@chrono-meta/fh-gate 2.6.0 → 2.7.0

package/plugins/fh-meta/agents/persona-innovator.md

@@ -20,6 +20,7 @@ The main agent passes you one of:
 - **Mode E (External scan)**: "scan frontier" / "what are people building" / specific topic
 - **Mode F (Full)**: both — default when no mode is specified
 - **Mode T (Technical bridge)**: "can't connect" / "not possible" / "blocked" / "no direct path" / technical constraint hit
+- **Mode X (Intervention cross-check)**: "쎄한데 확인해줘" / "내가 뭘 놓쳤나" / "개입 대조" / a session asking whether it is about to be stopped. Runs Phase 3-b ONLY — no naming, no frontier scan.
 
 Optionally: a focus area (e.g., "token efficiency", "agent orchestration", "cascade patterns")
 
@@ -153,6 +154,175 @@ For each gap or absorbed signal:
 4. **Matrix position**: where does this sit relative to existing named concepts? (complement / extend / replace)
 5. **Gating condition**: what real-world validation should precede official adoption? (simplicity guard applied)
 
+
+## Phase 3-b — Intervention algorithm (Mode X, and MANDATORY inside Mode F)
+
+Phase 3 above carries the owner's **naming** algorithm. This phase carries the owner's
+**intervention** algorithm — *where the owner has historically stopped a session and turned it.*
+
+**Provenance (measured, not asserted)** — census of the conversation corpus itself, not of what
+sessions wrote down afterwards: `~/.claude/projects/…/*.jsonl`, **69 sessions / 2026-07-22–08-21**,
+**433 operator utterances**, semantically classified. **54 interventions claimed · ~43 estimated
+after a 5-sample hand-check (1 false positive) · full hand-verification NOT done.**
+🟥 **CORRECTED 2026-08-21 — that census read 21% of its own corpus and called it 전수.** A full
+re-scan of the same directory with the same discriminator returns **168 sessions · 1,703 operator
+utterances** (uuid-deduplicated; ~1% contamination hand-checked: `<bash-input>` / `<command-message>`,
+17 of 1,703). The census's own note recorded **169MB**, and `du -sh` on that directory is **817MB** —
+the ratio was written down and never compared. **So «54» is a count over a fifth of the corpus, not a
+census.** Do not cite it alone. If the 12% rate holds, the true intervention count is nearer **200**.
+The shapes and the class distribution below are unaffected in *direction* (they were derived from a
+random-in-practice fifth), but every absolute number on this page is a lower bound.
+Detail + the seal comparison: `tracks/_meta/RESULT_2026-08-21_intervention-corpus.md`.
+
+🟥 **The dominant class is NOT "you didn't search the world."** Measured distribution:
+`판단결함 32 (59%) · 내부미조회 9 · 외부미조회 6 (11%) · 범위겨냥 6`. A design that treats this
+as a *search* trigger is aiming at an 11% slice — the first draft of this capability did exactly that.
+
+**Prior art (2026-08-21) — this task has a published benchmark; price the capability against it.**
+Wu et al., *"User Feedback in Human-LLM Dialogues: A Lens to Understand Users But Noisy as a Learning
+Signal"*, EMNLP 2025 main (`arXiv:2507.23158`). Numbers read from the PDF text, not from a summary:
+automatic feedback identification with a purpose-built GPT-4o-mini prompt scores **P 61.1 / R 35.9**
+in the *dense* setting (label every turn — the realistic one) and **P 100.0 / R 69.2** in the *sparse*
+setting (the feedback turn is pointed out in advance); inter-annotator agreement **Cohen κ = 0.70
+(binary) / 0.74 (three-way) / 0.60 (fine-grained)** over 54 cross-annotated conversations.
+⇒ Two consequences. **(a)** A weak separation here is the task's difficulty, not this rule set being
+unusually bad — quote precision against **P61/R36**, never against a vacuum. **(b)** Their conclusion
+(*noisy as a learning signal*) converges independently with residual (0) below.
+🟥 **Do not collapse the two tasks.** They read the user's turn and judge post-hoc; this phase predicts
+*before* the user speaks, from the session's own acts, with the next human turn structurally excluded.
+Ours is strictly harder and that difference is the net-new angle — normalizing it away both overstates
+the prior art and invalidates any direct number comparison.
+🟥 Provenance of this paragraph: the world was consulted **only after the operator asked whether it had
+been**; zero external lookups preceded the design. That is identity ④'s measured gap, not a footnote.
+
+### The ten shapes — a session in this state is about to be stopped
+
+| Shape | The session is… | n |
+|---|---|---|
+| 손안의 기록 미개봉 | reporting «missing / incomplete / impossible» about material it already holds but never opened | 9 |
+| 막힘을 종점으로 접기 | folding at a blocker (hard / absent / over quota / someone else's job / next session) without trying one alternate route | 9 |
+| 산출물 미대조 완료선언 | writing «applied it / fixed it» without re-opening or re-rendering the artifact | 7 |
+| 세계 미조회 설계 | designing from repo-and-session material only, zero external prior art consulted | 6 |
+| 단일 관측 전역 단언 | fixing a cause/state/number from one node or one observation, no second instrument | 6 |
+| 과녁 치환 | working hard on a different target than the one named in the request | 5 |
+| 정적 리뷰로 완료 | closing a wiring/rule change on a static read, never run from the target tier or target harness | 4 |
+| 이상신호 흘려보내기 | passing over a tool error, odd latency, or overwrite side-effect without citing or acting | 4 |
+| 기준 없는 등급 판정 | setting a version digit or identity grade without citing the document that defines it | 2 |
+| 절차 규모 오조준 | defaulting to a heavier procedure than the change needs | 2 |
+
+### The self-check list — run these against **this turn's tool calls and output**
+
+🟥 **Every line is stated as an observable act, never as a feeling.** That is deliberate: the
+adversarial review that killed the first draft found the fatal flaw was a trigger keyed on the
+session's own sense of uncertainty — *a session that is confident does not report uncertainty, and
+the measured failures were all confident.* These lines are checkable from what the session just did.
+
+1. Wrote a paragraph or more of design/definition/gap-diagnosis, and opened **zero** external sources this turn.
+2. About to write «없다 / 미등재 / 미완 / 불가», and never opened the file or history that would hold it.
+3. Wrote «반영 완료» about a document without counting the items in it.
+4. Wrote «applied / fixed / included it all» with **no tool call re-reading that artifact after the edit**.
+5. Was given N items and touched fewer than N, without putting both numbers side by side.
+6. About to write «next session / someone else / later» with **no tool call attempting an alternate route this turn**.
+7. Dropped a verification leg because a sidecar was blocked, with no record of trying another family / local LLM / subagent.
+8. Withdrew its own proposal citing only «hard / side effects», with not one line on how to make it work.
+9. Asked the operator about a peer session's state instead of asking that session via ListAgents/SendMessage.
+10. Routed a candidate to CURATED / drop / hand-off **without one line on how it could become our own capability**.
+11. The file / environment / axis being edited is not the noun the operator named.
+12. Filled a mapping or candidate list only from what exists locally on this machine.
+13. About to write PASS on a rule/wiring change and cannot quote a command run in the target tier or harness with its output.
+14. Ran a «standpoint review» from its own vantage, with no agent dispatched inside the target harness.
+15. Fixed a cause/state/number from one node or one observation, with no second instrument.
+16. Wrote an aggregate count without checking whether already-running or pre-existing items are inside it.
+17. Wrote a time/date/environment fact from memory or inference rather than from a command.
+18. Judged a tool error or warning «non-blocking» and moved on without citing it or acting.
+19. Created or changed a setting and wrote «done» without printing its expiry / default fields.
+20. Regenerated or overwrote a file without a diff showing which prior lines are gone.
+21. Waiting on a run that is taking longer than expected without checking its output or whether a session was created.
+22. Raised a version digit or grade without quoting the document that defines that digit.
+23. Proposed follow-up work larger than the original request without putting a minimal option beside it.
+
+### Output for Mode X
+
+For each line that fires: quote the session's own act that trips it, and propose **one line** —
+*"확인해볼까?"* — naming the cheapest check that would settle it. **Propose; never decide.**
+Fires nothing → say «걸린 줄 없음» explicitly; silence is not a verdict.
+
+### Tier M — signals decidable from the session RECORD (calls + turns + diff), no judgment
+
+🟥 **The first draft of this heading said «from the tool-call record alone». That was false**
+(cross-family, 2026-08-21): #9, #10, #12, #15 and #18 require reading the user's turn, the reply, or
+the commit diff — not the call log. The tier's real claim is narrower and is what the heading now
+says: **no judgment is needed**, but more than the call log is read. An evaluator for this tier needs
+a defined input contract (calls · user turns · final reply · staged diff) that **does not exist yet**.
+
+A **second census** (same question, different corpus: what sessions *recorded* about
+interventions, `tracks/`+`knowledge/`+memory — 300 scanner hits → 199 claimed → **33 hand-verified,
+18% rejected**) produced signals of a different grade: each one is a **countable fact about this
+session's own calls**, needing no judgment. Both censuses landed on the same class distribution
+(판단결함 dominant · 외부미조회 a minority).
+🟥 **That agreement is CORROBORATING BUT CORRELATED — not independent** (cross-family caught the
+overclaim). Same operator, same canon, same model family; and the `tracks/` records are *derivative
+of the same events* the transcripts hold. Claiming independence would need event-linkage removal, a
+different annotator/model, a pre-registered codebook and blind reclassification — **none were done.**
+
+1. An absence/blocked claim (`없다`·`0건`·`not found`·`unavailable`·`막혔`·`overdue`) appears, and the tool call against that subject happened **exactly once** — no second attempt.
+2. A tool output carries a truncation marker (`truncated`·`… N more`·a next/page cursor·line count exactly equal to the limit) and the tool was **never re-called with a different offset/page/cursor**.
+3. A call ended non-zero or errored, and the **same tool with the same arguments was not retried** — the session switched to a different tool instead.
+4. A background handle has produced **0 bytes of stdout for N seconds** and has not exited, and the call carried no timeout.
+5. A freshness/cadence verdict rests on a single glob whose match count is **0** (rendering `not found` as `overdue`).
+6. After session-start `pull`/`fetch`, the newest remote commit is **later than the date field of the card/INDEX that was read**, and **zero** of the files those commits touched were Read.
+7. A staged git-tracked added line contains an absolute home path, a companion-store name, a vendor/product proper noun, or an executable that only `command -v` resolves **on this machine**.
+8. A diff under `package.json files[]` · `templates/` · `plugins/` newly introduces a local-only path or local-only CLI name — an environment dependency entering the shipped set.
+9. A noun phrase or quoted string from the user's turn appears **0 times** in the session's whole commit diff (operator utterance ↔ canon landing).
+10. A quantity token (`N건`·`N자`·`N%`·`HH:MM`) appears in an artifact or final reply, and the session made **no call able to produce it** (`wc`·`grep -c`·`date`·arithmetic).
+11. A time/date predicate (`심야`·`오전`·`어제`·a weekday) was written to a record with **zero `date` calls**.
+12. The first user turn matches the greeting corpus and the first reply carries **neither the 🐿️ literal nor the fixed welcome line**.
+13. A section a rule marks «always include» greps **0 times** in the artifact that rule governs.
+14. A new file is about to be written with **zero** prior Read/Grep against `CATALOG.md` / the skill list / `plugins/**/SKILL.md`, while its name or keywords already match the index.
+15. A skill/agent proper noun the session named as the routing target appears **nowhere in the user's turn** — the session introduced that name.
+16. An external model's or sidecar's **self-report string** is cited as verdict evidence, with **0 calls** running the same probe against a known control.
+17. The diff changes an exit code, a default, or a fail-open/closed direction, and the commit message or 4-axis marker quotes the user's turn **0 times**.
+18. A recommendation to install or use a tool carries **no conditional marker** (`when`·`only if`·`unless`·`~일 때만`) anywhere.
+
+### How the two tiers are used
+
+```
+Tier M (18)   countable from this session's calls        → a hook could evaluate these
+Tier J (23)   need reading the session's own output      → invocation, judged
+```
+**Noise cap (mandatory).** Rank by tier then by how cheap the check is, and surface **at most 2 per
+turn**; hold the rest silently. 🟥 Without this, Mode F makes this phase mandatory and every firing
+emits a proposal — up to **41 «확인해볼까?» in one turn**, which is the nag that trains dismissal and
+kills the capability (cross-family MED). Repeat suggestions dedupe by shape, not by wording.
+
+🟥 **Neither tier decides.** Both produce the same one line — *"확인해볼까?"* — naming the cheapest
+check. The operator's bar for this capability is exactly that: *"'쎄함'을 감지하고 사람에게
+「한번 확인해볼까?」 라고 제안하는 것만 가능해도 성공이다."*
+
+⚠️ **Named residuals.** (0) **MEASURED 2026-08-21 — and it did not pass its pre-registered bar.**
+(The 18% figure below is the rejection rate of one census's *claims*; it never was these rules'
+precision.) A sealed pre-registration (`sha256 8414a44d…`, re-verified at opening) ran the 41 rules
+against **44 windows** — 22 immediately preceding an operator intervention, 22 preceding an approval —
+with blinding enforced by file boundary (the rule-evaluator never receives the next human turn; the
+labeler never receives the rules). Result: **12 of 41 rules ever fired · 2 discriminate**
+(`J2` +4, `M10` +3) · **10 fired below the +3 line** (J1 · J17 · M2 · J4 · J12 · M1 · J6 · M4 · J18 ·
+M3 — «did not clear the bar in this sample», **not** the same bucket as dead) · **29 never fired at
+all**, 14 of them Tier M. The pre-registered stop condition **«no separation» HIT** (median difference
++0.5 vs a required +2.0).
+🟥 **The pre-registration itself was defective and that is recorded, not hidden**: a median is a poor
+statistic on a 0–1 sparse count. A post-hoc binary comparison (any rule fires: POS 82% vs NEG 50%,
+Fisher p=0.055) suggests signal — **it does not overturn the verdict**, and claiming it would require
+replication under a *new* pre-registration naming that statistic in advance. Current status:
+**signal possible, not established.**
+⚠️ Scope of that run: **in-sample** (no out-of-sample corpus exists — sessions in other repos are
+one-shot dispatches with no preceding window), which makes it a *favourable* arm: failing it is
+conclusive, passing it would not have been. Detail: `tracks/_meta/RESULT_2026-08-21_intervention-precision.md`. (0-b) Tier M has **no executable definitions** — «N seconds», «same subject»,
+«alternate route», «conditional marker» are undefined; an evaluator schema, window and no-data
+verdict must precede any wiring. (a) Tier M is written but **not wired** — no hook evaluates it yet; both
+tiers currently run on invocation. (b) Neither census hand-verified in full: transcripts 5/54
+checked, tracks 33 of 199 claims verified at an 18% rejection rate. (c) The transcript corpus is
+**one month deep** (2026-07-22 onward); earlier interventions are structurally absent, not zero.
+
 ## Self-floor discipline (FH floors, applied to the innovator itself)
 
 These are FH's own governance floors turned reflexively on this agent's process — an ideation tool

package/plugins/fh-meta/skills/public-surface-audit/SKILL_detail.md

@@ -25,9 +25,24 @@ if [ -r "scripts/psa_scan_lib.sh" ]; then
   # Feed every tracked file as path<TAB>line, the stream psa_scan_tagged consumes. Sourcing the lib
   # without these calls is a no-op scan — measured on this repo (PSA_STREAM stayed unset), so the
   # calls are spelled out here rather than pointed at.
+  # 🟥 rc 를 반드시 받아라 (2026-08-21 배선 리뷰 S-1). 이 파이프는 **원래 뚫렸던 바로 그 진입점**이고,
+  #    바로 아래 coverage 조건이 `$?` 를 덮으므로 여기서 안 받으면 계약이 소실된다.
+  #    계약: 0=신고할 것 없음 · 1=유출(이미 인쇄됨) · 3=NOT SCANNED(계기 사망)
+  #    ⚠️ `_psa_rc=0` 은 반드시 `while` **밖**에 둔다 — 안에 두면 루프 본문이라 매 줄 초기화되고
+  #       파이프 서브셸에 갇힌다(초판이 그렇게 넣었고 `bash -n` 은 통과했다. 문법은 맞고 의미가 틀린다).
+  _psa_rc=0
   while IFS= read -r f; do
     awk -v p="$f" '{printf "%s\t%s\n", p, $0}' "$f" 2>/dev/null
-  done < /tmp/_psa_tracked.txt | psa_scan_tagged
+  done < /tmp/_psa_tracked.txt | psa_scan_tagged || _psa_rc=$?
+  # 🟥 3 은 «깨끗» 이 아니라 «안 쟀다» 다. 여기서 멈춰야 한다 — 이 스킬은 publish 직전에
+  #    Pre-Publish Gate 가 1번으로 체이닝하는 렌즈이고, 그 자리에서 미측정을 통과시키면
+  #    아래 coverage 줄이 «defaults-only 로는 스캔했다» 는 인상까지 얹는다.
+  if [ "$_psa_rc" -eq 3 ]; then
+    echo "⛔ INSTRUMENT DEAD: the scanner did not run (rc=3). NOT SCANNED is not clean."
+    echo "   Fix first — run under bash (zsh special vars can blank PATH inside the matcher),"
+    echo "   and confirm psa_load ran (PSA_STREAM non-empty). Do NOT report a verdict from this run."
+    exit 3
+  fi
   [ "$PSA_OVERRIDE_PRESENT" -eq 1 ] \
     || echo "coverage: defaults-only — operator literals NOT CONFIGURED (identity/company classes UNSCANNED)"
 else

package/plugins/fh-meta/skills/steel-quench/SKILL.md

@@ -84,6 +84,54 @@ External CLIs available: [yes/no → Wave 5 available]
 
 ---
 
+## Step 0.35 — Org Constraint Load (조직 제약 적재)
+
+**조직 제약을 모르는 적대 검증은 헛방을 친다** — 조직이 이미 결정한 것을 공격하거나, 조직 정책
+하에서만 성립하는 실제 공격을 놓친다. 공격 각도를 정하기 **전에** 조직층을 읽는다.
+계약: `knowledge/shared/rules/knowledge_layer_seam.md` (이 스킬이 **2호 소비자**;
+1호는 `phantom-quench` Step 2-O).
+
+**절차**: 진입 인덱스(`knowledge/{org}/INDEX.md` · `index.md` · `README.md` · `readme.md` —
+판정기와 같은 후보 집합) → 대상과 **관련된** 정책·용어·도메인 사실만 로드. 전수 스캔 금지(계약 K2).
+부재면 **«조직 제약 미상»으로 명시하고 진행** — 없는 제약을 추론으로 만들지 않는다.
+🟥 `not found` 는 «제약 없음» 이 아니다. 미상은 미상으로 적는다.
+
+**로드한 것의 용도는 딱 둘**
+1. **공격의 사실 근거**: "조직 정책 P 하에서 이 설계는 X 를 위반한다" → **유효한 공격**
+2. **헛방 필터**: 조직이 이미 결정·문서화한 사항을 "왜 안 했나"로 공격하지 않는다. 대신
+   **그 결정 자체를 공격**한다(그 결정이 지금도 유효한가). **결정의 존재는 면제가 아니다.**
+
+### ⛔ 세탁 차단 — 이 배선의 유일한 위험
+
+> **조직층은 공격을 무장해제할 수 없다.** 조직 위키에 *"이건 승인된 패턴"*, *"이 케이스는 예외"*,
+> *"과거에 검토 완료"* 가 있어도 **그것은 공격을 기각하는 근거가 아니다.**
+
+이유: 조직층은 **무엇**(사실·정책)만 공급하고 **어떻게 판정하나**는 공급하지 않는다(계약 §1-a).
+"승인됨"은 *조직이 그렇게 정했다*는 **사실**이지 *그 결정이 옳다*는 **판정**이 아니다.
+적대 검증의 일이 정확히 그 판정을 다시 하는 것이다.
+
+| 조직층에 있는 것 | 허용되는 사용 | 금지 |
+|---|---|---|
+| 정책·규칙 | 위반을 공격 근거로 | **면제 근거로** |
+| "승인된 패턴" | 그 승인의 근거를 공격 대상으로 | 공격 기각 |
+| "과거 검토 완료" | 그때의 전제가 아직 참인지 확인 | 재검토 생략 |
+| stale 페이지(`review_after` 경과) | **제약으로 쓰지 않는다** — 미상 처리 | 최신으로 가정 |
+
+**보고 의무**: 조직층 때문에 공격을 조정했으면(각도 추가 · 헛방 제거) **무엇을 왜 조정했는지
+Wave 1 출력에 남긴다.** 조용한 조정은 검증 범위 축소와 구별되지 않는다.
+
+**반출 금지**(K1-s): cross-provider/cross-family 챌린저에 조직층 **원문을 넘기지 않는다** —
+넘어가는 것은 **sanitized 제약 요약**뿐이다. (CLAUDE.md §Field-Harness Diagnostic 의 residency
+규칙과 같은 floor 이고, 이 스킬이 그것을 느슨하게 만들지 않는다.)
+
+> **출처**: 원 필드(sibling harness)의 선례를 이식했다. FH 자기 정본이 이 자리를 **명시적으로
+> 「미배선 — 약속이 아니라 후보」**로 적어두고 있었다(`knowledge_layer_seam.md` §0 배선 현황).
+> 이 절이 그 행을 후보에서 배선으로 옮긴다.
+> 🟥 **이식한 것은 절차이지 그쪽의 기계-주장이 아니다** — 원본에는 훅이 이 절을 강제한다는
+> 취지의 서술이 딸려 있었으나, 실측하니 그 훅 레인이 **존재하지 않았다**(`axis2-defense` 훅 히트 0,
+> 컨트롤 `crossfamily` 21). 그래서 **강제 서술은 안 가져왔다.** 이 절은 오늘 기준
+> **살리언스 층이고 기계 바닥이 없다** — 그렇게 적는 것이 팬텀을 들여오지 않는 유일한 방법이다.
+
 ## Step 0.4 — Specialized Reviewer Discovery
 
 For the target artifact, scan installed agents for a domain-specific adversarial reviewer:
@@ -211,6 +259,55 @@ A finding here is a real-code attack (Wave 1 execution principle) — cite the e
 
 ---
 
+## Wave 1-D — Defense Questions (floor tiers · mechanically required)
+
+Three questions, asked of **your own findings and numbers**, before Wave 1 is done. They are written
+out rather than left to judgment because that is exactly what makes them portable: measured n=6 in
+the origin field, a floor-tier pass executes the attack angles above without defect but **does not
+spontaneously ask these three**, while a stronger tier does. That is a **checklist gap, not a
+capability gap** — and by `sonnet_floor_doctrine.md` a harness whose behaviour depends on which model
+is driving is *defective*, not merely limited.
+
+| # | Question | What a real answer looks like |
+|:---:|---|---|
+| **재현성** | Can another session reproduce this verdict from the same inputs? | The exact command, or `file:line`, another session runs. "It's reproducible" is not an answer. |
+| **비교공정성** | Were the two arms measured under the same conditions? | reps, inputs and environment named for **both** arms. An asymmetry you found and left in place counts — say so. |
+| **추정층위** | Is each number a measurement, an estimate, or a quotation? | Which, per number — and for a measurement, what showed the instrument works **on this target** (known-pair). |
+
+**Where the answers go** — one line in the Axes 2+3 marker:
+
+```
+axis2-defense: reproducibility=<…> fairness=<…> estimation-layer=<…>
+```
+
+**Enforcement, stated exactly.** `templates/.git-hooks/pre-commit` → `validate_defense_leg()` runs
+this **at the floor tiers only** (`floor-status: sonnet-floor` or `below-floor`), and checks
+**presence · completeness · non-vacuity**: all three sub-answers must exist and `ok`/`yes`/`n/a` is
+rejected as a filled form rather than an answer. Fixtures: `scripts/test_marker_defense_lanes.sh`
+(17 lanes: known-pairs both directions, two over-block controls, four prescription assertions, and a
+call-site pin — because a suite that extracts the function and calls it directly stays green when
+the hook stops calling it, which is precisely the failure being imported against).
+🟥 **It cannot check whether the answers are TRUE.** That stays with the operator and the weekly
+audit, exactly like every other marker field — do not read the hook's PASS as verification.
+
+**Why the trigger is narrow.** `below-floor` occurs **0 times** across the existing marker corpus, so
+gating on it alone would be a decoration that never fires; `sonnet-floor` occurs 11. The wide reading
+("any marker carrying numbers") is deliberately **not** taken — pricing this axis at a near-universal
+rate is the over-trigger `field_verdict_crossfamily_gate.md §7` rejects, and a field required
+everywhere becomes a rubber stamp.
+
+> **Provenance, and what was deliberately NOT imported.** Absorbed from a sibling harness
+> (2026-08-20). That document additionally amends its floor rule so a below-floor pass carrying this
+> line **plus** a crossfamily record passes **without the operator ack**. 🟥 That is a *loosening* of
+> an existing FH gate and **was not adopted** — here the leg is purely additive and `below-floor`
+> still requires `below-floor-ack`, unchanged.
+> 🟥 The same document asserted its own hook read this field and its own fixture suite pinned it.
+> Measured twice with a control (`crossfamily` → 21 hits in the same files): **both were 0**. The
+> prose was portable; the machine was not. Everything claimed in this section's *Enforcement*
+> paragraph was built here, and the fixture file named there is the receipt.
+
+---
+
 ## Wave 2 — Defense Principles
 
 **3 Defense Principles**: (1) Reinforce with external cases via WebSearch — "unique to us" or "structural pattern"?

package/scripts/adapters/peer_resolve.sh

@@ -35,6 +35,38 @@
 #   rc=3   NO_ROOTS    — 루트 목록 자체가 비었다(전제 파손. 「없다」가 아니다)
 #   stderr : rc!=0 일 때 한 줄 사유
 
+# ── 단일 소스 배선 (2026-08-21 F-1) ──────────────────────────────────────────
+# 🟥 위 주석이 «스캐너와 같은 순서여야 한다» 고 **관례로** 적어 두었던 그 중복을 여기서 닫는다.
+#    `scripts/fh_track_resolve.sh` 가 track 이름 → 레포 루트 해석의 단일 소스이고,
+#    아래 `FH_PROJECTS_HOME` 팔은 그 라이브러리를 부른다.
+#
+# ⚠️ **`FH_CLUSTER_ROOTS` 팔은 닫히지 않는다 — 조회 모델이 다르기 때문이다.**
+#    라이브러리는 «루트 하나 밑의 `$root/$c`» 를 묻는다. 이 팔은 «임의 절대경로 목록의
+#    `basename` 이 별칭인가» 를 묻는다. 이름만 같고 의미가 다른 둘을 억지로 합치면 그건
+#    이 수리가 막으려는 결함과 같은 종류다. 그래서 그 팔은 자기 열거를 유지한다
+#    (⇒ `fh_peer_alias_candidates` 는 남는다. 부재 메시지와 강등 폴백도 이걸 쓴다).
+#
+# 라이브러리가 없으면 **수리 이전 동작**으로 강등한다 — 다른 세 소비자와 같은 형태다.
+# (폴백은 의도적으로 옛 구현 그대로다. 공백 결함까지 포함해서 «수리 이전» 이다.)
+_FH_TRLIB="$(cd -P "$(dirname "${BASH_SOURCE[0]:-$0}")/.." 2>/dev/null && pwd)/fh_track_resolve.sh"
+# shellcheck source=scripts/fh_track_resolve.sh
+[ -f "$_FH_TRLIB" ] && . "$_FH_TRLIB"
+type fh_resolve_track_root >/dev/null 2>&1 || fh_resolve_track_root() {
+  local n="$1" root="$2" c hits="" first="" nh
+  for c in "$n" "${n}-dev" "$(printf '%s' "$n" | tr '_' '-')"; do
+    [ -d "$root/$c" ] || continue
+    case " $hits " in *" $c "*) continue ;; esac
+    hits="$hits $c"; [ -n "$first" ] || first="$c"
+  done
+  nh=$(printf '%s' "$hits" | wc -w | tr -d ' ')
+  if [ "${nh:-0}" -gt 1 ]; then
+    printf '%s|AMBIGUOUS:%s' "$root/$n" "$(printf '%s' "$hits" | sed 's/^ //; s/ /,/g')"; return 0
+  fi
+  [ -n "$first" ] || { printf '%s|UNRESOLVED' "$root/$n"; return 0; }
+  [ "$first" = "$n" ] && { printf '%s|' "$root/$n"; return 0; }
+  printf '%s|alias:%s' "$root/$first" "$first"
+}
+
 # 별칭 후보 — 닫힌 목록. 스캐너의 `_resolve_root` 와 **같은 순서**여야 한다.
 fh_peer_alias_candidates() {   # $1=name → 후보를 줄 단위로
   printf '%s\n' "$1" "$1-dev" "$(printf '%s' "$1" | tr '_' '-')"
@@ -74,13 +106,33 @@ EOF
       printf 'peer-resolve: 프로젝트 홈이 없다: %s (전제 파손이지 «peer 없음» 이 아니다)\n' "$home" >&2
       return 3
     fi
-    while IFS= read -r c; do
-      [ -d "$home/$c" ] || continue
-      case " $hits " in *" $home/$c "*) continue ;; esac
-      hits="$hits $home/$c"; [ -n "$first" ] || first="$home/$c"
-    done <<EOF
-$(fh_peer_alias_candidates "$n")
+    # 🟥 여기가 단일 소스로 닫힌 자리다. 옛 인라인 루프(별칭 3종 + 공백 구분자 dedup)를
+    #    `fh_resolve_track_root` 로 대체한다. 출력 계약은 **아래에서 옛 표기로 복원**한다 —
+    #    소비자 4종과 레인은 여전히 rc(0/1/2/3) + 옛 stderr 문구를 읽는다.
+    local rr note
+    rr="$(fh_resolve_track_root "$n" "$home" dir)"
+    note="${rr##*|}"
+    case "$note" in
+      AMBIGUOUS:*)
+        # 옛 표기 복원: 라이브러리는 **후보 이름**을 콤마로 주고, 옛 메시지는 **절대경로**를
+        # «, » 로 이어 붙였다. 공백 없는 이름에서는 바이트 동일하다.
+        local _l="${note#AMBIGUOUS:}" _c _paths=""
+        while IFS= read -r _c; do
+          [ -n "$_c" ] || continue
+          if [ -n "$_paths" ]; then _paths="$_paths, $home/$_c"; else _paths="$home/$_c"; fi
+        done <<EOF
+$(printf '%s' "$_l" | tr ',' '\n')
 EOF
+        printf 'peer-resolve: 한 이름에 레포 여럿 — 고르지 않는다: %s\n' "$_paths" >&2
+        return 2
+        ;;
+      UNRESOLVED)
+        first=""   # 아래 공통 «부재» 분기로 떨어진다 (rc=1 · 옛 문구 그대로)
+        ;;
+      *)
+        first="${rr%|*}"
+        ;;
+    esac
   fi
 
   nh=$(printf '%s' "$hits" | wc -w | tr -d ' ')

package/scripts/cluster_capability_scan.sh

@@ -76,6 +76,31 @@ CAP_SUBDIR=".claude/capabilities"
 # 테스트·다른 배치를 위해 `FH_CLUSTER_ROOTS`(콜론 구분 절대경로 목록)로 통째 대체할 수 있다.
 PROJECTS_HOME="${FH_PROJECTS_HOME:-$HOME/projects}"
 
+# track 이름 → 레포 루트 해석은 scripts/fh_track_resolve.sh 가 단일 소스다 (2026-08-21 F-1).
+# 🟥 **이 파일이 그 라이브러리의 원본이다** — 닫힌 별칭 3종 + 별칭 표면화 + 모호거부는 여기서
+#    뽑아낸 것이고, 갈라져 있던 나머지 두 소비자(fh_session_load.sh · field_canon_preload.sh)가
+#    같은 벌을 쓰게 되는 것이 이 배선의 전부다. 이 파일의 동작은 바뀌지 않는다.
+# 라이브러리가 없으면 **수리 이전 동작**(= 아래 인라인 구현)으로 강등한다 — 다른 두 소비자와
+# 같은 형태다. 새 실패 모드를 만들지 않는다.
+_FH_TRLIB="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)/fh_track_resolve.sh"
+# shellcheck source=scripts/fh_track_resolve.sh
+[ -f "$_FH_TRLIB" ] && . "$_FH_TRLIB"
+type fh_resolve_track_root >/dev/null 2>&1 || fh_resolve_track_root() {
+  local n="$1" root="$2" c hits="" first=""
+  for c in "$n" "$n-dev" "$(printf '%s' "$n" | tr '_' '-')"; do
+    [ -d "$root/$c" ] || continue
+    case " $hits " in *" $c "*) continue ;; esac
+    hits="$hits $c"; [ -n "$first" ] || first="$c"
+  done
+  local nh; nh=$(printf '%s' "$hits" | wc -w | tr -d ' ')
+  if [ "${nh:-0}" -gt 1 ]; then
+    printf '%s|AMBIGUOUS:%s' "$root/$n" "$(printf '%s' "$hits" | sed 's/^ //;s/ /,/g')"; return 0
+  fi
+  [ -n "$first" ] || { printf '%s|UNRESOLVED' "$root/$n"; return 0; }
+  [ "$first" = "$n" ] && { printf '%s|' "$root/$n"; return 0; }
+  printf '%s|alias:%s' "$root/$first" "$first"
+}
+
 _die() { printf '❌ %s\n' "$1" >&2; exit "$RC_HARNESS"; }
 
 # ── `.cap` 파일 열거 — **단 하나의 소스** ────────────────────────────────────
@@ -150,20 +175,24 @@ _enumerate_harnesses() {
 # 🟥 그래도 «조용히 추측» 하지는 않는다. 별칭은 **닫힌 목록**이고, 맞은 별칭은
 #    `_ALIAS_USED` 로 표면화된다. 여러 개가 동시에 맞으면 **고르지 않고 모호로 낸다** —
 #    둘 중 하나를 조용히 고르는 것이 이 파일이 반대하는 그 접힘이다.
+# 🟥 이 함수는 이제 **어댑터**다 — 해석은 `fh_resolve_track_root`(단일 소스)가 하고, 여기서는
+#    이 파일의 **표시 계약**으로 옮긴다. 두 계약이 미묘하게 다르기 때문에 조용히 통과시키지
+#    않는다:
+#      라이브러리  `alias:<c>`   ↔ 여기  `경로 별칭 — alias:<c>`  (OK 행의 DETAIL 로 직접 출력됨)
+#      라이브러리  `UNRESOLVED`  ↔ 여기  빈 문자열              (부재는 `_harness_status` 가
+#                                                              UNREACHABLE 로 판정한다)
+#    `AMBIGUOUS:<...>` 는 양쪽이 같은 표기이고 `_harness_status` 의 `AMBIGUOUS:*` 분기가
+#    그대로 받는다. 술어는 `dir` — 이 스캐너는 «디렉토리가 있나» 를 묻는다(git 레포 여부가
+#    아니다). 그 차이는 의도이지 갈라짐이 아니다(라이브러리 헤더 참조).
 _resolve_root() {  # $1=track 이름 → "<root>|<alias-note>"
-  local n="$1" c hits="" first="" note=""
-  for c in "$n" "$n-dev" "$(printf '%s' "$n" | tr '_' '-')"; do
-    [ -d "$PROJECTS_HOME/$c" ] || continue
-    case " $hits " in *" $c "*) continue ;; esac
-    hits="$hits $c"; [ -n "$first" ] || first="$c"
-  done
-  local nh; nh=$(printf '%s' "$hits" | wc -w | tr -d ' ')
-  if [ "${nh:-0}" -gt 1 ]; then
-    printf '%s|AMBIGUOUS:%s' "$PROJECTS_HOME/$n" "$(printf '%s' "$hits" | sed 's/^ //;s/ /,/g')"
-    return
-  fi
-  [ -n "$first" ] && [ "$first" != "$n" ] && note="경로 별칭 — alias:$first"
-  printf '%s|%s' "$PROJECTS_HOME/${first:-$n}" "$note"
+  local rr note
+  rr="$(fh_resolve_track_root "$1" "$PROJECTS_HOME" dir)"
+  note="${rr#*|}"
+  case "$note" in
+    alias:*)    printf '%s|경로 별칭 — %s' "${rr%%|*}" "$note" ;;
+    UNRESOLVED) printf '%s|' "${rr%%|*}" ;;
+    *)          printf '%s' "$rr" ;;
+  esac
 }
 
 # 한 하네스의 상태를 판정한다. 출력 한 줄: "<name>\t<STATUS>\t<n>\t<detail>"