Coherence Audit — April 2026

Status: Two passes complete. Pass 1 (PRINCIPLES.md anchor hallucinations) is recorded immediately below. Pass 2 (PROBLEM_MAP.md anchor hallucinations + a real cross-document drift between PRINCIPLES.md §5 and FOUNDATIONAL_COMMITMENTS.md §5) was triggered by Pass 1's verification step and is recorded in its own section further down. Both passes share a root cause — a single early bulk-authoring pass for source digests cited anchors and section titles that had never existed in any commit of the target documents.

Trigger: While creating sources/source-weekly-show-stewart-platner-digest.md, four references to a Principle 5 anchor (#5-inclusive-institutions-with-bounded-rules) failed verification — the current PRINCIPLES.md anchor is #5-critical-systems-require-public-interest-governance. The steward requested a full coherence audit per the Coherence Audit Protocol before deciding how to proceed with Platner-digest follow-up exchanges.

Scope covered: All four protocol categories (cross-document consistency, recommendation tracking, exchange index integrity, source corpus integrity), plus terminological consistency, across both project-2028 and civicblueprint.org. The retained external source corpus (external-formation-docs) was checked at the registry-and-folder level — no source-handling errors found in this pass.

Headline finding (revised April 29, 2026 after steward git-history check): seventeen source-digest files in sources/ carry anchors and body-text principle titles that have never existed in any commit of PRINCIPLES.md. Initial framing called this "drift" (one document evolving past another). The git log shows otherwise: §1 has only ever been "Human dignity is not conditional on market productivity" or its current "Dignity is inherent and unconditional"; §2 has only ever been "Essential needs should not be hostage to artificial/avoidable scarcity"; §3 has only ever been "AI must expand/augment human agency, not replace democratic control." The phantom titles "human dignity is non-negotiable," "the inheritance of humanity belongs to all of humanity," and "human purpose is the point of civilization" exist nowhere in PRINCIPLES.md git history. The correct classification is Hallucinated reference — anchors and titles invented by the digest-authoring agent at digest-creation time (commit d45f718 and adjacent), citing principles the source PRINCIPLES.md never carried. The Coherence Audit Protocol was updated as a result of this audit to add Hallucinated reference as a named issue type.

Why the reclassification matters for resolution: the prior framing offered three options (re-attribute, footnote-as-prior-version, mechanical anchor swap). With no prior version to footnote against, option (b) disappears and option (c) becomes actively wrong — a mechanical anchor swap would land the link on a principle whose actual content (e.g. essential needs / avoidable scarcity) does not match the digest's body-text claim (e.g. "common inheritance"), propagating the hallucination instead of fixing it. The remaining option is to re-author the affected principle-mapping sections against the actual current PRINCIPLES.md, which is what should have happened at digest creation time.

---

Document classes covered

• Core normative documents: PRINCIPLES.md, PROBLEM_MAP.md, SYSTEMS_FRAMEWORK.md, CONTRIBUTING.md, FOUNDATIONAL_COMMITMENTS.md.
• Project-authored analysis & coordination: agent/exchanges/**, agent/process/**, sources/**, formation-docs/**, README.md, ROADMAP.md, ROADMAP_ARCHIVE.md, civicblueprint.org/docs/** (via cross-repo references).
• Retained external source corpus: external-formation-docs/** — registry-and-folder level only. No source-handling errors found in this pass; deeper alignment-memo audit deferred to a future lightweight corpus-integrity check.

---

Coherence checklist

---

Resolution rules applied

Per protocol §6:

• Findings #1, #2, #3 (hallucinated principle anchors and titles in 17 source digests): Fix by re-authoring. Each affected digest's principle-mapping section is re-written against the actual current PRINCIPLES.md. Mechanical anchor-only swaps are explicitly rejected because the body-text principle titles are also hallucinated and would propagate the misattribution.
• Finding #4 (Exchange #21 §5 vocabulary): Already fixed in the Platner digest. No further action; watch item carries forward to F1 spawn.
• Finding #5 (Platner digest in dependency graph): Defer. Tied to the open exchange-spawn decision.
• Findings #6–#12: Accept. No drift detected; recorded here for the audit log.

---

Recommended order of operations

The original audit framed this as a multi-stage decision tree. The corrected diagnosis collapses it into a single workstream:

1. Re-author the 17 affected digests' principle-mapping sections against the actual current PRINCIPLES.md, using the candidate-principle guidance in Findings #1–#3 above. Per-digest interpretive judgment is required (Bastani's substance maps differently than Sandel's even though both originally cited "common inheritance"), but the judgment is the normal authoring task — not a forensic reconstruction of a phantom prior principle set.
2. Add an authoring guard to sources/README.md (and to the digest-authoring practice generally) requiring the digest author to quote the actual current principle title when claiming a principle linkage. This structurally prevents the same hallucination class.
3. Watch item — Exchange #21 F1 follow-up: when F1 spawns, the v2 vocabulary either gets adopted into PRINCIPLES.md (§5 title changes) or doesn't. Both outcomes are coherent; the issue is only that future digests should cite whichever §5 title is current at that moment. The Coherence Audit Protocol's trigger on "core normative document substantively revised" will catch this if F1 lands.

---

---

Pass 2 — PROBLEM_MAP.md hallucinations and PRINCIPLES.md ↔ FOUNDATIONAL_COMMITMENTS.md §5 drift

Trigger: While extracting and verifying every PRINCIPLES.md#... anchor in use across the repo at the end of Pass 1, the same programmatic-extraction approach was applied to PROBLEM_MAP.md, SYSTEMS_FRAMEWORK.md, and FOUNDATIONAL_COMMITMENTS.md to confirm Pass 1 was not missing parallel issues in adjacent core documents. PROBLEM_MAP.md surfaced a much larger version of the same hallucination class, and FOUNDATIONAL_COMMITMENTS.md surfaced a real cross-document drift on §5.

Pass 2 scope

Pass 2 coherence checklist

Reference: current PROBLEM_MAP.md domain set (the re-authoring target)

For Pass 2 re-authoring:

1. Energy and critical infrastructure are too constrained for a resilient, abundant future
2. Money, credit, and capital allocation steer the economy in ways most people cannot see or influence
3. Information ecosystems are fragmented and easily manipulated
4. Institutional capacity is too weak for the demands placed on it
5. Housing is constrained where it should be abundant
6. Healthcare is too often rationed by price, complexity, and access
7. Education and opportunity pathways are uneven, rigid, and bottlenecked
8. Food systems are efficient in some ways, but fragile in others
9. Family support systems are misaligned with modern life
10. Wealth and power are concentrating faster than governance can respond
11. AI and compute power are concentrating faster than governance can respond
12. Ecological systems are under stress that constrains every other domain
13. Institutional distrust is becoming a governing condition
14. Talent is systematically routed away from public-interest systems
15. Democratic process cannot convert public need into institutional action at the speed or scale required

Phantom domain phrasings the re-authoring eliminates: housing-has-decoupled-from-wages, healthcare-costs-are-inflating-faster-than-wages, education-has-inflated-faster-than-it-has-improved, childcare-is-financially-inaccessible-to-most-working-families, urbanization-is-squeezing-labor-mobility, ecological-overshoot-is-accelerating, a-post-scarcity-future-is-increasingly-possible-and-is-not-being-prepared-for, work-and-purpose-are-decoupling-faster-than-our-institutions-can-absorb, institutional-capacity-has-decayed-faster-than-the-problems-requiring-it, transformative-technology-is-arriving-faster-than-its-governance, information-ecosystems-fragment-attention-and-narrow-the-shared-reality-needed-to-coordinate.

Pass 2 resolution rules applied

• P2-1 (PROBLEM_MAP hallucinations): Fix by re-authoring per digest. Same protocol as Pass 1 Findings #1–#3.
• P2-2 (PRINCIPLES.md ↔ FC §5 drift): Flag-only, gated on ROADMAP F1.
• P2-3, P2-4 (SF and FC anchor health): Accept. No issues found.

Pass 2 remediation summary (closed at audit time)

• P2-1 closed. Every PROBLEM_MAP.md#... anchor used anywhere in the repo now resolves to a real current heading. The 13 distinct anchors in use map 1:1 onto the actual current 15-domain set (no use of §8 Food, §14 Talent — the digest corpus does not currently substantively engage either, which is itself a useful signal for future-sweep planning). Body-text "Domain N (Topic)" references were re-authored alongside the anchors so the paraphrase always faithfully summarizes the actual current §N. The lone non-digest phantom (in government-overreach-ownership-ratchet-exchange.md Round 5 narrative) was also fixed.
• P2-2 carried forward. Recorded as a flag-only finding. No edits to either core normative document. Open until ROADMAP TODO #1 F1 resolves. F1 spawned April 30, 2026 as Exchange #23 — Principle 5 Revision; P2-2 closes when that exchange's Round 5 lands the v3 text in PRINCIPLES.md and FOUNDATIONAL_COMMITMENTS.md at the same commit.
• Authoring guard extended. sources/README.md authoring guard was rewritten to cover all four core normative documents (PRINCIPLES, FOUNDATIONAL_COMMITMENTS, PROBLEM_MAP, SYSTEMS_FRAMEWORK) with a per-document verification table and an explicit caveat about the §5 cross-document drift.

---

Items deferred to next audit

• Alignment-memo deeper check. This pass verified that every entry in formation-docs/SOURCE_REGISTRY.md has a matching folder in external-formation-docs/documents/. It did not verify that each analysis/principle-maps/*.md memo's interpretive claims still match the retained excerpt in the corresponding external-formation-docs/documents/* folder. That is the protocol's "lightweight corpus-integrity check" and should run on its own cadence — likely after the next formation-doc analysis update or after F1 changes the comparison target (PRINCIPLES.md §5 vocabulary).
• Filename typo iniital-systems-framework-review.md. Renaming would require updating every link to Exchange #2 across the project. Cost > benefit alone, but if a broader URL-cleanup pass is ever run, this should ride along.
• PRINCIPLES.md §5 vs FOUNDATIONAL_COMMITMENTS.md §5 drift (P2-2). Carries forward as an open finding. F1 spawned April 30, 2026 as Exchange #23; P2-2 will close when Exchange #23 Round 5 integrates the v3 text into both documents.

---

Next-audit trigger conditions

Per Coherence Audit Protocol §1, the next full audit should run when any of the following occurs:

• ROADMAP TODO #1 F1 lands (resolves the §5 drift between PRINCIPLES.md and FOUNDATIONAL_COMMITMENTS.md).
• A new exchange opens that consumes the Platner digest (resolves Pass 1 Finding #5).
• A core normative document is substantively revised.
• Approximately 3–5 new exchanges have completed since this audit.

A lightweight corpus-integrity check should run after any change to external-formation-docs or to formation-docs/analysis/.