Coherence Audit — June 2026

Status: Complete. This is the second full audit under the Coherence Audit Protocol (the first was April 2026). It was triggered by the June 13–14 2026 flurry: two new steward-anchor source digests (Suits Grasshopper, Stewart × Slobodian Muskism; the Brooks #1109 digest was also recent), three new exchanges (#29, #30, #31), and the accompanying index/roadmap updates. Per protocol §1, "a new exchange is created" and "approximately every 3–5 exchanges" both fired.

Method (steward-authorized): A deterministic mechanical pre-pass (scripts/coherence-check.py) was run repo-wide for link/anchor integrity and the Hallucinated-vs-Stale distinction. Then two independent interpretive sub-agents on different model lineages (one GPT-5.5, one Gemini 3.1 Pro), each blind to the other, ran the protocol's interpretive layer (exchange-index integrity, count/set-equality, cross-document drift, recommendation tracking, terminology) and returned coherence checklists. The auditing agent (Claude) reconciled the three outputs. The cross-lineage step earned its keep: the two interpretive agents converged on the high-risk mechanics being clean and diverged on two real findings neither would have caught alone (see Cross-lineage reconciliation).

Headline: The high-risk recent mechanics are clean — #29/#30/#31 are correctly indexed with statuses matching their status blocks, the dependency graph includes them, the "57 digests" and steward-anchor counts are accurate, and the newly created Muskism digest and Exchange #31 carry no broken links or anchors. The substantive finding is a drift: when the prior session opened #29 and #30, it updated the Brooks digest's candidates table and cross-references but left three downstream locations still describing those exchanges as a "candidate"/"future"/"not-yet-in-corpus" idea. That drift, one flurry-introduced broken link, and one metadata gap are fixed in this pass. The mechanical pre-pass also surfaced a cluster of pre-existing broken links/anchors (the April audit predates this script); they are catalogued and carried forward.

---

Document classes covered

• Core normative documents: PRINCIPLES.md, PROBLEM_MAP.md, FOUNDATIONAL_COMMITMENTS.md — referenced as link targets; no substantive revision occurred in the flurry, so they were checked for inbound-reference accuracy only.
• Project-authored analysis & coordination (primary scope): sources/** (the three new digests + the index/README), agent/exchanges/** (#29/#30/#31 + the index), agent/explorations/** (the new riffs + the cross-linked riffs), agent/process/**, ROADMAP.md.
• Retained external source corpus: external-formation-docs/** — not touched in the flurry; no source-handling, translation-status, or canonical-source findings in this pass. No lightweight corpus-integrity check was triggered.

---

Coherence checklist — fixed this pass

---

Verified clean (recorded for the audit log)

Independently confirmed by both interpretive lineages and spot-checked by the auditing agent:

• Digest count. 57 sources/source-*-digest.md files on disk; the "57" claim in SOURCE_INDEX.md and sources/README.md is correct.
• Steward-anchor counting. Friedberg #1, Acemoglu/Autor #2, Platner #3, Brooks #4, Muskism #5; Muskism is correctly the third Weekly Show anchor. Consistent across the Muskism digest, SOURCE_INDEX, and README.
• Exchange index integrity. #29, #30, #31 are present as numbered entries; their index Status values ("Active; Round 1 complete; Rounds 2–N reserved") match their file status blocks; all three appear in the dependency graph. Set-equality holds: 31 numbered exchanges, one index file, and three intentionally non-numbered discovery-principle-develop-leg-* support artifacts.
• New-artifact link health. The Muskism digest and Exchange #31 — the two artifacts created in this chat session — produced zero mechanical findings.
• Roadmap tracking. ROADMAP.md tracks #29/#30/#31, the Brooks/Suits/Muskism digests, the reserved Round-2 next step, and the open Suits-"transition" steward decision.
• Terminology. The Suits "a project about transition" framing is held consistently as an open/candidate lens (not a ratified thesis) across the Suits digest, the constitutional-ecology riff addendum, and the ROADMAP.

---

Coherence checklist — carried forward (pre-existing; newly surfaced by the mechanical checker)

The April 2026 audit predates scripts/coherence-check.py; this is the first audit to run it, so the script surfaced a backlog of pre-existing link/anchor breakage outside the flurry — ≈23 broken references across 12 files, grouped below into 5 clusters. None were introduced by the recent activity, and the two artifacts created in this session (the Muskism digest and Exchange #31) are clean. They are deferred (not silently auto-fixed) because each needs interpretive judgment or a content decision, and per protocol §3 a hallucinated/stale anchor must be re-evaluated against the actual target rather than substituted with the closest match. (An earlier draft of this audit under-counted these, having read only a tail-truncated checker run; the full untruncated run is recorded here.)

---

Cross-lineage reconciliation

The steward authorized parallel sub-agents to "compare notes." Outcome:

• Convergence (high confidence): Both the GPT-5.5 and Gemini 3.1 Pro agents independently verified the 57-digest count, the steward-anchor counting, the #29/#30/#31 index entries and status-block agreement, the dependency-graph inclusion, and the exchange set-equality. Independent agreement across lineages is the strongest signal that the recent mechanics are sound.
• Convergence (a real finding): Both flagged that the Brooks digest is described as cross-cutting sub-debates but is not represented the way the convention would suggest. Reconciliation refined this: the digest's own body claims Sub-debates 1 and 5 (the index adds 9), and the substantive issue was the missing sub_debates front matter plus an open categorization-convention question (do steward-anchor digests belong in the per-sub-debate tables, or only in the cross-cutting anchor table?). Fixed the metadata (F3); logged the convention question as a steward decision (below).
• Divergence — GPT-5.5 only: Caught the stale "candidate / not-yet-in-corpus" status language for #29/#30 across the Brooks digest, SOURCE_INDEX, and README (finding F2). This is the audit's headline substantive finding and would have been missed by the other lineage. Confirmed against the files and fixed.
• Divergence — Gemini 3.1 Pro only: Caught the missing sub_debates front matter (folded into F3) and raised whether the Round-3 (external-human) reviewer needs of #29/#30/#31 should be consolidated under ROADMAP TODO #11 (the human-reviewer recruitment workstream). Reconciliation: Accept/Defer — Round 3 is correctly gated behind the agent-runnable Round 2, which has not yet been run; consolidating into TODO #11 now would be premature (the same logic the April audit applied to deferring the Platner digest into the dependency graph). Revisit when any of the three Round 2 passes completes.

A single-lineage audit would have caught one of F2/F3 but likely not both. The compare-notes step is recorded here as evidence the method adds value for this corpus.

---

Steward decisions (resolved June 14, 2026)

1. Steward-anchor categorization convention → "Normalize" (executed June 14). The steward chose to dual-list all five anchors in their per-sub-debate tables (the Muskism precedent). Acemoglu/Autor (Sub-debates 3, 4, 6, 7, 8), Platner (3, 8), and Muskism (3, 4, 8) were already dual-listed; this pass added Friedberg to Sub-debates 1, 2, 3, 4, 5 (one dedicated cluster each) and Brooks to Sub-debates 1, 5, 9 (the last as a cautionary scientism cross-listing, with the §9 intro reworded from "single digest"). While normalizing, a fourth instance of the F2 stale-"candidate" drift was found in the SOURCE_INDEX status note and fixed (see F2b).
2. Round-3 human-reviewer tracking for #29/#30/#31 → accepted the deferral (June 14). The steward accepted that Round 3 stays gated behind the agent-runnable Round 2 and is not consolidated into ROADMAP TODO #11 now; revisit when any of the three Round 2 passes runs.

---

Carried forward from the April 2026 audit (not re-litigated)

• P2-2 — PRINCIPLES.md §5 ↔ FOUNDATIONAL_COMMITMENTS.md §5 drift: tracked to close when Exchange #23 Round 5 lands the v3 §5 text into both documents at one commit. Not revisited in this pass.
• Alignment-memo deeper corpus-integrity check (formation-docs/analysis/** vs retained external-formation-docs excerpts): still deferred to its own cadence; no external-formation-docs change triggered it.
• Filename typo iniital-systems-framework-review.md (Exchange #2): accepted; rename only as part of a broader URL-cleanup pass.

---

Resolution summary

• Fixed this pass (6): F1 (Suits broken link, flurry), F2a/F2b/F2c (Brooks→#29/#30 stale-status drift across three artifacts), F3 (Brooks sub_debates metadata), F4 (two safe pre-existing path typos).
• Carried forward (pre-existing, deferred with precise resolutions): ≈23 references in 5 clusters (C1–C5) — a deleted SCRATCHPAD.md, a never-created reviewer package, an anchor cluster from the research-protocol / constitutional-ecology / #24-exchange co-authoring pass, a set of renamed/archived ROADMAP.md anchors, and two never-written-section self-anchors.
• Carried forward from April: 3 items, unchanged.

The mechanical checker still exits non-zero (the C1–C5 pre-existing breakages remain). That is expected; closing C1–C5 is a recommended focused follow-up, ideally before the checker is wired into CI as a pre-commit gate. Note these are entirely pre-existing — the recent flurry introduced exactly one broken link (F1), now fixed.

---

Next-audit trigger conditions

Per Coherence Audit Protocol §1:

• Any Round 2 of #29/#30/#31 completes (changes their status and may spawn adopt-targets).
• The C1–C5 carried-forward clusters are cleaned up (re-run the checker to confirm a clean exit).
• A core normative document is substantively revised, or Exchange #23 lands the §5 v3 text (closes April P2-2).
• Approximately 3–5 further exchanges complete.
• A lightweight corpus-integrity check should run after any change to external-formation-docs or formation-docs/analysis/.