pages. papers/oecd-data-spaces-2026.md; concepts/data-space-boundary-conditions.md; concepts/bootstrapping-problem.md (two levels, the club-good structure, DGA neutrality disabling cross-subsidisation, Frischmann’s binding-condition test, five cases scored — Fintraffic’s measured 24 % CO₂ reduction and EUR 45 m/yr extrapolation, ABtC’s regulation-as-simultaneous-return, BoostAeroSpace’s eight years against a four-year grant cycle, Agdatahub’s November 2024 bankruptcy, Covisint); concepts/data-space-business-models.md (the marketplace model dying on three independent signals, the funding critique, the demand-side instruments, AI demand as the viability path, the unanswered IP question); people-partners/oecd.md.

Updated. market-competitors/idsa.md (Radar section, funding contradiction, membership count now four-way inconsistent, the EUR 1.7 bn provenance, ISO/CEN standardisation of DSP/DCP); market-competitors/catena-x.md (OECD case study — 15 May 2021, 17 founders incl. ARENA2036, 193 members Nov 2024, three-part structure, 10 Golden Rules, the March 2025 IPA PoC — with the note that the OECD’s design assessment and Handelsblatt’s adoption reporting are not in contradiction); market-competitors/gaia-x.md (the EUR 1.7 bn figure is Gaia-X’s; the Förderwettbewerb disbursed EUR 117.4 m of EUR 200 m and closed March 2025; Agdatahub dated and diagnosed; X-Road 8 converging toward Gaia-X); concepts/data-commons.md (third independent arrival, and the first with the governance layer intact and an economic derivation behind it); concepts/ostrom-commons-governance.md (data as the provision problem with appropriation removed; the eight principles scored against the cases — P2 congruence is where the failures sit, P4 without P5 is the enforcement layer; and external funding as a failure channel Ostrom did not name); concepts/coordination-vs-bargaining-failure.md (fifth worked case; the OECD restates the distinction as price-formation failure vs. asymmetric cost-value distribution, and Agdatahub exposes a third category the diagnostic lacks — a provision failure, worth having collectively and not worth paying for individually); concepts/cryptographic-trust-infrastructure.md (a three-way anchor comparison — ACN’s audit log, the data space’s contract-plus-connector, this stack’s signing sensor — and the OECD’s objection to us, which is “agreeing which PET”, a standards problem); concepts/confidentiality-gap-in-carbon-accounting.md (the counterexample addendum, and the migration from prevention to “deterrence through accountability” visible in adoption data: usage-policy blocks 35 %→28 %, provenance 25 %→48 %); concepts/digital-product-passport.md (product regulation as a bootstrapping instrument — it raises every anchor’s private return at the same moment — plus the Digital Omnibus risk and the EN 18235 / Data Act Art. 33 track).

Catalog. index.md — new Papers, Concepts (×3) and People entries, and the Open-tasks bullet rewritten; it had read “Seventeen as of the 2026-10-26 ACN ingest” while naming item 19, an internal inconsistency now corrected to nineteen. open-tasks.md — item 19 status changed to “open, narrowed” with the update section and a revised “done looks like”.

Not pursued, recorded here. The report flags twice that usage control “does not resolve who owns the value derived from” pooled data — model ownership, derived insights, downstream patent claims — with the Materials Genome Initiative (>USD 1 bn federal, persistent industry reluctance over IP attribution) as the worked case, and states that no widely adopted commercial model exists. That is adjacent to but distinct from the confidentiality gap: can my input be seen versus who owns what was learnt from it. Cryptography answers the first and is silent on the second. No page; a candidate.

Backlog unchanged. The Anthropic S-1 was downloaded on 2026-10-26 and has no wiki pages — still un-ingested.

Commit overdue. Now two ingests deep: the ACN ingest’s three new pages and eight modified files, plus this ingest’s five new pages and eleven modified files, all uncommitted.

[2026-10-29] query | German institutes with a profile comparable to IÖW

Asked by Martin Pompéry (Slack) to find German organisations with a profile comparable to IÖW. Scope gate agreed at tier M: a named candidate set, one fan-out round, no new primary sources. Six scoped Sonnet subagents surveyed eleven institutes plus the Ecornet network; IZT was covered the same day by a parallel chain and is not repeated.

Research filed immutably as sources/research/2026-10-29-german-sustainability-research-institutes.md with a provenance header and each subagent’s own limitations retained.

The finding: the peer group is already institutionalised. IÖW is one of eight members of Ecornet — independent, non-profit sustainability research institutes, ~800 staff combined, predominantly third-party funded, founded 1977–1995 “außerhalb des etablierten Wissenschaftssystems”. The other seven: Ecologic, ifeu, ISOE, IZT, Öko-Institut, UfU, Wuppertal Institut.

The cross-cutting inference: across all eleven institutes, the peer group researches data governance (Ecologic), builds methodology (ifeu/BISKO) or runs pilots (Wuppertal/DPP). None builds infrastructure — software, wire formats, conformance gates. That is SINE’s structural position against this group, and the reason the two closest institutes read as cooperation candidates rather than competitors.

New page on Martin’s follow-up (“Warum nicht Ecologic Institut ebenso mit in das Wiki aufnehmen?”): market-competitors/ecologic-institut.md. Index updated with a new “German sustainability research institutes (Ecornet peer group)” subsection, which also carries an explicit list of the pages this survey justifies but that were not written — Ecornet itself and the Wuppertal Institut being the two with the most behind them.

Research quality, recorded rather than smoothed over: ifeu.de blocks automated fetches (403, snippets only); foes.de was down (502, FÖS rests on background knowledge); Wuppertal headcount is contradictory across sources (250 vs. 320) and is carried as a contradiction; adelphi’s own site states neither founding year nor size. UfU was not researched — the one Ecornet member with no profile in the note.

[2026-10-29] query + ingest | IÖW — person-level overlap with SINE

Asked by Martin Pompéry (Slack): analyse www.ioew.de for thematic overlap with SINE, specifically at the level of individual people. Scope gate agreed at tier M. Follow-up the same day: “Analysiere noch die 3 Personen, und schreibe dann den Wikieintrag” — which turned a query into an ingest.

Method, and a deliberate departure from the gate. The proposed Sonnet fan-out was not used. Direct curl crawl plus local parsing read all 110 published person pages in full, the per-person publication and project lists for the three, four project detail pages and five institutional pages — more material, more accurately, for less than the fan-out would have cost. Filed immutably as sources/research/2026-10-29-ioew-person-level-overlap.md with the method, the complete scan table and the stated limits.

The load-bearing finding is a null result. Across all 110 profiles: Datenraum / data space, Gaia-X, Catena-X, Produktpass / DPP, CSRD, ESRS, Scope 3, Blockchain, Kryptograf*, Zero-knowledge, Rückverfolg*, Interoperab*, Datensouveränität, Datenökonomie, Datenteilung and vertraulich* each occur in zero profiles. Non-zero: Berichterstattung 5, Lieferkette 2, Ökobilanz 2, Zertifiz* 1, Commons 1, KI 1, Industrie 4.0 1. “ESRS” scores zero despite a multi-year ESRS project — the regulation is project vocabulary, not identity vocabulary.

The overlap is therefore thematically broad and technically null. The asymmetry, recorded on the page as the one-line reading: IÖW asks whether and why firms report; SINE asks how a reported figure can be believed without disclosure. Complementary, not competing.

New pages. market-competitors/ioew.md (modelled on ecologic-institut.md, same filing location — the Ecornet peer group, not organization/), plus three person pages: people-partners/patrick-schoepflin.md (ranked first — inside the ESRS text, and Nature Credits as a second bridge), people-partners/christian-lautermann.md (the Ranking der Nachhaltigkeitsberichte since 1994; CDR as the nearer bridge), people-partners/jonas-pentzien.md (platform cooperativism, Plattform Industrie 4.0 AG seat since 2022 — the closest institutional contact point found anywhere in the 110).

Second-order finding, not sought: the IÖW is researcher-owned — ~50 individual Gesellschafter*innen plus VÖW e.V. — and states explicitly that it pursues no growth strategy. Ecologic’s founder Andreas Kraemer is on that shareholder list; the Ecornet peer group overlaps at ownership level, not only at network level. Cross-linked from ecologic-institut.md.

Recorded rather than smoothed over: no IÖW publication was read — every publication claim is a list entry. Legal form, turnover, funding mix and offices were not established (/impressum and /forschung 404 under the crawled URL shapes). The team page publishes 110 profiles against a stated 69 staff (end 2024); the gap is unresolved and 110 is treated as “profiles published”, not “employees”. The three research-field descriptions that rendered are fewer than the fields the site navigates to.

Still open, in the order they are worth doing: (1) read the 2026 ESRS report — does it address verification at all; (2) does the Plattform Industrie 4.0 AG touch data spaces; (3) what has 30 years of the Ranking found about where reports are not believable. None of the three has a relationship path recorded.

[2026-10-29] lint | Verification pass over the ten sources ingested since 2026-10-15 (batch 1 of 2)

Why. Martin Pompéry asked for a re-verification of the wiki entries derived from the last fourteen days’ sources — one subagent per source, run as a loop rather than a fan-out, with a checkpoint after ten sources.

Scope. The ten sources with ingest entries since 2026-10-15: Hitachi Integrated Report 2026 · Hitachi Rail “Enhanced by HMAX” · HTR 412 Social Product Declaration (trade press) · Hitachi Sustainability Report 2026 · CE Association, Social Implementation of the CVP · H-AIST CVP research report (04/2026) · Driving Circularity Transitions (06/2025) · H-AIST CE Lab Third Open Forum (07/2026) · Li et al., Agentic Communication Networks · OECD, Data Spaces between Co-ordination Challenges, Financial Viability and Public Policy. Three further post-10-15 files were excluded as out of scope: the Hitachi Rail CSO question set (SINE work product), the Anthropic S-1 and Edenhofer’s Net Zero (neither ingested — grep returns no wiki pages citing them).

Each source was checked by its own subagent against the primary file; roughly 450 claims. Every finding touching attribution was re-verified in the source by the main thread before anything was edited — two subagent findings (that organization/hitachi.md miscites the Scope 1+2 trend and the assurance quote to p. 42) were false positives: that page’s p. 42 is the Integrated Report’s “5-Year ESG Data”, correctly. Those citations were disambiguated by naming the document rather than changed.

The one substantive error. The KPI “number of procurement partners that set GHG reduction plans and targets” (FY2027 target 700, FY2025 result 428) was presented across five pages as the compensation-linked supply-chain KPI. It is not. Both primary documents carry a dedicated table of indicators linked to executive compensation — Integrated Report p. 29, Sustainability Report p. 29 — listing five items, of which the only procurement entry is “number of procurement partners taking action on environment and human rights” (STI; FY2027 target 1.5× / 6,000+ partners, FY2025 result 5,232). The error originates in a merged cell in the Markdown twin’s PLEDGES table, where “STI” sits directly after the 64 % site-GHG figure and reads as belonging to the row below. ⟨This entry corrects the entry of 2026-10-15 at line 3006 of this file, which repeats the error; the log is append-only and that line stands.⟩ The argument it supported survives and is sharpened: a participation count is still not a measurement, and the correct KPI is the broader and softer of the two. Corrected in organization/hitachi.md, papers/hitachi-integrated-report-2026.md, concepts/verifiable-carbon-accounting.md, index.md, and the German dossier 2026-10-15-hitachi-customer-zero-signierte-sensordaten.md.

Everything else was an evidence pointer, not a fact. Across ~450 claims, no error was found in a number, name, date, ISO number, equation or quantity other than the one above. The defects cluster in four patterns:

1. Page pointers. The CE Association white paper has three unnumbered front-matter pages; the first half of its page cited printed pages one too high and the appendix cited the PDF’s physical index (printed p. 12 as p. 16, printed p. 15 as p. 19). Re-anchored on the paper page and in five dependent Sources footers, with the rule recorded. In the Sustainability Report page, the R3 Promotion Project Meeting quote moved p. 32 → p. 33. In market-competitors/idsa.md, the “>140 member organisations” endnote moved Ch. 4 → Ch. 1.
2. Inherited citations. When a second document about the same entity is ingested, citations migrate from the first without being re-anchored. Both Hitachi-cluster errors and both “the white paper” misattributions on organization/hitachi-aist-cecrl.md — a study-group roster and the whole ISO/WTO standards route, credited to the June 2025 overview, which contains neither — have this shape. The second of those produced an internal contradiction with papers/h-aist-cvp-overview-2025.md, which correctly recorded the absence.
3. Quotation discipline. Four cases of quotation marks around paraphrase or a silently dropped clause: the p. 52 Scope 3 target (SR), “maintaining equity among the stakeholders that comprise the VN” (CVP page), “preserve [its] own control” for “preserve each domain’s control” (ACN page), and the four-condition passage on concepts/data-space-boundary-conditions.md, set as a blockquote but written as a paraphrase — now replaced with verbatim §3.2.
4. Weight of evidence. concepts/bootstrapping-problem.md called Fintraffic’s 24 % CO₂ / 20 % energy / EUR 45 m-per-year the report’s “one hard outcome number”. OECD endnote 22 disclaims nearly every property that would make it hard: one shipping company’s own voyage data on one route, the EUR 45 m a counterfactual projection rather than measured savings, “not independently verified”, no peer-reviewed publication identified. The caveat is now on the page. This was the most externally dangerous finding of the pass — it is precisely the figure one would quote outside.

Two further fixes: one fabricated negative-search claim on papers/h-aist-cvp-overview-2025.md (the word “information” was said to appear once in Driving Circularity Transitions; grep returns zero occurrences), and the “Enhanced by HMAX” release, whose publication date (24 September) came from the source filename while the document’s own dateline reads Berlin, 22 September.

An unreconciled contradiction was also represented rather than left standing: market-competitors/gaia-x.md dated Agdatahub’s liquidation filing to December 2024 in one section and November 2024 in another, from two different sources, with no flag. Both now stand, flagged, under a dedicated heading.

Clean, no findings: H-AIST CE Lab Third Open Forum (~55 claims, both known failure modes checked and absent). Near-clean: HMAX release, HTR 412, the H-AIST research report, Agentic Communication Networks. Worst: Sustainability Report 2026, four confirmed errors.

Left open for Martin. The marker ⟨Note: …⟩ is used on about twenty pages but is not in the schema, which defines only ⟨Inference⟩ and ⟨Background⟩. That is too established to be drift and too widespread to rewrite unilaterally — it is a schema question, not a lint finding.

Batch 2 not yet run. Per the standing instruction, the pass pauses here for a go.

[2026-10-29] ingest | Pathfinder Network Vision Paper (WBCSD/PACT, June 2022)

Source 1 of 3 supplied by Martin Pompéry for the PACT topic complex (https://wbcsd.github.io/tr/websitedocs/PACT_network_vision_paper.pdf). Saved as sources/standards/2026-10-29-pact-network-vision-paper.pdf, converted with marker_single --disable_ocr; ingested from the Markdown twin. Takeaways discussed with Martin before writing; go given 2026-10-29.

Conversion exceptions, two. (1) Marker turned the Figure 5 table into a word-salad paragraph; the PDF’s text layer holds only the header cells (“Approach Benefit”), so the table is an image. Recovered by rendering page 9 with pymupdf at 130 dpi and reading it natively; the six approach/benefit pairs are transcribed in full on the paper page, and the corrupt paragraph in the Markdown twin is flagged there as not citable. (2) Figures 1 and 4 survive only as extracted JPEGs; both were read natively and transcribed as tables. Figure 2 (exemplary value-chain data flow) was not transcribed — illustrative. The page-2 collaborator roster is logos only and was read from a rendered page image.

What the source is. The founding vision paper, June 2022, written while the Network was still “in development”. Earliest primary PACT source in the wiki. Its value is as a baseline: stated intent, against which the 2023 Tech Specs, Methodology v3 and the ICC Review read as outcome.

Main finding. Figure 4 marks exactly one of six mechanisms as “already specified by PACT” — the data exchange mechanism. Identity management, authentication, plausibility/error-checking and security/fraud prevention were intent; auditing sat under “Add-on features”. The ICC’s four 2026 charges land on those white boxes, and verification is still a two-horizon roadmap in Methodology v3. Recorded as a new section on the PACT page rather than as a resolution of the ICC contradiction, which stays open.

Second finding. The paper rules out, by design, the layer the ICC proposes: the Network “cannot be used to calculate a carbon footprint or to store any individual footprint data”, is peer-to-peer “with no need for a central intermediary”, and is explicitly not setting technological standards. Fed into open task 1 as a fifth required element of SINE’s response — defend the boundary or move it.

Third finding. Confidentiality stated at full strength, answered with governance, legal frameworks, identity/fraud prevention and data minimisation; no cryptographic term anywhere. Logged as a fourth instance in open task 19, with the note that SINE was the named technology partner and had shipped a secure-MPC tool fifteen months earlier — which weakens reading 1 and makes the 2022 paper a clean test case for the sequencing reading. Contradiction with the Data Commons whitepaper’s “strong encryption” description held open, not resolved.

Smaller. SINE’s formal title inside PACT (“technology partner”, McKinsey Sustainability “knowledge partner”) now has a primary source. 38 core collaborators recorded, two of which change existing pages: Catena-X was a PACT collaborator in 2022, not only a parallel effort; Japan’s Green x Digital Consortium is the earliest PACT–Japan link in the wiki and sits unconnected to the Hitachi/CVP strand. First attestation of “internet for emissions data”. A third pillar, Pathfinder Ecosystem, has no page — the source defines it only by activity, so it is noted rather than created.

Pages. New: papers/pact-network-vision-paper-2022.md. Updated: concepts/pact-pathfinder-network.md (new section + two ⟨⟩ notes + sources), organization/sine-foundation.md, people-partners/wbcsd.md, market-competitors/catena-x.md, open-tasks.md (items 1 and 19), index.md.

Commit still overdue — now across three ingests.

[2026-11-01] ingest | Progress of “Inspire 2027” — CEO deck to financial analysts, 27 April 2026

Supplied by Tell Münzing (Slack, 2026-11-01) together with the Birje DSS deck, with the brief: “understand Hitachi digital business development … What are the implications of the DSS / CI split? Why would the CEO of digital business not talk about driving Lumada 3.0 forward assuming this is where we fit in strategically?” Ingested one source at a time; this is source 1 of 2.

Saved as sources/hitachi/2026-04-27-hitachi-inspire-2027-progress-financial-analysts.pdf, converted with marker_single … --output_format markdown --disable_ocr. Method note: the deck’s substance is in its figures — the Markdown twin carries only loose text — so every cited figure was read natively from the extracted JPEGs, and the footnotes, which Marker dropped entirely, were recovered with pdftext (plain_text_output, page by page) from Marker’s own venv. The footnotes turned out to be the two most load-bearing findings. Citations anchored to printed page numbers (= Marker _page_N_ + 1).

New page: hitachi-inspire-2027-progress-2026.md.

Updated: hitachi.md — new sections “The four sectors, side by side — FY2025” and “HMAX, sized”; status ⟨2026-11-01⟩. hitachi-rail.md — new section “Mobility in the CEO’s own numbers”. customer-zero.md — new section “How big it is: 175 sites”. Plus index.md (new paper entry; three entries extended) and open-tasks.md (new item 20).

What the source changed.

1. The four sectors are parallel, not nested — answering Tell’s premise directly. Sector revenues sum to ≈¥10.7tn against group ¥10,586.7bn (difference = eliminations); nesting Energy and Mobility inside CI would overshoot by ≈¥4.5tn. And Lumada is an attribute of sector revenue, not a fifth sector: each yen is booked once in one sector and additionally flagged, which is how digital growth on the physical installed base avoids double counting, and why group 40% is the weighted blend of Energy 25 / Mobility 33 / CI 43 / DSS 62.
2. HMAX is ¥300bn at over 20% adj. EBITA margin — ~7% of Lumada’s ¥4.1tn and under 3% of group revenue. The most profitable sliver, not the engine. This constrains the dossier’s part-2 argument: the Lumada 80-20 case survives, but it must be made against Lumada the aggregate, not routed through HMAX.
3. Customer Zero has a size: 175 in-house manufacturing sites, attached to HMAX Industry — confirming Connective Industries as its operational home and Mobility’s absence.
4. Two basis changes sit under the 29% → 40% jump. Group Lumada figures newly reflect “revenue consolidation offsets … started from this period”; CI and DSS alone are restated “based on the new reporting segment structure” with structural-reform operations (home appliances, ATM) out of the denominator. The deck never says what moved between CI and DSS → open task 20.
5. DSS owns none of the three HMAX products. On the deck’s own product slide they belong to Energy, Mobility and Industry; DSS appears as a financial sector and as the “Digital” domain band.
6. No sustainability vocabulary in 24 slides — no circular economy, carbon, PCF, DPP or supplier data. The growth story is Physical AI + Agentic AI, “over ¥100tn by 2030”. Not a contradiction of the Sustainability Report; a division of labour by audience, and a constraint on how a proposal has to arrive to reach this altitude.

Difference held open, not resolved. This deck (April 2026) names three HMAX products — Energy / Mobility / Industry — with Digital as a fourth domain band under Lumada 3.0. The Integrated Report (October 2026) names the HMAX family as Energy / Mobility / Industry / Cyber. Recorded on both pages as a difference between two documents five months apart.

Next: source 2, the Birje DSS deck of 28 October 2026 — which is where Tell’s DSS/CI question and the Lumada 3.0 question actually land. A DSS entity page and a CI entity page are deliberately deferred until it is read.

Commit due — four wiki pages changed, one added, plus two source PDFs and their Marker output directories.

[2026-11-01] ingest | Pathfinder Framework Version 2.0 (WBCSD/PACT, January 2023)

Source 2 of 3 from Martin Pompéry’s PACT batch (message 2026-10-29): “die PACT Methodologie V2, an der wir mitgearbeitet haben”. Downloaded from wbcsd.github.io/tr/2023/framework-20232601/framework.pdf to sources/standards/2026-10-29-pact-methodology-v2-framework-2023.pdf (20 MB, 78 pp.), converted with marker_single; ingested from the Markdown twin, read in full in the main thread.

Conversion recovery. Marker garbled three image-tables into word salad — Figure 17 (assurance roadmap, p. 46), Figure 19 (key values, p. 56), Table 2 (summary of guidelines, pp. 13–15) — and silently dropped “Version 2.0” from the title page. Pages were re-rendered via uv run --with pymupdf python -c "…get_pixmap(dpi=135)…" and read natively; Figures 17 and 19 are transcribed in full on the paper page and the garbled paragraphs are flagged non-citable. Same procedure as the Vision Paper’s Figure 5. Also recovered: the page-2 collaborator spread (50 logos + McKinsey as knowledge partner, SINE as technology partner).

New page: papers/pact-methodology-v2-framework-2023.md.

Updated:

• concepts/pact-data-quality-verification.md — new section “What changed from Pathfinder Framework v2”: the assurance roadmap shrank from three horizons to two and reasonable assurance (2030) disappeared; DQR deadline 2025 → 31.12.2027 (v3 attributes the slip to its own revision of the matrix); DQR indicators 5 → 3, scale 1–3 → 1–5; v3’s internal inconsistency on its short-term horizon (2025-2030 in prose vs 2023–2030 in the table) recorded as a held-open contradiction.
• concepts/pact-pcf-calculation-methodology.md — exemption rule v2 (<1%/process AND <5% aggregate) → v3 (<3% aggregate), i.e. tightened; v2’s six areas of extension over v1; endnote 38’s list of v2-new elements.
• concepts/pact-pathfinder-network.md — new section “The moving value set (2022 → 2023)” with the three-way comparison (5 design principles → 6 key values → 3 Data Commons values); Figure 19 definitions; and the fifth instance of the confidentiality finding appended to “Confidentiality, stated without cryptography”.
• organization/sine-foundation.md — role title stable across the document (Foreword “tech collaborator”, page 2 and back matter “Technology partner”), the Disclaimer’s McKinsey-only credit, and SINE’s own January 2023 self-description (“ready-to-use governance tools and open-source software”).
• open-tasks.md item 19 — fifth instance. Zero cryptographic vocabulary in 78 pages; the value defined down to adjustable access rights; the one technical control optional, about authenticity, and dropped by v3. SINE’s self-description supports the sequencing reading; the fact that the governance prerequisite was met by 2025 and the instrument still did not appear pushes the PACT case toward reading 3 (not wanted). Tech Specs v2.2 named as the direct test.
• index.md.

Discussed with Martin Pompéry before writing (seven findings); go given 2026-11-01.

⚠ Commit still due — now across four ingests. Repo has uncommitted work.

[2026-11-01] ingest | Hitachi Digital — Anand Birje presentation, 28 October 2026

Source 2 of 2 from Tell Münzing’s 2026-11-01 brief, ingested after source 1 was completely finished. Saved as sources/hitachi/2026-10-28-hitachi-digital-anand-birje-dss.pdf, converted with marker_single … --output_format markdown --disable_ocr; 12 slides, footer “Hitachi Digital | Q2”. Slides 4, 5 and 10 were read as rendered pages (pypdfium2 from Marker’s venv) because their meaning is in layout.

⚠ Conversion defect worth remembering: on slide 5 the Markdown twin detaches both headline figures from their captions and places them adjacently, which reads as though “+46% YoY” were the top-100 account growth rate. It is the opposite — 46% is the internal synergy channel, and the top-100 accounts are the single-digit CAGR. Noted on the paper page.

New pages: hitachi-digital-birje-2026.md, hitachi-digital.md.

Updated: hitachi.md — new section “The DSS / CI division of labour”. hitachi-rail.md — new section “Rail as a client of the digital business”. customer-zero.md — new section “Practised without the name”. Plus index.md (two new entries, three extended).

What the source changed.

1. The DSS/CI relationship is settled, and it is not a mirror. The three HMAX products belong to Energy, Mobility and Industry; DSS owns the horizontal underneath — Physical AI FDE service, HMAX AI Operations, HMAX Data Fabric (data platform + IWIM/AI, “ontologies and physical models”), plus HARC as the industry-ontology layer. CI owns the sites and the Customer Zero population; DSS owns the mechanism. ⟨Inference: a proposal that needs a site goes to CI; a proposal that is a mechanism goes to DSS. The wiki had been treating these as one conversation.⟩
2. The growth of the digital business is internal. Sell-to/sell-with Hitachi front business units: +46% YoY (FY2026/H1 actual). External top-100 accounts, 65% of the business: single-digit 2-year CAGR. Three of four named reference cases are Hitachi companies. ⟨Inference: an internal-first proposition runs with this BU’s incentives, which strengthens the Customer Zero framing on economic rather than doctrinal grounds.⟩
3. Why Lumada 3.0 is never mentioned — answering Tell’s second question. Four reasons, none of them de-prioritisation: Lumada is a reporting attribute with no product surface; Birje runs a BU one level below the P&L that reports a Lumada ratio; the deck’s own benchmark is “best-in-class multiples” against IT-services peers, for whom the word means nothing; and what he owns is the machinery, not the narrative. The consequence is a vocabulary one — pitch the mechanism, not the programme.
4. Customer Zero is practised without the name, by the sector that the Integrated Report says runs it.
5. The counterpart function does not exist yet: an “HMAX-focused CTO office”, to be stood up in FY26. That is the most specific address any source gives, and it is unstaffed.
6. Rail’s live digital engagement is in the factory — the FDE model for Smart Operations, Factory IQ, vehicle manufacturing optimisation — not on the fleet.
7. No confidentiality, provenance, verification, security or standards vocabulary anywhere in the deck. The only trust word is “trusted AI”, about model behaviour. Recorded as the falsifier against the structural fit, and as a fourth instance of the pattern in open task 19.

Deliberately not written: a Connective Industries entity page. CI is now described from two sources but always as a counterparty to something else; there is no source in the repo written from CI’s own side.

Commit due — both sources of the 2026-11-01 brief are now ingested: 2 source PDFs plus Marker output, 3 new wiki pages, 5 updated.

[2026-11-01] ingest | DSS Business Strategy — Jun Abe, Hitachi Investor Day, 10 June 2026

Source supplied by Tell Münzing as 20260610_04_digital_en.pdf, saved as sources/hitachi/2026-06-10-hitachi-dss-sector-briefing-abe.pdf (25 pages). Marker twin written to the sibling directory.

⚠ Conversion caution, recorded on the paper page. Marker flattened this deck almost throughout — figure captions on pp. 2, 4, 6, 7, 9, 13, 14, 16 and 18 collapse into structureless adjacency. Nine slides were rendered natively with pypdfium2 at scale 2.0–2.2 and read as images; every number and quotation on the paper page comes from the rendered slide, not the twin. Page indexing for this deck: printed slide number == 0-indexed pypdfium2 page index (verified at indices 2, 5, 19) — the opposite of the convention that held for the April and October decks.

New page: papers/hitachi-dss-abe-2026.md.

New source (SINE-authored): sources/hitachi/2026-11-01-questions-hitachi-digital-addendum.md — four questions (section VIII) extending the consolidated set, which could not ask them before the two DSS decks were held.

Updated: organization/hitachi-digital.md (Status; Where it sits; the ¥30bn money bullet; the Abe-pillars-vs-Birje-layers comparison; the “DSS Sector” band on the Data Fabric; the GlobalLogic full-year table; four new open questions — plus a new section, “Where SINE attaches, and to whom”, filed at Tell Münzing’s request: the Data Fabric-vs-AI-Operations layer answer, the CE / value-chain nuance for Hitachi Rail’s CSO, the narrative for a counterpart function that does not exist yet, the six-audience internal map, and six translations for the DSS CEO/CTO). organization/hitachi.md (fifth primary source; the DSS/sector division of labour is no longer an inference — Abe p. 13 tags each contributed technology “(DSS)”). organization/hitachi-rail.md (Edge AI (DSS) → HMAX Mobility; the Energy ↔ Mobility railway-power synergy arrow). concepts/customer-zero.md (175 sites corroborated at sector level). open-tasks.md item 20 (does not close: the Lumada FY27 target was revised 65% → 75–80% without a stated basis, which is the task restated at target level; the ATM/OKI integration now has a date, October 2026, but still no revenue effect). index.md.

Retired inference: the DSS / physical-sector division of labour, previously derived by reading the April and October decks against each other. Settled: Data Fabric ownership (p. 16 band reads “DSS Sector”). Found: the funded line — “enhance data fabric capabilities”, first technology bullet of the ¥30bn FY2026 organic investment. Found and bounded: the CE hook — “optimal production linked to green power supply and demand” is a value-chain sustainability case in Hitachi’s own hand, but all three Data Fabric arrows are Hitachi-to-Hitachi.

Commit due.

[2026-11-01] dossier | Where SINE attaches inside Hitachi’s digital business

Written for Tell Münzing to share with the team, covering the three points answered in the channel earlier today (layer, audiences, legibility) plus the circular-economy / value-chain nuance for Hitachi Rail’s CSO. dossiers/2026-11-01-hitachi-where-sine-attaches.md, rendered to PDF with --external --no-toc (pandoc 3.10.1 from /workspace/agent/tools/).

It is a briefing, not a source: everything in it is already filed on wiki/organization/hitachi-digital.md under “Where SINE attaches, and to whom”, which cites the Abe and Birje decks directly. Indexed under ”## Dossiers”.

Length note: two full pages plus a four-line spill onto a third, after several compression passes. Dropping the falsification paragraph would make it exactly two — not done without asking, since the dossier schema requires it.

Commit due.

[2026-11-02] lint | Eleventh health check — a page was linked as a concept before it existed, a rename was dated from nowhere, and one source of a three-source bundle is still sitting there

Why this one covers two weeks. The 2026-10-26 lint never ran: it failed on an API error (OneCLI gateway, 502) before doing any work. This check therefore spans 2026-10-19 → 2026-11-02, the busiest fortnight in the wiki’s history — six ingests, two queries, one verification pass, one dossier, and 17 pages touched.

Method. Structural checks on the main thread with a rewritten link/anchor walker (/tmp/lint/lint.py); three sonnet subagents for the content layer, scoped to the clusters that moved — Hitachi, PACT, and a wiki-wide date sweep. Every subagent finding was re-verified against the files before anything was edited.

⚠ Checker method note, for the next lint. My first run reported 19 broken anchors; 18 were false. Two bugs, both already recorded by earlier lints and both re-introduced because the script is rebuilt from scratch each time (/tmp does not survive the container): (1) the slugger must map each whitespace character to a hyphen, not collapse runs — ## EcoSpace — the precedent slugs to ecospace--the-precedent, with the double hyphen the em dash leaves behind; (2) relative links beginning ../ that resolve back inside wiki/ are wiki links, not external files — treating them as external also inflated the terminal-page count from 3 to 21. A third, found at the end of this run: (3) strip fenced blocks and inline code spans before matching links — this entry quotes [IWIM](../concepts/…) as prose about a link, and the checker read it as a link and reported the log as broken. The checker belongs in the repo, not in /tmp. Recorded as a recommendation, not done unilaterally — it would be the first script in a prose wiki.

Structural result (after fixes): 137 pages (134 content + index, log, open-tasks). 0 broken links, 0 broken anchors, 0 orphans, 0 index gaps, 0 weakly-linked pages. Three terminal pages — ada-lovelace-institute, gs1, mozilla-foundation — unchanged for eight checks. Clean for the eleventh consecutive run, with one exception, below.

The one real break: a concept page linked before it was written

organization/hitachi-digital.md, created 2026-11-01, links [IWIM](../concepts/iwim.md) — a page that did not exist. IWIM was discussed on four pages (organization/hitachi.md, papers/hitachi-integrated-report-2026.md, papers/hitachi-digital-birje-2026.md, organization/hitachi-digital.md) without a canonical page of its own, so this is the “concept without a page” check and the broken-link check reporting the same defect.

Fixed by writing the page, from the primary source rather than from the dependent pages: concepts/iwim.md, built on the Integrated Report’s R&D Strategy section (printed pp. 21–22) plus Birje’s slide 4. The reading it records: IWIM is a safety argument before it is a product — its stated reason to exist is preventing “large-scale power outages and railway accidents resulting from incorrect AI decisions”, and the asset it claims is OT knowledge and physical causality, not data volume. Also noted: the SPP/NVIDIA −80 % study time is an anticipated reduction stated by the partner, the same evidential shape as the avoided-emissions headline already flagged on organization/hitachi.md; and the “IWIM framework” and “Social challenges” panels survive only as extracted images and are not transcribed. Linked from hitachi.md and the Birje paper page, catalogued in index.md.

Findings fixed directly (17 pages, plus index and log)

1. An unsourced date on a one-day-old page. organization/hitachi-digital.md opened with “renamed Hitachi Digital in the second half of FY2026”. No source the wiki holds says that — and H2 FY2026 (Oct 2026–Mar 2027) sits after the September 2026 appointment to the merged organization that the same cluster’s paper page records from the deck. Replaced with what the sources carry: the combined organization is called Hitachi Digital; Birje was appointed to the BU in August 2026 and to the combined organization in September 2026 (Q2 FY2026); the deck is footered “Hitachi Digital | Q2”. The renaming itself is undated here. The tagline “Engineer what the world depends on” is sourced (deck, p. 5 area of the twin) — but “adopted with the rename” was not, and is gone.
2. The Integrated Report page was never revisited in the 2026-11-01 batch. Its DSS paragraph (p. 20) carries the ¥30bn / ¥1tn figures and a four-name partner ecosystem, all of which the Abe and Birje ingests covered in far more depth the same day — while organization/hitachi.md was updated to link them and this page was not. Added the cross-references plus a note that the Integrated Report’s “Anthropic, Google Cloud, NVIDIA, OpenAI and others” is a selection: Abe’s deck gives six, alphabetical, adding AWS and Microsoft. Not recorded as a contradiction — “and others” plausibly covers them, and Anthropic/Claude is company vs. product.
3. An unlinked evidence cell. In the Hitachi Digital audience map, the Group Standardization Committee row cited “wiki, CECRL pages” as prose while every other row in the table linked its evidence. Now points to hitachi.md and hitachi-aist-cecrl.md.
4. Three PACT pages name the new v2.0 Framework page without linking it. papers/wbcsd-pact-methodology-v3-2025.md says “Supersedes PACT Methodology v2 (2023, then called ‘Pathfinder Framework’)” in plain text — the page it supersedes went live on 2026-11-01; papers/pact-network-vision-paper-2022.md calls v3 “the Framework’s descendant, three generations on” with no entry for the middle generation; people-partners/wbcsd.md says v3 “improves on v2” unlinked. All three linked, both inline and in Cross-references. The new page’s own outbound links were checked and are complete.
5. PACT’s founding year, disambiguated. concepts/pact-pathfinder-network.md juxtaposed “launched in 2020” and “launched at COP26” without naming the year of COP26, which reads as a contradiction on its own page; organization/sine-foundation.md resolves it (“launched in 2020 and went public at COP26 in 2021”). Added “(2021)”.
6. Two pages still read a missed gate as a future one. CCTI Milestone 1 fell due 30.09.2026. projects/ccti.md and open-tasks.md item 8 have carried it as overdue since 2026-10-12, but the two cross-references to it kept the future tense: organization/aria.md (“due M1, 30/9/2026”) and concepts/epistemic-integrity-certificate.md (“M1 (30/9/2026) is the gate”). Caveat and link propagated to both. This is the second time the lint has found this exact shape — a correction applied to the canonical page and not to the pages that point at it.
7. An expiry date that arrived. concepts/ai-stack-market-structure.md and concepts/adtech-as-regulatory-precedent.md both rest a “forming exception” on CME Group / Silicon Data compute futures announced for 2026-10-05, “pending regulatory review”. That date is four weeks past and the wiki holds no source on what happened. The adtech page’s own ⟨Inference⟩ called it “a momentary reading with a visible expiry date” — correctly, and then the expiry passed unnoticed. Both now say so.
8. Milestone arithmetic. Item 8 and projects/ccti.md updated: 33 days past due, eight asks in nine weeks, with the 2026-10-26 gap explained.

Reported, not fixed — these need a decision or a source

1. The PACT bundle is one source short of finished. Martin supplied three on 2026-10-29. Two are ingested (Vision Paper 2022, Framework v2.0 2023); sources/standards/2026-10-29-pact-data-exchange-protocol-v2.2.md (2,421 lines, Markdown already) is not — no page, no index entry, no log entry, no citation anywhere. The Vision Paper page itself anticipates it in writing: “the Methodology v2 and Tech Specs v2.2 ingests that follow this one.” It is the document that would settle several open questions, since it is where the data model and the REST API actually live. (Its comments address is public-dev@pathfinder.sine.dev — SINE’s own domain.)
2. A relevant standard has sat unread for three weeks, and a page on its exact subject was written without it. sources/standards/2026-05-22-gbz-185.1-2026-agent-interconnection-general-architecture/ — GB/Z 185.1—2026, Artificial intelligence — Agent interconnection — Part 1: General architecture, a Chinese national standard, manually transcribed from page images on 2026-10-12 with an exemplary acquisition note. It appears nowhere: not in index.md, not in open-tasks.md item 11, not in log.md. Two weeks after it was filed, the 2026-10-26 ACN ingest created concepts/agent-protocol-interoperability.md — the O(n²) problem, protocol convergence — from a different source entirely. A national standard for agent interconnection is not a footnote to that page; it is a second answer to it. Extended during this lint rather than left as a finding — registering a source that exists is a factual correction, not a judgement call. Item 11 now reads “Nine un-ingested sources, filed 22 September – 29 October” and carries both the Tech Specs and the GB/Z, with a note on why neither fits the item’s own retirement logic: both have a wiki page already waiting for them, and the Tech Specs have a requester, which breaks the item’s standing ⟨Inference⟩ that what keeps this backlog unread is the absence of one. concepts/agent-protocol-interoperability.md now carries a warning that it has heard only one side. index.md’s open-task count is corrected nineteen → twenty (item 20 arrived with the 2026-11-01 ingest and the index was not updated). What is not fixed is the ingest itself — both still need reading, one at a time.
3. Verification batch 2 was never run. The 2026-10-29 pass covered ten sources and paused for a go per the standing instruction. Six sources have been ingested since, five of them on 2026-11-01 — the Hitachi Digital cluster and the two PACT papers — and none has been through a verification pass. The one substantive error batch 1 found (the compensation-linked procurement KPI) had propagated across five pages and a client-facing dossier before it was caught.
4. The ⟨Note: …⟩ marker is still not in the schema. Carried from the 2026-10-29 pass, unanswered: it is used on roughly twenty pages, the schema defines only ⟨Inference⟩ and ⟨Background⟩. A schema question for Martin, not a lint fix.
5. The commit is due and is now large. Nothing has been committed since 1d0f876 on 2026-10-29. Pending: five new pages from the 2026-11-01 ingests, this lint’s concepts/iwim.md, ~17 modified pages, one new dossier with its PDF, and four new source bundles.
6. CCTI Milestone 1, eighth ask. Unchanged, and the oldest unanswered question in the wiki.
7. One external check would change a page. Whether the CME/Silicon Data compute contract launched on 5 October. A launched contract is a price-formation mechanism at compute; an abandoned one is not, and the “forming exception” on two pages turns on it.

Clean, no findings: PACT version drift — every v2-era rule (assurance horizons, DQR deadline and indicator count, exemption threshold) is confined to the three dedicated pages and tagged by version on all three; no page states a superseded v2 rule as current. Hitachi figures — the DSS revenue/EBITA/ROIC/Lumada-ratio/backlog series, HMAX at ¥300bn / >20 % margin, the GlobalLogic table, 175 Customer Zero sites, Mobility’s 2,500 trainsets — consistent across all twelve cluster pages. Named roles and org structure, consistent throughout. PACT member counts (2,500+ companies, 150+ stakeholders, 11+ initiatives) and the collaborator-circle growth (38 → 50), consistent. Six contradictions were found already flagged as held-open and correctly left alone — Mobility’s 8.2 %/9.2 % two bases, the HMAX product-family vs. Digital-domain difference, the Security pillar routing fact, the CI/DSS segment restatement (open task 20), the 2,500-trainset reading, and PACT’s encryption contradiction.

[2026-11-05] ingest | Unlocking the circular economy (EEA Report 02/2026)

Source: sources/circular-finance/2026-11-05-eea-unlocking-the-circular-economy.pdf plus its Markdown twin in the sibling directory of the same name. Downloaded from eea.europa.eu, converted with marker_single --disable_ocr. 32 pages, ISBN 978-92-9480-767-0, doi:10.2800/5422329, CC BY 4.0.

Found during the Ellen MacArthur Foundation / financial-sector search run for Tell Münzing earlier the same day; ingested on his instruction (“Ingest”) as the first of two sources, the second being Toxopeus et al. (2021).

Why it was worth an ingest. The wiki’s two circular-finance concept pages both rested entirely on interested parties — Circle Economy, which sells circularity assessment, and DKB, a bank describing its own product gap. The EEA states the same findings as a European agency with a statutory reporting mandate and nothing to sell. Three claims changed status from practitioner opinion to official finding:

1. The definition problem, instrument by instrument (§3.1.1): EU Taxonomy criteria “only partially cover circular activities”; financial ecolabels “not currently designed in a way that allows investments linked to circular economy strategies to be mapped”; SFDR “do not fully allow the benefits of circular businesses to be quantified. Consequently, the benefits cannot be factored into financing decisions”; CSRD “still immature”. Plus a dimension the wiki did not have: the problem is also jurisdictional — 27 divergent national definitions of ‘recyclability’, ‘repairability’, ‘end-of-waste’ (§3.1.3).
2. The valuation gap (§3.1.2): “traditional credit assessments remain anchored in solvency metrics and physical collateral”; finance teams “rely on tools unsuited to integrating circular business models into credit assessment”.
3. The bootstrapping problem for physical infrastructure (§3.2.2): circular infrastructure “displays characteristics of public goods … require collective investment but generate dispersed benefits”; pioneer knowledge “spills over to competitors, reducing first-mover incentives”. This is Frischmann’s test failing, stated without being named.

The finding by absence. 32 pages on why circular value cannot enter financing decisions contain no digital product passport, no materials or building resource passport, no ISO 59000 and no data space. The enablers chapter answers an information problem with taxes, EPR fee modulation, blended finance and broader taxonomy coverage. The single line where SINE’s instrument is implied and not named: “results-based loans tied to verified waste reduction” (ch. 4). The only place data appears as a viability condition is Box 3.1, the ReCiPSS case study.

Figures recorded: +68 % / +EUR 82 bn per year; EUR 1,229 bn gap 2025–2040; 51 % of the gap in CEAP value chains; current investment EUR 120–121 bn at ≈93 % private; private investment flat at ≈0.8 % of GDP since 2019; circular material use rate +1.5 pp 2010–2024; cost of inaction EUR 84.5 bn/year; EIB EUR 5.1 bn over 153 projects 2020–2024. The EUR 1,229 bn figure is the same one the Wechselgeld report carries; both derive from EIB-EC (2026), which is now the attribution to use.

Pages written/updated. New: papers/eea-unlocking-circular-economy-2026.md. Updated: concepts/circularity-definition-problem.md (new section “Independent confirmation”), concepts/circularity-valuation-gap.md (new section “The agency-side confirmation”, plus the scale figures), concepts/bootstrapping-problem.md (new section on circular infrastructure as the same provision problem, read against the Agdatahub and ABtC patterns), concepts/ecosystem-financing.md (new section with the EU’s own lifecycle instrument inventory and why it does not answer the open question), concepts/esg-reporting-financial-sector.md (SFDR and CSRD measured against circularity). Index and this log.

Method note. Tables 2.1, 3.1 and 3.2 came out of the conversion with collapsed columns and are unusable; everything cited is from running prose. If a table figure is ever needed, read it from the PDF.

Two open questions tightened, neither closed. “Is there a documented case of a bank changing a credit decision because of asset-level circularity data?” and “Is there a working example of an ecosystem-level financing vehicle in the circular economy?” — the EEA’s EU-wide inventory contains neither, which is stronger evidence for “no” without being proof of it.

Two defects found in the existing wiki while doing this, both now fixed or filed:

• The Wechsel-Geld ingest of 2026-11-04 was never catalogued in index.md and never logged here. An index entry has now been written (dated 2026-11-04 with a note); this log entry is the first record of the omission. Nothing was lost — the page itself was complete — but the catalog was wrong for a day.
• That page links to three entity pages that do not exist (organization/roechling-stiftung.md, organization/circle-economy.md, people-partners/silke-stremlau.md), and POLYPROBLEM reports 06 and 07 have no wiki page at all although both have been read and analysed. Filed as open task 21; needs a decision from Tell Münzing.

The three-report dossier dossiers/2026-11-04-polyproblem-06-07-08-drei-reports.md has also been catalogued under ”## Dossiers” per the schema.

Commit due — the repository has uncommitted changes across wiki/, dossiers/ and sources/circular-finance/.

[2026-11-06] ingest | Der Circularity Code (POLYPROBLEM Report 06, Röchling Stiftung / Wider Sense, Dez 2023)

Ingested sources/roechling/2023-12-01-polyproblem-06-circularity-code.md (2,557 lines / ~19,200 words, 60 pp.) on Tell Münzing’s instruction “06 07 ingest” (Slack, 08:53). Read in full in the main thread, nothing delegated. Takeaways were discussed with Tell before writing; he corrected the first pass for under-weighting the financing material (“gibt es in 06 nichts relevantes als Link zu Finanzierung/bankability?”), and the page set below was written only after his go.

Source caveat carried onto the page. The file is a full machine text extraction of a two-column magazine layout, explicitly not an edited full text. Only running prose is quoted; no figure, table or footnote is cited anywhere.

New pages. papers/polyproblem-circularity-code-2023.md (the summary, 497 lines); organization/roechling-stiftung.md (closes three broken links from the Wechsel-Geld page — the foundation’s own self-description, “nicht als reiner Förderpartner, sondern … Aufbau neuer sektorenübergreifender Kooperationen und Netzwerke”, is the explanation for why its reports read as convening instruments).

Nine concept pages updated. confidentiality-gap-in-carbon-accounting.md — the gap is now attested in plastics (Claussen-Markefka/R-Cycle on Betriebsgeheimnisse) and the wiki has its first named mechanism in production: föderiertes Lernen at SKZ with the Hamburg supplier Katulu. Recorded with the inference that it protects the inputs of a model, not the provenance of a claim, so it does not close the gap. bootstrapping-problem.md — the ABtC inference now has source backing: the 2023 Mehrweg-Angebotspflicht “hat … nichts geändert”, a duty that moved no participant’s expected return; offtake guarantees and an earmarked single-use levy added as instruments that do. ecosystem-financing.md — two instruments no finance source here lists (Helftewes’ langfristige Abnahmegarantien, p. 13; a noticeable single-use levy, p. 25), plus a new subsection cost-bearer ≠ value-captor enumerating five cases, offered as a sharper diagnosis than either 06’s “Haltung” or 08’s “Lenkung”. circularity-valuation-gap.md — Helftewes’ operating-side instance (“solange Rezyklate teurer sind als Neuware”) distinguished from Gruber/Nusseck’s unbookable value: a visible negative spread versus an invisible positive one. data-space-business-models.md — three field instances, including attestation as a fourth revenue model (KOLEKT, CleanHub), payer = whoever bears a reporting liability. digital-product-passport.md — Moesslein’s (Polysecure) attachment objection, the sharpest in-source challenge to SINE’s own passport-first position: over half the mass is shredded fragments, and a passport is a record without a reliable binding to the physical object at the moment of sorting. Left standing, not smoothed over. residual-value-grading.md — cirplus initiated and financed DIN SPEC 91446 privately: the strongest evidence in the wiki that the beneficiary will fund verification, and the European sibling of H-AIST’s grading. data-commons.md — the report as a negative case: it demands a shared architecture over “Insellösungen” and never asks who owns or governs it. coordination-vs-bargaining-failure.md — a mixed case, with the inference that classifying everything as coordination and therefore recommending convening is the characteristic error of a convening organisation.

Two further pages picked up cross-references: organization/giz.md (the “more than twenty collection apps in Indonesia alone” measurement, plus the Waste Flow Diagram) and papers/polyproblem-wechsel-geld-2026.md (06 and 08 now point at each other, which neither report does).

Contradictions represented, not resolved. Urff (ReFrastructure, “Diese Daten sind ein gesellschaftliches Gemeingut”) against Barthel (Vytal, “keine gemeinsame Daten-Plattform”) — held open, with their shared constraint named: “Niemand verdient mit der Rücknahme”. And an internal inconsistency in the source itself: DIN SPEC 91446 on p. 13, DIN Spec 19446 on p. 38, recorded on both affected pages rather than silently corrected.

Open task 21 half-closed. 06 now has a page and the Röchling Stiftung page exists; organization/circle-economy.md, people-partners/silke-stremlau.md and report 07 remain.

Next: POLYPROBLEM 07 “Die Kluft im Kopf” the same way, then Toxopeus et al. 2021.

Commit due — wiki/ and sources/roechling/ carry uncommitted changes.

[2026-11-06] synthesis | Circular ecosystem maturity stages — POLYPROBLEM 06+07+08 read as one set

Instruction ⟨Tell Münzing, Slack, 2026-11-06, 10:41 and 10:49⟩: do not follow 07 on its own terms; bring 06 and 08 logic in, read the three as “ein ganzheitliches Set für die Transition”, show fit and contradictions, present the paradoxes named and unnamed. Then: “Die Stimmen in den Reports zeigen die unterschiedlichen Bewertungsmuster bzw. Reife des Systems wie die Stufen und Akteure ineinandergreifen. Da entsteht ein temporales to-do set. SINE bietet mit seinem Absatz für Teile davon Governance und Koordinationscapability.” Preceded by his correction that I had under-covered accounting/financial innovation and value-chain ecosystem development — that correction is the reason this page exists.

Reading done first, from the primary sources rather than the existing wiki pages: 07 in full (first complete read; lines 1–1480, the rest endnotes), 08’s financing and ecosystem passages again (pp. 4, 6, 9–10, 15, 22–24, 27–28, 30–35), and the existing 06 and 08 wiki pages.

New page: circular-ecosystem-maturity-stages.md. A SINE-authored five-stage model (S0 unvalued → S1 single-actor isolated → S2 bilaterally underwritten → S3 multilaterally synchronised → S4 bookable), each stage anchored in quoted voices from all three reports. The page’s own claims, marked as inference: (a) the stages interlock rather than sequence — one chain holds actors on different rungs simultaneously and performs at its least mature link, which is the mechanism behind 08’s “synchronisiert … entlang der Wertschöpfungskette” (p. 23) and behind Schuck’s 0–80 % variation across stores of the same chain (07, pp. 26–27); (b) each financier class can only underwrite at particular stages, so the financing gap is a stage mismatch, not a capital shortage — which reconciles Hafenmayer (pp. 30–31) and Betancourt Munoz (p. 34) with Nusseck’s valley of death (pp. 27–28); (c) a temporal to-do set T1–T7 with SINE coverage marked, including where it is absent (T1 pricing, T6 regulation, T7 behaviour execution are not ours); (d) the capability window — the offer binds across S2→S3 and nowhere else.

The strongest cross-report finding, recorded on the page and sent to Tell: all three reports state the same externalisation defect from three directions and none cites the others — 06’s cost-bearer ≠ value-captor, 07’s Rembe (“Die Kosten tragen Städte und Kommunen bzw. wir als Steuerzahlerinnen und Steuerzahler … Einweg muss stärker eingepreist werden und von den finanziell profitierenden Unternehmen getragen werden”, p. 27), 08’s “Folgekosten … nicht in den Bilanzen” (p. 4).

Unnamed paradoxes surfaced in the set reading (sent to Tell; the page carries the structural ones): the system paradox (08 quotes PREVENT’s “Finanzierung des Systems … anstatt einzelne Akteure”, then all three reports profile only individual actors); the disclosure paradox (Stremlau: banks “erheben die Daten, nutzen sie aber kaum aktiv”, taxonomy without “echte Steuerungsfunktion” — while Intesa Sanpaolo deploys €11.7bn on a private EMF-built methodology); the value-chain paradox (Jackson’s “je weiter man die Wertschöpfungskette hinaufgeht, desto schwerer wird es, Profitquellen zu erkennen”, 07 p. 10, is 08’s open distribution question restated behaviourally); the enforcement paradox (Wermter’s Indonesian burn ban with ~30 % still burned, 07 pp. 20–21, against 08’s regulation-as-de-risking, with 06 holding the only answer in the set and neither other report reaching for it); the taxonomy paradox (Betancourt wants more of what Stremlau calls steering-free, and a positive list compels disclosure where a negative list shifts the burden of proof).

Cross-references added: circularity-valuation-gap, ecosystem-financing, and the 06 paper page all now link to the new page.

Index: entry added under ## Concepts — plus an orphan fixed: ecosystem-financing.md had no index entry at all despite being linked from several pages. It now has one.

Open and unchanged: 07 has no wiki page yet (ingest pending, the source is now fully read); the 08 page still needs the accounting/financial-innovation nachtrag Tell asked for (Rabobank €270m, Intesa €11.7bn, 13 % vs 46 % bank-credit use, first-loss/90-10, taxonomy without steering function, Betancourt’s “andere Art von Risiko”, the Schwalbe chain); and a stress test of this model against biodiversity/nature-finance maturity is requested ⟨Tell Münzing, 2026-11-06⟩ and not yet done.

Commit due.

[2026-11-06] synthesis | Stress test — the maturity ladder against biodiversity

Instruction ⟨Tell Münzing, Slack, 2026-11-06, 11:02⟩: “Stress-Test Haltbarkeit Analogie zur maturity im Biodiversity topic.” Run against the wiki’s existing biodiversity material only (essential-variables, nature-related-reporting-frameworks, geo-bon, kim-essential-variables-2026) — no new sources fetched.

Result: the ladder transfers, and the test improved it. Filed as a new section on circular ecosystem maturity stages.

Holds: stage definitions map intact (S0 = Aichi missed with “the dearth of reporting measures” named as a contributing cause; S1 = voluntary pilots and no formal EESV variable list, the same fragmentation signature as 06’s twenty-plus collection apps; S3 = the EV layer, crosswalks, GBiOS, SEEA EA). The interlocking claim survives its strongest test: biodiversity carries S3 structure on S1 supply under S4 demand, and the wiki had independently recorded the predicted outcome (“a disclosure regime arriving ahead of the data layer … where carbon was before PACT”).

Two amendments forced, both now in the page:

1. S4 splits into S4a (disclosure demanded) and S4b (bookable). Biodiversity has S4a without S4b; the circular reports conflate the two and treat PPWR/quotas/EPR as investment signals when they are compliance demands landing on S1 supply. This is the mechanism behind Pöppelmann’s “erfolgreich gescheitert” and it was not visible in the circular sources alone.
2. S2 exists only where a bargaining component does. No S2 instance in state-side biodiversity — states and statistical offices have no rent to divide, which is the wiki’s own near-pure-coordination finding. So the domain runs S1→S3 directly, and the presence of an S2 rung is diagnostic of failure shape rather than a measure of progress.

Net effect on SINE’s capability window: it is not a stage but a stage plus a bargaining component — S2→S3 and a reporting entity exposed by disclosure. In biodiversity that holds at the corporate boundary only (TNFD), which converges with open task 10 and the still-unread TNFD/NGFS primary sources.

Counter-finding carried to the camp: biodiversity’s shared layer is funded because it was claimed as a public statistical function (SEEA EA under the UN Statistical Commission, sibling of the SNA). The circular shared layer was never claimed that way and is expected to find a private business model — which may make “who pays for the shared layer” the wrong question. The one privately funded circular instance in this wiki (cirplus financing DIN SPEC 91446, because its business model is transparency) shows how narrow the private route is.

Honest limit recorded on the page: one source community (Kim et al., written by the prospective operators of the layer they propose), a stub GEO BON page, no TNFD/NPI/NGFS primary source, and no source at all on biodiversity credits, offsets or debt-for-nature swaps — so the missing S2 rung is argued structurally, not demonstrated.

Cross-references added from nature-related-reporting-frameworks and essential-variables; index entry amended.

Commit due.

[2026-11-06] update | Wechsel-Geld (08) — Nachtrag on accounting/financial innovation and value-chain ecosystem development

Tell Münzing (Slack, #66): “08 Nachtrag und 07 ingest”. The Nachtrag closes the under-coverage he named in #60 — balance-sheet mechanics, financial innovation, and ecosystem development along the value chain.

Added to papers/polyproblem-wechsel-geld-2026.md:

• Two new ## Nachtrag ⟨2026-11-06⟩ sections, placed before “What this source does not contain” so the original summary is unchanged and the additions are visibly dated.
• Accounting and financial innovation — SINE’s own split of the report’s single “financing” problem into an accounting problem (no ledger line → no risk model) and an instrument problem (wrong duration, wrong collateral), with the finding that the second is already being solved privately and the first nowhere. Instrument table (Intesa Sanpaolo €11.7bn 2023 on its own EMF-built methodology; Rabobank Circular Entrepreneurship Desk €270m; blended finance/first-loss; the 90-10 rule; green-loan steering; the Einspeisevergütung analogue; 06’s offtake guarantee and levy). Stremlau’s taxonomy critique decomposed into coverage/usage/polarity. VC’s Asset-Light preference against “Infrastruktur ist jedoch … unerlässlich”.
• The value chain as the unit of ecosystem development — Schwalbe (p. 22), Stammer/TecPart (p. 24, newly quoted in full), Betancourt (p. 34) and the report’s own open distribution question (p. 23) read as one unit-of-account mismatch: cost-bearer ≠ value-captor.
• One contradiction newly represented, not resolved: Stremlau’s misestimated risk vs Betancourt’s “andere Art von Risiko” — calibration fix vs new risk category.
• Five rows added to “Key quantitative claims”: 13% vs 46% bank-credit use, Rabobank €270m, Intesa €11.7bn, CE private funds tenfolding 2016→mid-2020, +5pp peer outperformance.
• Cross-reference added to concepts/circular-ecosystem-maturity-stages.md.

Index entry for the page extended with the Nachtrag. Link check clean except the two pre-existing missing targets (organization/circle-economy.md, people-partners/silke-stremlau.md) already tracked in open tasks.

Commit due.

[2026-11-06] ingest | Technical Specifications for PCF Data Exchange v2.2.0 (WBCSD/PACT, April 2024)

Ingested sources/standards/2026-10-29-pact-data-exchange-protocol-v2.2.md at Martin Pompéry’s request, the day after the open-task-11 status review. Native Markdown, 2,421 lines, no conversion needed. Read in full in the main thread, no delegation. Source 3 of 3 of the 2026-10-29 PACT bundle — the bundle is now complete.

Framing corrected mid-ingest by Martin Pompéry before any page was written, and the corrections are load-bearing for everything below:

1. There are no canonical product IDs — only several. The Tech Specs are in that sense an interoperability profile. My first reading had filed the absence of a canonical identifier as a gap against the DPP; it is not a gap, it is the condition every DPP is in, and profiling is the available answer.
2. Postel’s law is a basic design principle of the Tech Specs generally. My first reading treated the protocol’s tolerance (best-effort filtering, 202 partial results, “should strive to accept”) as conformance guaranteeing syntax but not behaviour. Under Postel’s law that tolerance is the load-bearing principle, not a loophole: a small mandatory floor with liberal acceptance above it is what lets heterogeneous hosts interoperate and evolve independently.

Neither term occurs in the document (grep: zero hits for postel|robustness|liberal in what|conservative in what|profile), so both are written as ⟨Per Martin Pompéry, 2026-11-06⟩ wherever they carry a claim.

New page: papers/pact-data-exchange-protocol-v2.2.md.

Updated:

• concepts/pact-pathfinder-network.md — new “What the protocol actually is” subsection (profile, Postel, §7.2.1, SINE inside the spec); sixth and sharpest instance of open task 19 in the confidentiality section; discretionary disclosure (§4.1.2).
• papers/pact-network-vision-paper-2022.md — its five unspecified trust mechanisms answered by exclusion: §7.2.1 turns the white boxes of Figure 4 into scope exclusions. Also closes the page’s own open question about the third pillar: “ecosystem” occurs zero times in the Tech Specs.
• concepts/pact-data-quality-verification.md — how PDS, dqi and assurance travel on the wire, plus the deadline divergence (data model mandates dqi for reporting periods including 2025; Methodology v3 defers DQR to 31.12.2027), level: "reasonable" alive in the schema, the admitted providerName publication error, and the exemption cap moving in the opposite direction from the methodology.
• concepts/confidentiality-gap-in-carbon-accounting.md — addendum ⟨2026-11-06, the gap read at the level of the wire format⟩: the complete security inventory, confidentiality as advisory access control, non-disclosure as a specified outcome, and the argument that a profile declining to profile JWS is a dated selection decision.
• concepts/decision-grade-carbon-data-criteria.md — criteria B, C, E, F, H checked against a shipped protocol rather than a programme’s self-description. E is met well, F is not addressable at all (documents move, not quantities), B is out of scope by statement.
• concepts/digital-product-passport.md — profiling rather than harmonisation as the route to the page’s four use cases; sharpens the attachment objection by separating agreement on scheme from binding to the object.
• open-tasks.md — item 11 nine → eight (title, status line, table row); item 19 gets the verdict its revised “done looks like” asked for: (b) deliberately scoped to access control, written down in §7.2.1, with reading 1 dead for PACT and a fourth reading added (deliberately pushed below the interoperability boundary). Also corrects the wiki’s implication that SINE authored the specification — the document supports technology partner and reference host (pathfinder.sine.dev), not authorship.

Four divergences with already-ingested PACT material recorded as contradictions rather than resolved: DQR shape, DQR deadline, exemption cap direction, assurance.level: "reasonable".

Index updated (papers entry + the open-task counts in the header).

Commit due.

[2026-11-06] wiki | New concept page: interoperability profile

Follow-up to the PACT Tech Specs v2.2 ingest earlier today, at Martin Pompéry’s go. No new source — the page generalises a frame he supplied orally during that ingest (“there are no canonical product IDs … the Tech Specs are in that sense an interoperability profile”) into the standalone term the wiki had been missing.

Created: concepts/interoperability-profile.md.

Why it exists: the wiki kept generating the same invalid finding in different places — standard X does not define Y, therefore there is a gap. For a profile that inference does not hold; the substance is which pieces it picks, how tightly it constrains them, and what it leaves below.

What the page carries:

• Four moves — selection, constraint, obligation levels, tolerance. Tolerance is the one normally omitted from descriptions of profiles and the one that decides whether a heterogeneous network interoperates; in PACT it is Postel’s law. ⟨Per Martin Pompéry, 2026-11-06.⟩
• The representation / accountability boundary, lifted from the criteria check on decision-grade-carbon-data-criteria.md and stated as a general property: a profile fixes symbols, and accountability is a question about state. PACT has no state by the 2022 architectural exclusion, so conservation of quantities was unreachable before the protocol was written. The frame defends §7.2.1 against “incomplete” and sharpens it into “these are choices” — JWS was as available as OData.
• Profiles stack: CX-0136 on the Pathfinder Framework, iLEAP inside PACT via §5.5 extensions. “Both conform to PACT” ≠ “their data is exchangeable”. Connects to the standing line on agent-protocol-interoperability.md — an architecture’s real n² count is the count of its profiles.
• A profile is a frozen cut of moving standards — the 01.04.2026 PACT v2 sunset and iLEAP’s decoupling as the case; therefore governance (version policy, sunset policy, notice to downstream profiles) outweighs content.
• Conformance certifies the envelope, not the method — two separate certifications in PACT; iLEAP §8 says it in bold. So “40+ conformant providers” needs the question which conformance.
• Consequence for harmonisation: identifiers stay plural and interoperability happens anyway; and the honest limit of GS1 Digital Link — it settles the scheme, not the binding of symbol to object.

Cross-linked from (eight pages): papers/pact-data-exchange-protocol-v2.2.md, concepts/pact-pathfinder-network.md, concepts/ileap.md, concepts/decision-grade-carbon-data-criteria.md, concepts/digital-product-passport.md, concepts/agent-protocol-interoperability.md, and — both of which had no cross-reference section at all, so one was added — market-competitors/catena-x.md, market-competitors/gs1.md.

Index: new concepts entry, placed in the PACT/iLEAP cluster.

Deliberately not pulled in: GB/Z 185.1—2026, which is being ingested in a parallel thread and whose convening-authority angle belongs there first.

Commit due (together with the Tech Specs ingest).