Plan review — Reviewer Y

Date: 2026-05-12 Plan version reviewed: v2 (2026-05-12)

Summary

The plan is a structurally elegant adaptation of math-olympiad to bridge-card sign-off — dual context isolation, asymmetric quorum, hypothesis-note discipline, pattern-armed adversaries. As internal procedural mechanics it is solid. My adversarial focus, however, is on what the procedure does not see: the cases it would silently pass, the corpus-specific failure modes it does not name, the gray-zone writes it implicitly authorizes, and the metric vacuum at the pilot-to-codify boundary.

The top concerns, in priority order: (1) the procedure does not reckon with positive-result publication bias — main-thread Claude wrote the cards in Pass 3 and now adjudicates sign-off, which is a structural conflict the plan handwaves with "subagents are isolated"; the adjudicator is not isolated. (2) Several corpus-specific failure modes (philological inversion, MP-period miscoding, secondary-author scaffolding mistaken for MP's own register) are folded into pattern #1 "editorial-frame laundering" but are sufficiently distinct that one pattern under-detects all three. (3) The carve-out boundary leaks: a "subsection write on What the Concept Does" can in practice be a corpus-level thesis claim, especially for HUB pages, and the plan offers no guardrail beyond apply-as wording. (4) Pilot success/failure metric is undefined. (5) The math-olympiad analogy breaks at a load-bearing place: math has crisp truth conditions; bridge cards do not, so "DEFECT FOUND" is calibrated against a soft target and "APPROVED" means much less than HOLDS. Without a calibration step that maps DEFECT/APPROVED back to interpretive risk levels, the procedure may produce false confidence.

Verdict: ACCEPT-WITH-CHANGES. The plan is fundamentally sound; the changes are additive (extra patterns, an adjudicator-isolation step, a defined pilot metric, three carve-out tightenings) and can be absorbed before the v0d.7 codification step.

Failure scenarios the plan does not cover

Scenario 1 — Adjudicator self-confirmation bias

Description. Main-thread Claude drafted the Pass 3 cards. The cards' framing, primary-home selection, and counterpressure phrasing are main-thread's own. When 2 APPROVED + 1 DEFECT arrives, main-thread reads the DEFECT report and decides. Main-thread's prior interpretive commitments to the card content (it just wrote them) bias adjudication toward "the dissent is mild" / "concern already in card §Counterpressure" / "bridge survives at subsection level."

Why the plan misses it. §2.2 ("dual context isolation") isolates the subagents but says nothing about the adjudicator. §11 names main-thread as adjudicator without any structural counterbalance. The example sign-off lines in §11 even model this bias: "adjudication: B1's editorial-frame concern already in card §Counterpressure; bridge survives at subsection level; sign-off proceeds with apply-as item 5 narrowed" — the model adjudication is to absorb the dissent rather than to be moved by it.

Fix. Either (a) require that the adjudicator agent be invoked with fresh context and no knowledge that it wrote the Pass 3 cards — the subagent receives the 3 reports + the card text + the schema gate + the failure-mode taxonomy, and produces an adjudication report; main-thread then ratifies the subagent's adjudication only when adjudication = APPROVED, and escalates to user when adjudication = BLOCK or UNCLEAR. Or (b) require any 1-DEFECT outcome where the DEFECT is at high confidence to default to BLOCK unless the user is consulted, regardless of A and C's verdicts. (a) is preferable; it preserves the dual-isolation discipline at the adjudication step.

Scenario 2 — Same-gap-twice fires but the gap is the plan's gap, not the cards'

Description. §10 "same-gap-twice" assumes that if 3+ cards fail the same pattern, the upstream silent-keys report or the schema is buggy. There is a third option: the failure-mode taxonomy itself is mis-tuned and is over-flagging a pattern that's actually fine. The pattern most at risk: #1 editorial-frame laundering. The Pass 3 report itself acknowledges editorial-frame counterpressure in Cards 1, 2, and 4. If all three trigger #1 because the taxonomy requires MP's own register and these cards genuinely operate at the editorial-register-but-anchored-in-MP level, the procedure may block productive bridges.

Why the plan misses it. §10's "the schema or the upstream silent-keys report likely has a systematic bug" is binary; it does not include "the taxonomy is mis-tuned." §13's pilot framing also does not enumerate this.

Fix. Extend §10's "if same-gap-twice fires" branch to include: (c) the failure-mode taxonomy is mis-calibrated for this card cohort; the same-gap-twice check should produce a diagnostic note that names which of the three options (schema bug / upstream bug / taxonomy mis-tuning) the pause reflects, and require an explicit choice rationale before re-piloting. Without this, the procedure could keep firing same-gap-twice and rewriting upstream reports that aren't broken.

Scenario 3 — Card cross-dependency rewrites itself in adjudication

Description. §14 acknowledges Card 5 cross-references Card 1's déposer. Cards 1, 5 (and arguably Card 2 via déposer's "deposit a sense") form a dependency triangle. If Card 1 is APPROVED but Card 5 is BLOCKED, Card 1's apply-as item 4 (reprise as "takable-up dimension") might no longer be self-consistent with the apply-mode write set. The procedure says "evaluate each card independently; cross-card dependencies are flagged as concerns but do not affect the per-card verdict." But apply-mode writes are not independent — Card 5's open-question entry on [[institution]] cross-references Card 1's déposer, so if Card 5 is blocked, the cross-reference becomes a dangling pointer in any apply-mode write Card 1 authorizes.

Why the plan misses it. The independence convention is appropriate for verdict but inappropriate for apply-as scope. The plan does not have a post-adjudication consistency check.

Fix. Add a post-adjudication "dependency reconciliation" step (§11.5 or §11a): after all 5 cards are adjudicated, main-thread re-reads any apply-as items that cross-reference other cards; if a referenced card was BLOCKED or RETIRED, the cross-referencing apply-as item is narrowed or struck. Recorded in the adjudication.md file.

Scenario 4 — Anchor-cite drift between extraction note and raw

Description. The plan's Role C verifies that anchors exist. It also (§6) caps raw reads at 5 per card per reviewer and forbids re-ingest. But the Phase 8 cite-back failure modes I have on record (per my MEMORY.md) include "iterate Wiki homes fields not prose; verify stale statuses; watch for dead claim-refs." A specific failure mode: an extraction-note anchor cites I&P p. 77, the raw passage exists at p. 77, but the meaning Caraus assigns to the passage in extraction-note line 285 is more interpretive than the passage warrants. Role C can verify "the cite is accurate" without catching "Caraus's gloss is over-reading." The cap-of-5 raw reads makes that catch unlikely.

Why the plan misses it. §5 Role C item 3 ("verify the anchor says what the card claims it says") implicitly delegates the interpretive verification to Role C; but Role C is empirical, and the empirical check at 5-raw-reads-per-card is short of the interpretive-philological check needed.

Fix. Either (a) add an interpretive sub-check to Role C (cite checks include a 1-paragraph philological gloss: "the passage says X; the card claims it says Y; X and Y match because Z"); or (b) split Role C into C-empirical (anchor existence) and C-interpretive (passage-says-what-card-claims); two reviewers per role would be 8 instead of 6 invocations for the 5-card batch. Option (a) is lighter; option (b) is more robust.

Scenario 5 — Single-source promotion blocker honored at sign-off, then quietly bypassed at apply-mode

Description. Card 5 explicitly carries a single-source promotion blocker. Sign-off says: apply-mode is limited to subsection + open-question writes; no claims.md entry, no motif promotion. Then in the apply-mode follow-up branch, the agent doing the writes interprets "subsection on [[transtemporality]] under What the Concept Does" as authorizing a paragraph-length, thesis-shaped statement of "the past that could have been otherwise" as the modal hinge. That subsection paragraph IS a thesis claim, even though it's not in claims.md. The single-source blocker is honored literally but evaded substantively.

Why the plan misses it. §12 "what sign-off authorizes" lists "subsection writes on concept pages (under existing or new headings within 'What the Concept Does' / 'Stakes' / 'Open Questions')." It does not constrain the shape or epistemic force of the subsection content. The shape constraint is implicit in the bridge card's "what it makes visible" field but is not enforced.

Fix. Add §12a "subsection-write epistemic-force constraint": single-source-blocker cards may write subsections under Open Questions only (not What the Concept Does or Stakes), and the subsection prose must use provisional framing ("Larison reads X as Y; the claim is not yet corroborated outside M-C 2026"), not declarative framing ("X is Y"). Same constraint for any single-source card. Sign-off shall include explicit reference to the constraint in the Approved by maintainer: line for blocker-bearing cards.

Scenario 6 — Hypothesis note becomes ritual

Description. Reviewer reads the card, forms an initial impression, performs the anchor checks, and the hypothesis does not change. The note is written as if the convergence step did work, but it's actually post-hoc justification of the initial impression. The procedure cannot tell.

Why the plan misses it. §7 says "the note IS the reasoning trace" but does not require the note to evidence update. If the note's "Initial reading" and "Convergence" are substantively identical, the convergence step was a ritual.

Fix. Add explicit field: "What changed between initial reading and convergence" or "If nothing changed, explain why the anchor checks did not move the impression." A reviewer that consistently writes "nothing changed" on every card is a calibration signal (likely under-engaged).

Scenario 7 — Pilot evaluation criteria undefined

Description. §13 says "if pilot fails, revise; if pilot succeeds, codify." "Fail" and "succeed" are undefined. Concretely: what counts as failure? (a) Any DEFECT in any card? (b) Same-gap-twice firing? (c) Adjudicator needing to BLOCK 2+ cards? (d) The procedure surfacing a schema bug that requires editing this plan?

Why the plan misses it. §13 names the categories of outcome (gate-too-lax, gate-too-strict, schema issue) but offers no specific threshold or metric.

Fix. Define pilot success metric explicitly. My suggested criteria: pilot SUCCESS = (i) ≥3 of 5 cards reach a verdict (APPROVED or BLOCK) via the procedure as written without escalation to user; (ii) no same-gap-twice firing that diagnoses a taxonomy issue (schema or upstream issues count as procedure-working-as-intended); (iii) no card requires more than 1 revise cycle to reach a stable verdict; (iv) the human user reviewing adjudication.md does not override any per-card decision. Pilot FAILURE = any of the above fails. Pilot AMBIGUOUS = mixed (e.g., 1 card escalates but 4 reach verdict cleanly); in ambiguous cases, codify with documented caveats rather than re-piloting.

Scenario 8 — Drift across pilot → live use

Description. The pilot operates on 5 cards from a single upstream report (silent-keys-2026-05-07.md). Live use will operate on cards from multiple upstreams — silent-keys-delta, structural-isomorphism scans, motif aggregation. Live cards will not share the same evidential profile.

Why the plan misses it. §13 frames pilot-vs-live as procedure-validation only, not as evidential-profile generalization.

Fix. Codify the procedure with an explicit "card-cohort fingerprint" requirement: every run records (a) source-report type, (b) number of cards, (c) bridge-type distribution, (d) primary-home overlap. When cohort fingerprint differs significantly from prior runs, expect higher escalation-to-user rate and budget accordingly. Document this in the schema-changelog entry, not as a process metric.

Scenario 9 — The plan does not say what happens to BLOCKED cards

Description. §11 "if block: write rationale; card status remains proposed; flag for revision in a follow-up Pass 3 run." But a Pass 3 run is heavy infrastructure to revise a single card. The plan does not specify whether main-thread can revise a card in-line, or whether the card sits at proposed indefinitely.

Why the plan misses it. Card-lifecycle is treated as binary (sign-off OR retire); the revise-and-resubmit path is gestured at but not specified.

Fix. Add §11b "card revision pathway": a BLOCKED card may be revised in the same run by main-thread (with explicit log of the revision rationale in adjudication.md) and re-submitted to a fresh-context round of A/B/C review (not the same agents — they have stale verdicts). Cap: 1 revise-and-resubmit per card per run. If the second round also BLOCKS, the card retires.

Adversarial bridge-card construction

Here is a bridge card that would satisfy all 6 schema gate tests AND survive all 11 adversarial patterns as currently formulated, but is nevertheless a bad bridge.

Card X — l'invisible du visible / "the invisible of the visible"

Status: proposed Bridge type: mechanism Primary home: [[indirect-ontology]] Secondary homes: [[chair]] (or [[flesh]]), [[empietement]]

Relation:

• [[indirect-ontology]] is the condition of intelligibility of the visible-as-bearing-invisibility structure.
• [[chair]] enacts the same structure at the carnal-perceptual level (flesh as visible-with-invisible-depth).
• [[empietement]] enacts a structural variant (overlap-with-non-coincidence).

Evidence anchors:

• MP V&I (real page reference, real passage where MP uses "l'invisible du visible")
• SA-2006 extraction note line (real, a passage where Saint Aubert discusses this structure)
• Lanzirotti chapter (real, M-C 2026, a passage gesturing at the same structure)

What it makes visible: indirect ontology's positive content (not just refusal of presence-metaphysics) requires a structural account of how invisibility belongs to visibility rather than standing outside it. Without naming the structure, indirect ontology reduces to negative phenomenology.

Counterpressure: the "invisible of the visible" formula is famously associated with MP, but its function in late ontology has competing readings (Dillon's flesh as element, Lawlor's imperceptible, Barbaras's desire). The bridge could be reading one school's gloss back into MP's text. Counter-test: check whether MP's own usage in working notes (§"L'invisible" 1959-60) deploys the formula with the structural force the bridge claims.

Apply as:

1. Subsection on [[indirect-ontology]] (What the Concept Does): name the "invisible of the visible" as the structural mechanism.
2. Typed connection on [[chair]]: enacts indirect ontology's invisibility-in-visibility structure.
3. Typed connection on [[empietement]]: enacts a structural variant.

Why this passes all 11 patterns:

• #1 editorial-frame laundering: clears it — MP's own working notes use the formula (anchor 1).
• #2 single-source promotion creep: clears it — apply-as items are subsection + typed connections, no claims.md / motif moves.
• #3 false genealogy: clears it — relation language is "is the condition of intelligibility of" / "enacts" / "enacts a structural variant" — no genealogical chain claimed.
• #4 anchor-counterpressure inversion: clears it — counter-test is a specific check on working notes.
• #5 primary home arbitrariness: clears it — indirect-ontology is justified as the "condition-of-intelligibility" home, with chair and empietement as enactment-pages.
• #6 apply-as scope creep: clears it — each apply-as item has a directly-warranting anchor.
• #7 bridge-type conflation: clears it — "mechanism" matches the claimed structural account.
• #8 hovering register: clears it — each connection has a specific direction (intelligibility / enactment / structural variant).
• #9 fabricated citation: clears it — all anchors are real.
• #10 longstanding controversy: marginally clears it — counterpressure explicitly names Dillon / Lawlor / Barbaras as competing readings, and the bridge declines to adjudicate.
• #11 too-clean-bridge: clears it — the bridge is not clean; it's textured with three secondary homes and an explicit acknowledgment of competing readings.

Why it is nevertheless a bad bridge. The formula "l'invisible du visible" is one of the most-glossed phrases in late MP scholarship. The bridge claims to name a "structural mechanism" but the actual interpretive work has been done by 50 years of secondary commentary — what the wiki adds is a vocabulary choice rather than a new sense-distinction. The bridge is well-formed but it does not earn its keep: it occupies the structural-mechanism slot for an already-busy phrase and locks in a vocabulary at the wrong level of generality. A reader looking at the resulting concept-page subsection learns nothing they could not have learned from a competent MP secondary text.

What pattern would catch it. Call it #12 saturation-bridge / over-glossed-phrase pattern: a bridge whose primary phrase is so heavily glossed in secondary literature that no single bridge can clarify its function without taking a position the wiki has not earned. Check: search secondary-source attestations for the bridge's primary phrase; if the phrase appears as a heading-level or chapter-organizing term in ≥3 distinct interpretive schools, the bridge is in saturation territory and must either (a) take an explicit position in the dispute with extra counterpressure work, or (b) downgrade from "mechanism" or "structural" to "vocabulary-track" — a different and lighter bridge type.

What's missing from the taxonomy more broadly. Pattern #10 already gestures at "longstanding controversy" but #10's check is about whether the bridge's general form encroaches on a controversy — i.e., whether the bridge takes a position. The saturation pattern is different: the bridge can decline to take a position (by listing the competing schools in counterpressure) and still be a bad bridge, because the absence of a position becomes the position. #10 under-catches this; a distinct pattern is needed.

Missing patterns / tests

Missing pattern A — Philological inversion

Failure mode. A French/German/Greek/Latin term gets paired with an English gloss that inverts its semantic register. Example: déposer renders most naturally as "to deposit" or "to lay down" in non-philosophical French; its philological force as MP uses it has an active-but-not-volitional valence that the English "deposit" partially conveys but the noun-form "deposit" entirely loses. A bridge that pairs déposer with "deposit a sense" risks turning a verb-mechanism into a noun-residue, which is exactly what Card 1's counterpressure tries to prevent — but the counter-test is structural, not philological.

Suggested test. "Does the card's vocabulary work consistently across the original language and translation? If the primary term is in another language, does the English gloss preserve register, tense/aspect, voice, and grammatical class?" This is an existing concern in Layer 2's bilingual-corpus discipline (Page Format §Terminology); the bridge-card layer should inherit it.

Where in taxonomy. New pattern #13: "philological inversion / translation-register drift." Role B applies it primarily on cards with non-English primary terms.

Missing pattern B — MP-period miscoding

Failure mode. MP's published, unpublished, and working-note material spans 1935 (early manuscripts) → 1961 (last notes). The "1953–55 period" is a recognized transition zone (CLAUDE.md mentions it explicitly: "is a middle term between... known transition zones when relevant"). A bridge that conflates PhP 1945 register with V&I 1959-60 register is mishandling MP-corpus chronology — even when the cited passages are genuine. Card 2 (haunting / obsessive presence) explicitly cross-references PhP "obsessive presence" with I&P "haunts" and V&I working notes' "phantom" — three different periods. Is the modal-temporal register the same across all three? The counterpressure flags this but the counter-test is "raw-source check of local context," not "check periodization."

Suggested test. "If the bridge crosses MP-period boundaries (PoP / 1953-55 lectures / V&I-era / institution-lectures / late notes), does the bridge name the periodization explicitly, and does the relation language (is a reformulation of / is a middle term between) acknowledge the chronology?" A bridge that asserts enacts across periods without naming the periodization is mishandling the corpus.

Where in taxonomy. New pattern #14: "MP-period boundary mishandling / chronology elision." Role B applies it on multi-period cards (Cards 2 and 4 in this batch are at risk).

Missing pattern C — Secondary-author scaffolding mistaken for MP's own register

Failure mode. Distinct from #1 editorial-frame laundering. #1 catches "secondary author X uses term Y, card claims Y is MP's term." The missing variant catches: secondary author X uses term Y as a scaffolding term that organizes X's reading of MP, but Y is not MP's own term; the card claims Y is a stable MP-corpus register; X did not intend Y as MP's term — X coined Y as scaffolding. Card 1's déposer is at risk: the verb-form is MP's (I&P p. 77), but the centralization as institution's master mechanism across 4 chapters is the M-C 2026 editorial frame's scaffolding, and the bridge would promote the scaffolding to a stable MP register. Counterpressure acknowledges this; the counter-test is "does déposer operate this way in PoP / Husserl-limits / Inédits I-II?" — which is the right counter-test but is not enforced by any pattern.

Suggested test. "Is the term's function in the cited secondary literature scaffolding (organizing the secondary author's reading) or attested (a stable MP-corpus register)? If scaffolding, the bridge is staging the secondary literature's reading, not MP's own."

Where in taxonomy. Refine pattern #1 to have two sub-patterns: #1a "lexical laundering" (secondary author's term presented as MP's) and #1b "scaffolding promotion" (secondary author's organizing scaffolding presented as MP-corpus register). #1b is the more common failure mode for Cards 1, 2, 4.

Missing pattern D — Status-gate-mismatched apply-as

Failure mode. The bridge card's apply-as items, executed as written, would write content that implicitly requires live or supported claim status — but the claim-promotion gate has not been run. Example: an apply-as item "subsection on [[indirect-ontology]] (Stakes section): name philosophie / ontologie militante as the attitudinal mode of indirect ontology" reads as a thesis claim about MP's late ontology, with cross-source genealogical (1947→1958→post-1958) structure. If this were a claim entry, it would be promotion-from-candidate-to-live at minimum; if it claimed to "name MP's attitudinal mode," potentially supported. But the carve-out (§12) authorizes the subsection write without invoking the claim-promotion gate.

Suggested test. "Does any apply-as item write content whose epistemic force matches a live or supported claim, without the claim-promotion gate being run? If yes, either (a) downgrade the apply-as item to Open Questions framing, or (b) require the claim-promotion gate before sign-off."

Where in taxonomy. New pattern #15: "claim-status laundering via subsection-write carve-out." Critical for cards 1 and 4 in this batch.

Missing pattern E — Wiki-internal terminology collision

Failure mode. The bridge introduces a term that is already in wiki-internal use under a different sense. Example: if Card 1 wrote a subsection on [[institution]] introducing déposer as "the verb-form distinguishing institution from constitution and sedimentation," and the wiki already uses "verb-form" elsewhere to mean something else (e.g., the wiki's existing distinction between constitution-as-noun and constituer-as-verb in the Husserl-MP chapters), the bridge creates an internal terminology collision that future readers will trip on.

Suggested test. "Does the bridge's load-bearing vocabulary (verb-form / mechanism / call-side / residue-side / etc.) collide with existing wiki usage of the same words in different senses? Grep wiki/concepts/ and wiki/motifs.md for the bridge's distinctive vocabulary."

Where in taxonomy. New pattern #16: "wiki-internal terminology collision." Role A (gate-applier) can run this mechanically; it's a structural check.

Missing pattern F — False-friend caution evaded

Failure mode. CLAUDE.md §Sourcing Rules for Claims includes "false-friend caution": "When assembling claims from existing artifacts, mark tempting-but-unsafe analogies as false-friend cautions inside the relevant concept page's Connections or Open Questions section, rather than promoting them to candidate claims." A bridge card is an analogue of a candidate claim. The bridge-card layer offers a tempting end-run around the false-friend rule: a candidate claim would have been blocked but a bridge card with "similar" apply-as items survives.

Suggested test. "If this bridge were a candidate claim entry instead of a bridge card, would General Rules 17 / 18 or the Sourcing Rules for Claims block it? In particular: is the bridge a tempting-but-unsafe analogy that should be a false-friend caution rather than an apply-as set?"

Where in taxonomy. New pattern #17: "false-friend evasion via bridge-card route."

Wiki-schema interaction concerns

Concern 1 — Bridge-card sign-off is procedurally a promotion event but the schema does not record it as one

The wiki's status-gate logic (claim-status-gates, motif weight class judgments) is structured: every status change has a documented event. Bridge-card sign-off creates apply-mode authorization but is not currently part of the wiki's status-gate machinery. The Pass 3 report's Approved by maintainer: YYYY-MM-DD line is a sign-off marker but is not exposed to lint or to audit-time review.

Conflict. Without sign-off events appearing in some status-gate machinery, future audits cannot distinguish between (a) apply-mode writes anchored in a signed-off bridge card vs. (b) apply-mode writes that landed without bridge-card sign-off. The wiki loses provenance.

Fix. §15 says "Does not modify CLAUDE.md or weave/SKILL.md... those updates land at the v0d.7 codification step." The v0d.7 codification must add: (i) a wiki/.audit/bridge-cards-applied.md register that lists every bridge card that has been signed off and applied, with apply-date and apply-as scope; (ii) a lint check that flags concept-page subsections whose anchor is "bridge card X" but bridge card X is not in the signed-off register. Without these, the bridge-card layer becomes a black hole for provenance.

Concern 2 — General Rule 17 ("user memories nominate; artifacts authorize") and bridge cards

General Rule 17 is explicit: an audit report itself authorizes a claim only when its content cites traceable anchors. The Pass 3 report (which contains the 5 cards) is an audit report. The cards cite anchors that themselves trace to the silent-keys report and to extraction notes. The chain is: silent-keys-2026-05-07.md → extraction notes → raw passages. This chain is artifact-backed.

Subtlety. Card 2's apply-as item 5 mentions "Possible motifs.md candidate (hauntology-in-MP / phantom register)." If the apply-mode follow-up adds a hauntology motif entry to motifs.md, the evidence chain for that motif entry rests on the bridge card's anchors. But §12 of the plan says sign-off does NOT authorize motif HUB/STRUCTURAL upgrades. The card is fine; the future motif promotion needs its own gate. The plan is OK on this, but the boundary is so subtle that an inattentive apply-mode agent might draft a motifs.md entry at BRIDGE weight, treating the bridge card as authorization. The plan should explicitly say: bridge-card sign-off authorizes apply-as items literally, not downstream motif/claim promotions even when "apply-as item 5" gestures at them. Card 2's apply-as item 5 currently reads "Possible motifs.md candidate" — the apply-mode agent should treat "possible" as "do not write a motifs.md entry," not as "write a motifs.md entry at BRIDGE weight."

Fix. Add §12b: "Apply-as items that flag future promotion (Possible motifs.md candidate, flag for second-source confirmation, claim candidate slug X) authorize the flag itself, not the future promotion. Apply-mode writes implementing such items must write only the flag, not the promotion."

Concern 3 — General Rule 18 (artifact conservatism / retrospective relevance) and the 5-raw-read cap

General Rule 18 says: when a later source turns an older passage into evidence for a new synthetic claim, do a targeted raw-source check of the older passage's local context. The bridge cards are doing exactly this work — using older extraction-note anchors as evidence for new synthetic bridges. §6 of the plan caps raw reads at 5 per card per reviewer, which is fine for verification but may be too tight for the retrospective-relevance check that Rule 18 requires.

Conflict. The cap-of-5 means at most 15 raw reads per card across A, B, C reviewers, which is sufficient. But Rule 18 specifies which raw reads are needed: the older passage's local context. Role C's protocol §5 item 3 ("Verify the anchor says what the card claims it says") gets at this, but is not framed as the Rule 18 check.

Fix. Add to Role C protocol §5: "Item 7: For any anchor that originates in an extraction note ingested before the bridge's primary motivating source, perform a General Rule 18 local-context check: read the cited passage in context (preceding and following paragraphs in the raw file) and verify that the passage carries the cross-source weight the bridge assigns." This is a small textual addition that grounds Role C in the existing Rule 18 discipline.

Concern 4 — Confidence field omitted from card schema

Concept pages and claim entries carry a confidence field. Bridge cards do not (the 10 fields in §1 of the plan do not include confidence). The bridge card's "evidence status" field (extraction-note anchored vs. source-page anchored) is a sourcing fingerprint, not a confidence rating.

Conflict. The wiki's status-gate machinery uses confidence as a calibration signal. Bridge cards have no calibration signal beyond "evidence status." This is fine while bridge cards stay as proposed-status artifacts, but once they authorize apply-mode writes, the apply-mode subsection on (e.g.) [[institution]] will need a confidence value, and there's no obvious value to assign without re-deriving it from scratch.

Fix. Add confidence (medium / low / speculative) as an 11th bridge-card field, set at draft time, re-verified at sign-off. Single-source-blocker cards default to low or speculative. Subsection writes implementing a low-confidence card should annotate the subsection's affected page with a confidence: low increment on the page's frontmatter — or, more practically, an explicit (live claim, see [[claims#X]]) or (bridge card, see [[.audit/...]]#card-1) annotation in the subsection prose.

Concern 5 — The 5-test gate for supported and bridge cards

The carve-out (§12) explicitly says sign-off does NOT authorize promotion to supported. Good. But: subsections that paraphrase a bridge card's "What it makes visible" prose can functionally encode a thesis at supported-level epistemic force, even without a claims.md entry. The wiki has prior incidents on this — my MEMORY.md flags Phase 8 cite-back failure modes including "verify stale statuses." A bridge card that does not exist in claims.md but whose content is supported-force in a subsection is harder to verify and more likely to drift.

Fix. Same as Scenario 5's fix: epistemic-force constraint on subsection prose. Plus: lint check that finds subsection prose with supported-level declarative language anchored only to a bridge card.

Predicted pilot outcomes (per card)

Card 1 — déposer / "deposit a sense"

Role A (gate-applier) verdict: APPROVED, medium confidence.

• 6-test schema gate: all anchors traceable; counterpressure has counter-test ("does déposer operate this way in PoP / Husserl-limits / Inédits I-II?"); apply-as specificity reasonable; no single-source blocker needed (4-chapter convergence); primary home justified; adversarial restatement light is met. The card is structurally clean.

Role B (adversarial-attacker) verdict: DEFECT FOUND, medium-to-high confidence.

• Pattern #1b (scaffolding promotion sub-pattern, if added): the 4-chapter convergence is the M-C 2026 editorial frame's scaffolding, not a stable MP-corpus register. Counterpressure flags this; the counter-test is right, but the bridge already commits to déposer as "the verb-form distinguishing institution from constitution and sedimentation" without running the counter-test. This is the heart of #1b.
• Pattern #15 (claim-status laundering, if added): apply-as item 1 ("subsection on [[institution]]: name déposer as the verb-form distinguishing institution from constitution and sedimentation") encodes a thesis-claim about institution's mechanism. Without the claim-promotion gate, this lands in a HUB concept page at thesis-claim force.
• Pattern #10 (longstanding controversy): the distinction institution-vs-constitution-vs-sedimentation is itself heavily contested in MP scholarship (Lawlor, Toadvine, Saint Aubert). The bridge takes an implicit position on this contestation.

Role C (evidence-tracer) verdict: APPROVED, high confidence.

• All anchors verified PRESENT (silent-keys-2026-05-07.md §"Source 2" Tier 1 PASS row 1; extraction notes for León, Caraus, Mendoza-Canales, Pagan; I&P p. 77).
• Role C's empirical check is satisfied; if Role C is split into C-empirical and C-interpretive (per Scenario 4 fix), C-interpretive might flag Caraus's gloss on line 285-286 as more interpretive than the I&P passage warrants. But as currently specified, Role C passes.

Quorum: 2 APPROVED + 1 DEFECT → adjudicate. Likely adjudication outcome: main-thread reads B's report; B's defects (under current taxonomy) reduce to pattern #1 editorial-frame laundering, which is the same concern the card's counterpressure already names; main-thread likely sign-off proceeds with apply-as item 1 narrowed to Open Questions framing or with explicit "M-C 2026 editorial-register pending PoP / Inédits check" annotation. Confidence in prediction: medium. The adjudication outcome depends on whether the taxonomy includes #1b and #15 — under the current 11-pattern taxonomy, B's defects are weaker and main-thread will likely approve with light narrowing. What would change prediction: (a) adding patterns #1b and #15 to the taxonomy strengthens B's verdict to high confidence and may flip adjudication to BLOCK; (b) Role C split into empirical and interpretive may surface Caraus-gloss concerns that flip C to DEFECT FOUND.

Card 2 — haunting / "obsessive presence" / "haunts"

Role A verdict: UNCLEAR or DEFECT FOUND (low-medium confidence).

• 6-test schema gate has problems. The bridge crosses three MP-periods (PhP 1945, I&P late 1950s, V&I working notes 1959-60). Test 5 (primary home selection justified) is OK. Test 2 (counter-test specific) is OK. But test 1 (anchor traceability) is interesting: anchor "MP PhP 'obsessive presence'" lacks a page or paragraph reference; that's a fabricated-citation-risk pattern (#9) at the margin — not a fabrication, but an underspecified cite.
• Multi-secondary-home (4 secondary homes: passence, depth-of-time, transtemporality, plus the primary institution-of-the-proletariat) creates apply-as scope ambiguity.

Role B verdict: DEFECT FOUND, high confidence.

• Pattern #1 (editorial-frame laundering): cross-source convergence is "secondary-author-mediated" by Caraus and Dufourcq; MP himself does not name "haunting" as a thematic register. Counterpressure flags this; counter-test is appropriate but not yet run.
• Pattern #14 (MP-period miscoding, if added): conflates PhP / I&P / V&I-era registers without explicit periodization.
• Pattern #11 (too-clean-bridge): 4 secondary homes for a single bridge type (modal-temporal) is suspicious — the bridge spans too much without earning each home.

Role C verdict: APPROVED with caveat, medium confidence.

• Anchors traceable; "obsessive presence" in PhP and "haunts" in I&P are MP's own terms; Dufourcq's Schwebung and Lanzirotti's "thickness" are real. But cross-source convergence is interpretive (Caraus's reading), not textual.
• Role C may flag the cross-source-convergence claim as "anchored, but convergence is interpretive scaffolding."

Quorum: Mixed; possibly 2 DEFECT + 1 APPROVED → BLOCK. Likely adjudication outcome: BLOCK or revise-and-resubmit. The MP-period-conflation problem is structurally serious; the multi-period bridge cannot survive at apply-mode without explicit periodization. Confidence in prediction: high (this is the most clearly defective card). What would change prediction: if Role A misses the underspecified PhP cite, and if the taxonomy as currently written (no #14) lets Role B settle for pattern #1 editorial-frame laundering (which the card already acknowledges), then 2 APPROVED + 1 DEFECT is plausible, and adjudication would likely approve with narrowing.

Card 3 — endurance du chaos

Role A verdict: APPROVED, high confidence.

• 6-test schema gate clean. Primary home selection extensively justified (the "Note on Primary home selection" paragraph carefully addresses the silent-keys ambiguity).
• Single-source attestation (SA-2006 only) is acknowledged; no explicit "single-source promotion blocker" but the counterpressure section gestures at it.

Role B verdict: UNCLEAR or DEFECT FOUND (medium confidence).

• Pattern #1: SA-2006 is single secondary source; the endurance register may be SA's archival-philological reading. Counterpressure flags this.
• Pattern #2 (single-source promotion creep): the card claims "single-volume primary attestation" in counterpressure but does NOT carry an explicit single-source promotion blocker like Card 5 does. Apply-as items include "Subsection on [[good-ambiguity]] (under What the Concept Does or new Stakes section)" — Stakes is thesis-force; without a blocker, this is single-source promotion creep.
• Pattern #11 (too-clean-bridge): 3 secondary homes (hyper-dialectic, ineinander, indirect-ontology) for a SA-2006-only attestation is suspicious.

Role C verdict: APPROVED, high confidence.

• SA-2006 Ch IV §1 section title verified; extraction-note primary-concepts list verified; silent-keys verdict verified.

Quorum: 2 APPROVED + 1 DEFECT or UNCLEAR → adjudicate. Likely adjudication outcome: main-thread sign-off with apply-as item 1 narrowed to Open Questions only (not What the Concept Does or Stakes) and with explicit single-source-blocker annotation added. Confidence in prediction: high. What would change prediction: if Role B adopts pattern #2 strictly (single-source blocker required), B flips to DEFECT high confidence and quorum may BLOCK.

Card 4 — ontologie / philosophie militante

Role A verdict: APPROVED, high confidence.

• 6-test schema gate clean. Three SA-2006 raw passage attestations (1947, 1958, post-1958) form a chronological arc; primary home (indirect-ontology) is justified.

Role B verdict: DEFECT FOUND, medium confidence.

• Pattern #1: same as Card 3 — SA-2006 is the single secondary source; the centralization of militante as a stable late-MP register is SA's emphasis.
• Pattern #14 (MP-period miscoding, if added): the 1947 attestation predates MP's late-ontology turn; pairing 1947 with 1958+ requires periodization care. Counterpressure flags "adjacent risk of confusion with philosophie militante in earlier French Marxism (Politzer, Lefebvre)" but does not flag the MP-internal periodization.
• Pattern #15 (claim-status laundering, if added): apply-as item 1 lands on [[indirect-ontology]] Stakes section as a thesis claim about MP's attitudinal mode — thesis-force without claim-promotion gate.

Role C verdict: APPROVED, high confidence.

• All three SA-2006 raw attestations verified. Silent-keys verdict verified.

Quorum: 2 APPROVED + 1 DEFECT → adjudicate. Likely adjudication outcome: sign-off with narrowing — apply-as item 1 downgraded from Stakes to Open Questions or Stakes-with-provisional-framing. Confidence in prediction: medium-high. What would change prediction: if pattern #14 is added and Role B explicitly checks periodization, B's defect strengthens and the 1947 attestation may need to be split out from the 1958+ arc.

Card 5 — "the past that could have been otherwise"

Role A verdict: APPROVED, high confidence.

• 6-test schema gate clean — this card is the cleanest in the batch. Explicit single-source promotion blocker; apply-as scope limited to subsection + open-question writes; counterpressure with second-source-required counter-test; primary home (transtemporality) justified.

Role B verdict: APPROVED, medium confidence.

• The single-source-blocker neutralizes most adversarial patterns. Pattern #2 cleared explicitly. Pattern #11 (too-clean-bridge) marginally — the bridge is structurally simple (one anchor, one modal-hinge claim, narrow apply-as).
• Pattern #10 (longstanding controversy): the modal-vs-epistemological distinction in MP's late ontology has competing readings, but the card declines to adjudicate.

Role C verdict: APPROVED, high confidence.

• Larison extraction-note line 4464 verified. M-C 2026 extraction-note tracker §"trans-temporality" verified. Silent-keys verdict verified.

Quorum: 3 APPROVED → sign off. Likely adjudication outcome: sign-off clean. Confidence in prediction: high. What would change prediction: unlikely to change. The single-source-blocker discipline is what makes this card clean — it's the model card for what the layer can produce.

Batch-level prediction

Same-gap-twice check (§10): pattern #1 (editorial-frame laundering) fires on Cards 1, 2, 4. That's 3 cards failing the same pattern. The procedure pauses the entire batch.

• Main-thread reads aggregate reports.
• Under the current §10 framing, main-thread diagnoses either schema bug or upstream silent-keys bug. Neither is correct: the upstream silent-keys report is sound; the schema is sound; what's actually happening is that the taxonomy is mis-tuned (under-distinguishing scaffolding-promotion from lexical-laundering, per Scenario 2 fix).
• Likely outcome: pilot reveals taxonomy needs refinement (#1 → #1a + #1b split, plus #14, #15 added). Pilot does NOT codify cleanly; revision required before re-pilot.

Aggregate prediction: 1 card (Card 5) signs off clean. 4 cards adjudicated; of those, 2-3 sign-off-with-narrowing and 1-2 BLOCK or revise-and-resubmit. Pilot reveals at least 2 taxonomy gaps. Pilot SUCCESS by the criteria I suggest in Scenario 7: borderline. Codification likely delayed by one revision cycle.

Where math-olympiad analogy breaks

Break 1 — Math has crisp truth conditions; bridge cards do not

In math-olympiad, a proof is correct or it's not. HOLE FOUND = there is a specific step the proof does not justify. The verifier's job is to find the hole; the hole is an objective property of the proof relative to the theorem's truth conditions.

Bridge cards are interpretive philosophical claims. There is no analogue of "the truth conditions for déposer as institution's mechanism." There is more or less philosophically defensible; there is more or less philologically warranted; there is more or less consistent with MP's published corpus. None of these have HOLDS/HOLE-FOUND truth conditions.

Risk. The procedure treats DEFECT FOUND as analogous to HOLE FOUND, but DEFECT is calibrated against a softer target. "APPROVED" in math means "the verifier found no specific gap"; "APPROVED" for a bridge card means "the reviewer judged the bridge to be defensible." The first is approximately a property of the proof; the second is approximately a property of the reviewer. This invites a different failure mode: confident-approval-of-the-wrong-bridge, because the reviewer's calibration is the failure surface.

Fix. Calibrate verdict semantics explicitly. APPROVED for bridge cards means "I (reviewer) judge this bridge defensible under the named tests; my judgment may be wrong, and the bridge's content may turn out poorly under future evidence." DEFECT FOUND means "I have identified a specific structural feature of this card that points to a calibratable concern under named patterns." The procedure should require, for any APPROVED with high confidence, a sentence explaining why the reviewer's confidence is high — what evidence specifically warrants high confidence vs. medium.

Break 2 — Math problems are bounded; bridge-card stakes are open-ended

A math-olympiad proof's stakes are local: the proof is right or it isn't. A bridge card's stakes are open-ended: a poorly-vetted bridge plants vocabulary in a HUB concept page that will be cited from claims, paper drafts, and downstream concept pages. The cost of a wrong-but-approved math proof is one wrong answer. The cost of a wrong-but-approved bridge card is corpus-level contamination.

Risk. The plan treats sign-off as a local decision (per-card quorum vote). The downstream consequences of an apply-mode write on [[institution]] are not local.

Fix. Weight HUB-page apply-as writes more heavily than non-HUB writes. Cards whose primary or secondary home is a HUB page (per motifs.md) should require at least one DEFECT FOUND verdict overturned at adjudication before sign-off proceeds — i.e., raise the bar for HUB-page-affecting cards. Cards 1, 2, 5 all touch HUB-weight concept pages (institution, transtemporality); apply higher scrutiny.

Break 3 — Math has training-set transfer; bridge-card patterns are corpus-specific

The math-olympiad failure modes (#4 famous theorem specialization, #40 too-clean-lemma) are general patterns that transfer across mathematical domains because mathematics has shared structure. The bridge-card failure modes are MP-corpus-specific: philological inversion, MP-period miscoding, secondary-author scaffolding. The taxonomy needs to be MP-tuned; future corpus extensions (Spinoza, Hegel) would need different taxonomies.

Risk. The plan presents the 11-pattern taxonomy as if it were the analogue of math-olympiad's 12 verifier patterns. It's not. The 11 patterns are early-stage corpus-tuned and missing several major corpus-specific patterns (Scenarios A-F above).

Fix. Acknowledge the taxonomy is corpus-tuned and version it. The schema-changelog entry for v0d.7 should include a per-pattern attribution: which patterns are general (transfer to any wiki) vs. which are MP-specific. When the wiki extends to another author/corpus, the taxonomy will need a fresh tuning pass.

Break 4 — Math abstention is calibrated against historical hit rate; bridge-card abstention has no baseline

Math-olympiad calibrated abstention is grounded in a known failure rate: when the procedure says "solved," it's right at some measurable frequency. Bridge-card sign-off has no such baseline. The pilot is N=5 cards; even after the pilot, we don't know how to calibrate "sign-off with high confidence" against historical bridge-card outcomes.

Risk. "Confidence" labels in the verdict format (§8) are uncalibrated.

Fix. Establish a baseline by retrospective audit. After 3-6 months of bridge-card-applied writes, audit a sample for "did the apply-mode writes hold up under subsequent ingest and audit?" Use that to calibrate confidence labels.

Break 5 — Math verifiers don't have to know mathematics's "literature"; bridge-card verifiers do

A math verifier's job is to attack a specific proof; mathematics's larger literature is irrelevant. A bridge-card verifier needs to know MP scholarship's competing readings to apply pattern #10 (longstanding controversy) correctly. This is much harder to embed in a subagent prompt — the subagent does not have a Lawlor-Dillon-Barbaras-Saint Aubert-Toadvine context.

Risk. Pattern #10 is under-applied because subagents don't carry the scholarly context.

Fix. Pattern #10 should be applied at the adjudication step (main-thread Claude does carry some MP-scholarly context), not at the subagent step. Move it from Role B's taxonomy to the adjudicator's pre-quorum scan.

Proposed improvements

Improvement 1 — Adjudicator isolation

What. Run adjudication via a separate fresh-context subagent that receives only (3 reports + card text + schema gate + failure-mode taxonomy + dependency map). Main-thread then ratifies subagent-adjudication = APPROVED outcomes; main-thread escalates BLOCK and UNCLEAR to user.

Why. Main-thread drafted the cards in Pass 3 and is structurally biased toward approval. Fresh-context adjudication preserves the dual-isolation discipline that makes the rest of the procedure work.

Where. Replace §11 ("Main-thread Claude is the adjudicator") with the subagent-mediated workflow. Main-thread remains the coordinator (launching subagents, collecting outputs, writing adjudication.md) but the decision is delegated.

Improvement 2 — Expand the failure-mode taxonomy

What. Add patterns #12 (saturation-bridge), #13 (philological inversion), #14 (MP-period miscoding), #15 (claim-status laundering), #16 (terminology collision), #17 (false-friend evasion). Split #1 into #1a (lexical laundering) + #1b (scaffolding promotion).

Why. Several corpus-specific failure modes are folded under #1 (editorial-frame laundering), which under-detects them. Adversarial Card X (constructed above) demonstrates a pattern not covered.

Where. §4 (the adversarial taxonomy). Renumber after additions; preserve stable numbering by appending (not renumbering existing).

Improvement 3 — Pilot success metric

What. Define pilot success as: (i) ≥3 of 5 cards reach verdict without user escalation; (ii) no same-gap-twice firing requires schema revision; (iii) no card requires more than 1 revise cycle; (iv) user reviewing adjudication.md does not override decisions. Pilot failure is any of the above. Mixed cases codify with caveats.

Why. §13 names the categories of outcome but provides no threshold. Without a metric, "fail" and "succeed" are post-hoc.

Where. §13.

Improvement 4 — Hypothesis-note convergence field

What. Add an explicit "What changed between initial reading and convergence" subsection to the hypothesis-note template. If nothing changed, the reviewer must say so and explain why the anchor checks did not move the impression.

Why. Without this, the hypothesis note risks becoming post-hoc rationalization (Scenario 6).

Where. §7 hypothesis-note template.

Improvement 5 — Carve-out tightening (epistemic-force constraint)

What. Add §12a: subsections written by apply-mode under single-source-blocker cards may only appear under Open Questions (not What the Concept Does or Stakes), and must use provisional framing. Single-source-but-no-blocker cards (Cards 3, 4) should be retroactively assigned blockers before sign-off if Role B's analysis surfaces single-source concerns.

Why. The carve-out boundary (§12) authorizes subsection writes but does not constrain their epistemic force. Without the constraint, single-source theses land in HUB concept pages at thesis-claim force, bypassing the claim-promotion gate.

Where. §12.

Improvement 6 — Post-adjudication dependency reconciliation

What. After all per-card adjudications, main-thread re-reads any apply-as items that cross-reference other cards; if a referenced card was BLOCKED, the cross-referencing item is narrowed or struck.

Why. Card cross-dependencies are flagged but not reconciled (Scenario 3).

Where. New §11a between §11 and §12.

Improvement 7 — Apply-as-flag vs. apply-as-write distinction

What. Add §12b: apply-as items that flag future promotion (Possible motifs.md candidate, flag for second-source confirmation) authorize only the flag, not the future promotion. Apply-mode agents implementing such items write only the flag.

Why. Card 2's apply-as item 5 contains the word "Possible motifs.md candidate" which an inattentive apply-mode agent could read as authorization to write a motifs.md entry at BRIDGE weight.

Where. §12.

Improvement 8 — Bridge-cards-applied register

What. Add a wiki/.audit/bridge-cards-applied.md register that lists every bridge card that has been signed off and applied. Add a lint check that flags concept-page subsections whose anchor is "bridge card X" but bridge card X is not in the register.

Why. Bridge-card sign-off creates apply-mode authorization but is not currently exposed to lint or audit-time provenance checks (Concern 1).

Where. v0d.7 codification step (the schema bump itself, not the plan); the plan should anticipate it in §15.

Improvement 9 — Confidence field on bridge cards

What. Add confidence (medium / low / speculative) as an 11th bridge-card field, set at draft time and re-verified at sign-off. Single-source-blocker cards default to low or speculative. Subsection writes implementing low-confidence cards must annotate the subsection prose with provisional framing.

Why. Bridge cards have no calibration signal; subsection writes have no obvious confidence value to inherit (Concern 4).

Where. v0d.5 bridge-card template extension; the plan should authorize this in §15 as part of the v0d.7 update.

Improvement 10 — Local-context check inheritance from Rule 18

What. Add to Role C protocol §5: Item 7 — for any anchor that originates in an extraction note ingested before the bridge's primary motivating source, perform a General Rule 18 local-context check (read the cited passage in raw, with paragraph context, and verify the cross-source weight the bridge assigns).

Why. Rule 18 mandates this check; bridge cards systematically reuse older extraction-note anchors but the plan's Role C protocol does not explicitly invoke Rule 18.

Where. §5.

Improvement 11 — Pattern #10 moved to adjudication

What. Move pattern #10 (longstanding controversy) from Role B's subagent taxonomy to the adjudicator's pre-quorum scan. Reason: subagents do not carry the MP-scholarly context to apply #10 correctly.

Why. Math-analogy break 5.

Where. §4 (remove from Role B) and §11 (add to adjudication pre-quorum scan).

Improvement 12 — HUB-page apply-as escalation

What. Cards whose primary or secondary home is a HUB-weight concept page (per motifs.md) require at least one DEFECT FOUND verdict to be overturned at adjudication, OR the apply-as item targeting the HUB page must be downgraded to Open Questions.

Why. HUB pages have downstream impact disproportionate to the card's local stakes (math-analogy break 2). Cards 1, 2, 5 are at risk in this batch.

Where. §11 or §12.

Verdict

ACCEPT-WITH-CHANGES.

The plan's architecture is sound: dual context isolation, asymmetric quorum, hypothesis-note discipline, pattern-armed adversarial review. These are the right structural choices for adapting math-olympiad to bridge-card sign-off, and the plan articulates them clearly.

What the plan needs before v0d.7 codification:

1. Critical (must-fix before pilot): adjudicator isolation (Improvement 1); pilot success metric (Improvement 3); apply-as-flag vs. apply-as-write distinction (Improvement 7); HUB-page apply-as escalation (Improvement 12).

2. High (should-fix before pilot or as outcome of pilot): expand failure-mode taxonomy with #1b, #12, #13, #14, #15, #16, #17 (Improvement 2); carve-out epistemic-force constraint (Improvement 5); confidence field (Improvement 9); Role C Rule 18 inheritance (Improvement 10).

3. Medium (can land at codification): hypothesis-note convergence field (Improvement 4); post-adjudication dependency reconciliation (Improvement 6); bridge-cards-applied register (Improvement 8); pattern #10 to adjudication (Improvement 11).

The pilot, run as currently specified, will likely reveal taxonomy gaps that justify Improvement 2 retrospectively. That's an acceptable workflow if the pilot is treated as a calibration step rather than a green-light. If the pilot is treated as authorization for v0d.7 codification without the taxonomy revisions, the bridge-card layer will ship with a known under-detection problem on the highest-frequency failure mode (#1 editorial-frame laundering and its scaffolding-promotion subtype).

The core philosophical risk — that math-olympiad's crisp truth conditions don't transfer to interpretive philosophical claims — is structural and cannot be fully eliminated by procedural fixes. The procedure should be ambitious about reducing it, but humble about not eliminating it: bridge-card sign-off is a calibrated probability of philosophical defensibility, not a proof of correctness. The schema-changelog entry for v0d.7 should say so explicitly.