@event4u/agent-config 14.10.0 → 14.12.0

package/src/config/preamble-payload-budget.json

@@ -1,76 +1,102 @@
 {
-  "_comment": "Per-spawn preamble payload ceiling (road-to-cache-economy Phase 3), unblocked by C-3 measuring 38.0% against a 15% bar. This is a RATCHET, not the roadmap's literal candidate numbers: those (median 40k / p95 50k) were anchored to an upstream ~37k baseline, while this package's own deterministic in-repo payload is ~102.5k tokens \u2014 a hard 40k gate would be red on the day it lands and would teach the reader to ignore it. The ratchet fails on GROWTH instead, so the number can only walk down, and the 40k/50k target stays recorded as the destination. Only the deterministic, in-repo buckets are gated: user-scope rules depend on what the developer happens to have installed globally and are therefore not CI-checkable.",
-  "schema_version": 1,
-  "registered_at": "2026-07-30",
-  "owner": "maintainer",
-  "review_by": "2027-07-30",
-  "metric": {
-    "basis": "chars/4 token estimate over the projected tree, produced by src/scripts/preamble_byte_census.ts",
-    "gated_buckets": [
-      "project-scope rules (dist/agent-src/rules/*.md)",
-      "preloaded skills catalog (name + description, dist/agent-src/skills)",
-      "CLAUDE.md hierarchy (project + project-local)"
-    ],
-    "excluded_buckets": [
-      "user-scope rules (~/.claude/rules/*.md) \u2014 machine-dependent, not CI-checkable",
-      "tool definitions + dispatch prompt \u2014 no local source carries the request payload; the census reports it as an explicit residual",
-      "personas (src/agent-src/personas, 36 files, ~37,181 tok) \u2014 measured ZERO standing cost 2026-08-22: .claude/personas/ is written by this repo's generators (5 in-repo references) and read by no host contract; the host's real subagent surface is .claude/agents/, which this package writes 1 file into. Reversal condition: a host publishing a personas contract, or a generator folding persona prose into an instruction file that IS read. Evidence: agents/evidence/analysis/uncounted-prose-corpora-load-semantics.md",
-      "contexts (src/agent-src/contexts, 58 files, ~114,042 tok) \u2014 measured ON-DEMAND 2026-08-22: 0 rules declare load_context_eager: (17 declare the lazy load_context:), and rule.schema.json:49 documents eager as the variant that counts against the per-rule budget. This records that no eager declaration exists, NOT that eager loading would be free \u2014 nobody has declared one, so nobody has measured it. Evidence: agents/evidence/analysis/uncounted-prose-corpora-load-semantics.md",
-      "skill bodies (src/skills/*/SKILL.md, 290 files, ~595,843 tok) \u2014 measured ON-DEMAND 2026-08-22: censusSkillsCatalog (preamble_byte_census.ts:216-234) builds the gated catalog bucket from `- name: description` per skill and reads no body byte, which is why 290 skills totalling ~595,843 tok report as ~14.4k standing. Corollary: the description cap is the WHOLE skill-side lever, not a fragment of it. Evidence: agents/evidence/analysis/uncounted-prose-corpora-load-semantics.md"
-    ],
-    "_bucket_classification_note": "All three additions above are EXCLUSIONS, not gated buckets, and that is the finding rather than a shortcut. A census bucket over a corpus measured at zero is a permanently green gate certifying nothing. The exclusions carry their measurement so the next pass that proposes one of these corpora as a token lever meets the number instead of the silence."
-  },
-  "baseline_tokens": 102520,
-  "baseline_history": [
-    {
-      "tokens": 102599,
-      "at": "2026-07-30",
-      "why": "registered with the gate"
-    },
-    {
-      "tokens": 102520,
-      "at": "2026-07-31",
-      "why": "free tightening: the measured tree came in 79 tok under the registered baseline, so the ratchet walks down to meet it. Not a claimed reduction \u2014 nothing was optimised; the point of a ratchet is that a lower measurement becomes the new ceiling instead of becoming unused headroom."
-    }
-  ],
-  "headroom_pct": 5,
-  "target_tokens": {
-    "median": 40000,
-    "p95": 50000,
-    "_comment": "The roadmap's candidate numbers, kept as the destination. Reaching them needs reductions this package has not built yet; scope-dedup (C-3) removes a redundant COPY at the consumer, it does not shrink the suite itself."
-  },
-  "target_schedule": {
-    "_comment": "road-to-inbox-harvest-2026-08-d-context-ledger Step 5.3. The target above was a destination with no route: the ratchet fails on GROWTH only, so the corpus can drift away from 40k indefinitely without any gate objecting. These milestones give the destination a schedule. They are DATES WITH NUMBERS, not a derivation - no measurement in this tree says the estate can reach 85k by a given day, and pretending otherwise would be the invented-threshold failure this file's own baseline_history is careful to avoid. What they buy is that a miss becomes a published fact with an owner instead of a silent drift.",
+    "_comment": "Per-spawn preamble payload ceiling (road-to-cache-economy Phase 3), unblocked by C-3 measuring 38.0% against a 15% bar. This is a RATCHET, not the roadmap's literal candidate numbers: those (median 40k / p95 50k) were anchored to an upstream ~37k baseline, while this package's own deterministic in-repo payload is ~102.5k tokens \u2014 a hard 40k gate would be red on the day it lands and would teach the reader to ignore it. The ratchet fails on GROWTH instead, so the number can only walk down, and the 40k/50k target stays recorded as the destination. Only the deterministic, in-repo buckets are gated: user-scope rules depend on what the developer happens to have installed globally and are therefore not CI-checkable.",
+    "schema_version": 1,
+    "registered_at": "2026-07-30",
     "owner": "maintainer",
-    "measured_at_registration": {
-      "tokens": 125593,
-      "at": "2026-08-15",
-      "note": "MEASURED total at registration time, from check_preamble_payload_budget. It is ~23k ABOVE both baseline_tokens (102520) and the derived ceiling, i.e. the gate is red on the tree as it stands. The baseline is deliberately NOT raised to meet it - raising a ratchet baseline to clear a failing check is the config-weakening move this repo blocks by construction. It is recorded here so the milestones below are planned against the real starting point rather than a stale one. Per-rule ranking: agents/evidence/analysis/rule-paths-coverage-census.md"
+    "review_by": "2027-07-30",
+    "metric": {
+        "basis": "chars/4 token estimate over the projected tree, produced by src/scripts/preamble_byte_census.ts",
+        "gated_buckets": [
+            "project-scope rules (dist/agent-src/rules/*.md)",
+            "preloaded skills catalog (name + description, dist/agent-src/skills)",
+            "CLAUDE.md hierarchy (project + project-local)"
+        ],
+        "excluded_buckets": [
+            "user-scope rules (~/.claude/rules/*.md) \u2014 machine-dependent, not CI-checkable",
+            "tool definitions + dispatch prompt \u2014 no local source carries the request payload; the census reports it as an explicit residual",
+            "personas (src/agent-src/personas, 36 files, ~37,181 tok) \u2014 measured ZERO standing cost 2026-08-22: .claude/personas/ is written by this repo's generators (5 in-repo references) and read by no host contract; the host's real subagent surface is .claude/agents/, which this package writes 1 file into. Reversal condition: a host publishing a personas contract, or a generator folding persona prose into an instruction file that IS read. Evidence: agents/evidence/analysis/uncounted-prose-corpora-load-semantics.md",
+            "contexts (src/agent-src/contexts, 58 files, ~114,042 tok) \u2014 measured ON-DEMAND 2026-08-22: 0 rules declare load_context_eager: (17 declare the lazy load_context:), and rule.schema.json:49 documents eager as the variant that counts against the per-rule budget. This records that no eager declaration exists, NOT that eager loading would be free \u2014 nobody has declared one, so nobody has measured it. Evidence: agents/evidence/analysis/uncounted-prose-corpora-load-semantics.md",
+            "skill bodies (src/skills/*/SKILL.md, 290 files, ~595,843 tok) \u2014 measured ON-DEMAND 2026-08-22: censusSkillsCatalog (preamble_byte_census.ts:216-234) builds the gated catalog bucket from `- name: description` per skill and reads no body byte, which is why 290 skills totalling ~595,843 tok report as ~14.4k standing. Corollary: the description cap is the WHOLE skill-side lever, not a fragment of it. Evidence: agents/evidence/analysis/uncounted-prose-corpora-load-semantics.md"
+        ],
+        "_bucket_classification_note": "All three additions above are EXCLUSIONS, not gated buckets, and that is the finding rather than a shortcut. A census bucket over a corpus measured at zero is a permanently green gate certifying nothing. The exclusions carry their measurement so the next pass that proposes one of these corpora as a token lever meets the number instead of the silence."
     },
-    "milestones": [
-      {
-        "by": "2026-11-10",
-        "tokens": 102520,
-        "why": "Return to the registered baseline. This is a RECOVERY milestone, not progress toward 40k - it undoes drift rather than shrinking the estate. The date is the review_by both budget files already carry, so it needs no new calendar."
-      },
-      {
-        "by": "2027-02-10",
-        "tokens": 85000,
-        "why": "First real reduction. The rule-paths census shows 92.1% of rule weight sits in the auto tier and the top 20 files carry 35.5% of it, so a scoped tranche is the plausible mechanism - GATED on the paths-scoping consumer decision, which is a human call and may resolve against scoping entirely."
-      },
-      {
-        "by": "2027-05-10",
-        "tokens": 60000,
-        "why": "Second reduction. No mechanism is committed for this one; it exists so the destination keeps a date after the first tranche lands, and it is the milestone most likely to be revised on evidence."
-      }
+    "baseline_tokens": 102520,
+    "baseline_history": [
+        {
+            "tokens": 102599,
+            "at": "2026-07-30",
+            "why": "registered with the gate"
+        },
+        {
+            "tokens": 102520,
+            "at": "2026-07-31",
+            "why": "free tightening: the measured tree came in 79 tok under the registered baseline, so the ratchet walks down to meet it. Not a claimed reduction \u2014 nothing was optimised; the point of a ratchet is that a lower measurement becomes the new ceiling instead of becoming unused headroom."
+        }
     ],
-    "on_miss": "A missed milestone is PUBLISHED - recorded here with the measured number and the reason - and never silently re-dated. Two consecutive misses mean the schedule was wrong or the destination is unreachable with the mechanisms available; either outcome is a decision to surface, not a date to move again.",
-    "revisit_if": "The paths-scoping decision resolves against scoping (removing the only named mechanism for the 85k milestone), or a measured reduction lands ahead of schedule and the remaining milestones are re-derived from it."
-  },
-  "reference_measurements": {
-    "measured_cold_start_median_tokens": 230930,
-    "c3_scope_dedup_reduction_pct": 38,
-    "c3_condition": "FIXTURE-ONLY as of 2026-07-31: the 38% was measured on a byte-identical two-scope fixture, and that condition is currently unreachable for every consumer because install.ts:2723/2725 stamp package:/source_path: into every installed rule while the in-repo projection stamps nothing. Aligning versions yields 0/110 twins, not 110/110. The figure is the mechanism's ceiling, NOT a realised saving, and this baseline therefore assumes no dedup. Reachability is the subject of road-to-dedup-reachability; run measure_scope_dedup for the live split.",
-    "_comment": "The measured median comes from Claude Code transcripts (deduped by message.id + requestId) and is context for the ratchet, never the gated number \u2014 it moves with the developer's own session history."
-  }
+    "headroom_pct": 5,
+    "target_tokens": {
+        "median": 40000,
+        "p95": 50000,
+        "_comment": "The roadmap's candidate numbers, kept as the destination. Reaching them needs reductions this package has not built yet; scope-dedup (C-3) removes a redundant COPY at the consumer, it does not shrink the suite itself."
+    },
+    "target_schedule": {
+        "_comment": "road-to-inbox-harvest-2026-08-d-context-ledger Step 5.3. The target above was a destination with no route: the ratchet fails on GROWTH only, so the corpus can drift away from 40k indefinitely without any gate objecting. These milestones give the destination a schedule. They are DATES WITH NUMBERS, not a derivation - no measurement in this tree says the estate can reach 85k by a given day, and pretending otherwise would be the invented-threshold failure this file's own baseline_history is careful to avoid. What they buy is that a miss becomes a published fact with an owner instead of a silent drift.",
+        "owner": "maintainer",
+        "measured_at_registration": {
+            "tokens": 125593,
+            "at": "2026-08-15",
+            "note": "MEASURED total at registration time, from check_preamble_payload_budget. It is ~23k ABOVE both baseline_tokens (102520) and the derived ceiling, i.e. the gate is red on the tree as it stands. The baseline is deliberately NOT raised to meet it - raising a ratchet baseline to clear a failing check is the config-weakening move this repo REFUSES. Corrected 2026-08-24 (AC-3): this used to assert the repo blocks it structurally -- the exact phrase is deliberately not reproduced here, because AC-3's check is a grep and quoting a retracted claim would leave the string in the file it was removed from. It over-claimed the mechanism. `block_config_weakening` is a PreToolUse guard that warns from five allowlist entries and blocks past twenty, is honoured only on the one host that denies, and does not gate a baseline edit at all -- a rising count there may be a legitimate ratchet reset the edit alone cannot distinguish. The refusal is real and it is carried by review and by this file's own record, not by construction. It is recorded here so the milestones below are planned against the real starting point rather than a stale one. Per-rule ranking: agents/evidence/analysis/rule-paths-coverage-census.md"
+        },
+        "milestones": [
+            {
+                "by": "2026-11-10",
+                "tokens": 102520,
+                "why": "Return to the registered baseline. This is a RECOVERY milestone, not progress toward 40k - it undoes drift rather than shrinking the estate. The date is the review_by both budget files already carry, so it needs no new calendar."
+            },
+            {
+                "by": "2027-02-10",
+                "tokens": 85000,
+                "why": "First real reduction. The rule-paths census shows 92.1% of rule weight sits in the auto tier and the top 20 files carry 35.5% of it, so a scoped tranche is the plausible mechanism - GATED on the paths-scoping consumer decision, which is a human call and may resolve against scoping entirely."
+            },
+            {
+                "by": "2027-05-10",
+                "tokens": 60000,
+                "why": "Second reduction. No mechanism is committed for this one; it exists so the destination keeps a date after the first tranche lands, and it is the milestone most likely to be revised on evidence."
+            }
+        ],
+        "on_miss": "A missed milestone is PUBLISHED - recorded here with the measured number and the reason - and never silently re-dated. Two consecutive misses mean the schedule was wrong or the destination is unreachable with the mechanisms available; either outcome is a decision to surface, not a date to move again.",
+        "revisit_if": "The paths-scoping decision resolves against scoping (removing the only named mechanism for the 85k milestone), or a measured reduction lands ahead of schedule and the remaining milestones are re-derived from it."
+    },
+    "reference_measurements": {
+        "measured_cold_start_median_tokens": 230930,
+        "c3_scope_dedup_reduction_pct": 38,
+        "c3_condition": "FIXTURE-ONLY as of 2026-07-31: the 38% was measured on a byte-identical two-scope fixture, and that condition is currently unreachable for every consumer because install.ts:2723/2725 stamp package:/source_path: into every installed rule while the in-repo projection stamps nothing. Aligning versions yields 0/110 twins, not 110/110. The figure is the mechanism's ceiling, NOT a realised saving, and this baseline therefore assumes no dedup. Reachability is the subject of road-to-dedup-reachability; run measure_scope_dedup for the live split.",
+        "_comment": "The measured median comes from Claude Code transcripts (deduped by message.id + requestId) and is context for the ratchet, never the gated number \u2014 it moves with the developer's own session history."
+    },
+    "ci_delivery": {
+        "workflow": ".github/workflows/standing-payload-delta.yml",
+        "trigger": "pull_request",
+        "posture": "blocking_on_ceiling_breach_with_grace_ceiling",
+        "design_ceiling": 107646,
+        "grace_ceiling": 138212,
+        "grace_measured_at": "2026-08-24",
+        "grace_end_date": "2026-11-10",
+        "decision": "OPTION (b) of b-preamble-ci-placement: fold into the existing standing-payload-delta.yml rather than add a per-PR job or a scheduled main job. That workflow already computes the neighbouring number and already DECLINED to fail on the premise that this gate 'already fails on growth' -- a premise that was false, because nothing in .github/ ran this gate at all. Making it fail is the smallest change that removes the contradiction instead of adding a third surface. AI council 2026-08-24, 2/2 convergent, on a maintainer delegation.",
+        "why_a_single_ci_run_is_enough": "The 'per-spawn' in this metric's name describes WHERE the cost is paid, not random sampling. The measurement is a deterministic census of revision-controlled inputs -- rules, the skill catalogue and CLAUDE.md, all read from the checkout -- so one CI run represents the checked-out revision exactly. Variation between two runs on the same commit would be a CHECKER DEFECT, never budget noise, and is asserted against by the repeatability test in tests/scripts/check_preamble_payload_budget.test.ts. Both council seats rejected the single-sample objection on this ground.",
+        "why_a_grace_ceiling": "HEAD measures 138,212 against a design ceiling of 107,646 -- 28.4 % over. Making the gate blocking at the design ceiling would fail EVERY pull request from the moment it lands, which is a self-inflicted CI stop rather than a gate: nobody can reduce 30,566 tokens inside the PR that turns the gate on. The grace ceiling is set AT the measurement so growth beyond today reds immediately while today's tree passes, and it ratchets DOWN toward the design ceiling as reductions land. It may never move UP. It expires at the milestone-1 date, at which point the design ceiling applies.",
+        "honest_limit": "Editing this workflow makes it FAIL; it does not make it REQUIRED. Whether a failing check blocks a merge is branch-protection configuration, which is a repo-admin action outside this change and outside an autonomous run. So the claim this record supports is 'the gate can now fail a PR', never 'the gate now blocks a merge'. Verify the latter with `gh api repos/:owner/:repo/rulesets` rather than by reading this file."
+    },
+    "status_2026_08_24": {
+        "state": "at_risk",
+        "measured": 138212,
+        "target": 102520,
+        "target_date": "2026-11-10",
+        "delta_from_target": 35692,
+        "delta_from_registration": 12619,
+        "registration_measurement": 125593,
+        "registered_at": "2026-08-15",
+        "decision": "PUBLISH AND HOLD. The adverse drift is recorded now and the original milestone date is retained. This is an EARLY RISK PUBLICATION, not a milestone miss -- 2026-11-10 has not arrived, and recording a miss before its date would be as wrong as re-dating silently. The file's own on_miss clause ('a missed milestone is PUBLISHED ... and never silently re-dated') applies at the date; this entry is the early warning that keeps the date falsifiable. AI council 2026-08-24, 2/2.",
+        "trajectory": "+12,619 tokens in the nine days from 2026-08-15 to 2026-08-24, i.e. roughly +1,400/day. Reaching 102,520 by 2026-11-10 from 138,212 needs a NET REDUCTION averaged over the remaining 78 days, against a measured net growth. One seat argued the derivative already falsifies the schedule model and that holding for a second miss preserves a signal nobody will act on; the other argued that holding is what keeps the checkpoint falsifiable at all. Recorded unresolved rather than settled, because both readings are defensible and the measurement is what decides between them on 2026-11-10.",
+        "committed_reduction_mechanism": "NONE. Stated explicitly because its absence is the finding: no mechanism in the tree is committed to closing 35,692 tokens by 2026-11-10. road-to-cost-parity-1-rule-payload-diet and road-to-standing-context-40k both name reductions and neither is executing. A schedule with no mechanism is a date, not a plan, and this field exists so the next reader does not have to infer that from silence."
+    }
 }

package/src/config/quorum-attendance-budget.json

@@ -1,11 +1,11 @@
 {
-    "_comment": "Pre-registered council ATTENDANCE metrics (road-to-inbox-harvest-2026-08-b Phase 1.4). Registered here rather than in roadmap prose because a roadmap is archived when it closes and a metric outlives it — the sibling budget files (hook-token-budget.json, recycle-threshold-budget.json) already carry the owner / review_by / honest-gap discipline this reuses. The instrument is the quorum_result line appendQuorumEvent writes to agents/runtime/council/events.log (schema v4 since the ADR-224 shadow floor added gate_class + floor_would_hold at v3 and min_present at v4; v2 lines carry none of the three and v3 lines carry no min_present, so a rate over them must exclude rather than assume them); nothing here is measured yet, and no threshold is committed before data exists.",
+    "_comment": "Pre-registered council ATTENDANCE metrics (road-to-inbox-harvest-2026-08-b Phase 1.4). Registered here rather than in roadmap prose because a roadmap is archived when it closes and a metric outlives it — the sibling budget files (hook-token-budget.json, recycle-threshold-budget.json) already carry the owner / review_by / honest-gap discipline this reuses. The instrument is the quorum_result line appendQuorumEvent writes to agents/runtime/council/events.log (schema v5 since road-to-council-evidence-integrity Phase 3 added stance_agreement, on top of the ADR-224 shadow floor which added gate_class + floor_would_hold at v3 and min_present at v4; v2 lines carry none of the three, v3 lines carry no min_present, and v4-and-earlier lines carry no stance_agreement, so a rate over them must exclude rather than assume them); nothing here is measured yet, and no threshold is committed before data exists.",
     "schema_version": 1,
     "registered_at": "2026-08-10",
     "owner": "maintainer",
     "review_by": "2026-11-10",
     "unit": "quorum_result event lines in agents/runtime/council/events.log",
-    "privacy": "a quorum_result line carries {phase, command, dispatch, verdict, threshold, configured_total, total, present, solo, gate_class, min_present, floor_would_hold, absent[{member, reason}]} — counts, closed vocabularies, booleans, and member ids only. The type has no field able to hold free-form content, so the CLI's own `detail` string (provider error text, which can embed paths and prompt fragments) is dropped at the boundary rather than scrubbed. The two v3 additions are booleans and the v4 addition is a small integer read from config, none of which can widen that surface.",
+    "privacy": "a quorum_result line carries {phase, command, dispatch, verdict, threshold, configured_total, total, present, solo, gate_class, min_present, floor_would_hold, absent[{member, reason}]} — counts, closed vocabularies, booleans, and member ids only. The type has no field able to hold free-form content, so the CLI's own `detail` string (provider error text, which can embed paths and prompt fragments) is dropped at the boundary rather than scrubbed. The two v3 additions are booleans, the v4 addition is a small integer read from config, and the v5 addition is a three-value closed vocabulary (consensus | split | not_tallied) derived from tally_stances — none of which can widen that surface. stance_agreement in particular carries no option LABEL: the labels members vote on are free-form member text, so recording which option won would have opened exactly the free-form hole the type is built to refuse.",
     "phase_split": "A pass emits `pre_run` (build_members — who could be constructed) and, on the `run` path, `post_run` (_postRunQuorum — who actually returned a usable response). They disagree whenever a member constructs and then fails mid-flight, so a rate computed over both double-counts. A `pre_run` line is NOT evidence a pass ran: `command: estimate` is a spend-free preview, and a `run` / `debate` line is written before the `--confirm` gate (which defaults to false and returns before any provider call) and before the necessity and size-fit gates. A `post_run` line is the only line that proves providers were called — which is why every rate below is post_run-scoped, and why `command` exists at all.",
     "attendance_rate": {
         "definition": "of the members that entered the pass, how many answered: present / total per post_run pass. Bounded by 1 by construction (evaluateQuorum clamps present into [0, total]).",
@@ -42,6 +42,14 @@
         "`command: debate` carries the pre_run half ONLY. There is no _postRunQuorum on the debate path, and defining `usable response` for a multi-round debate is a separate question this registration does not answer. So debate contributes to absent_reason_distribution and to nothing else: attendance_rate and solo_conclusion_rate cover the `run` path, and a reader must not read them as covering every metered pass.",
         "A pre_run line proves construction happened, never that a pass followed — the `--confirm` gate (default false), the necessity gate and the size-fit gate all abort afterwards and nothing on the line records it. Any count over pre_run lines is an upper bound; pair it with a post_run line before treating it as a pass.",
         "`gate_class` is `false` on EVERY line written so far and will stay false until a caller declares otherwise — nothing in the tree branches on QuorumStatus to hold a gate, so there is no gate-class caller to mark. shadow_floor_fire_rate is therefore a rate over ALL post_run passes, not over gate-class ones, and the gate-class-scoped rate ADR-224 actually cares about is UNMEASURED, not zero. Do not read a `false` majority as evidence that gate-class passes are rare; it is evidence that the concept has no consumer yet.",
-        "The CLI emit wiring itself has no test. The two modules are covered directly, but nothing exercises build_members / cmd_run / cmd_debate end-to-end against a captured log — and `_emitQuorumEvent` forwards no AppendEventOptions, so there is no target to capture into. Two defects in this file's first version (a `dispatch` written from the flag rather than the outcome, and a rate mixing config entries with clients) were found by review rather than by a test, which is what that gap costs."
-    ]
+        "The CLI emit wiring itself has no test. The two modules are covered directly, but nothing exercises build_members / cmd_run / cmd_debate end-to-end against a captured log — and `_emitQuorumEvent` forwards no AppendEventOptions, so there is no target to capture into. Two defects in this file's first version (a `dispatch` written from the flag rather than the outcome, and a rate mixing config entries with clients) were found by review rather than by a test, which is what that gap costs.",
+        "stance_agreement measures agreement among the members that ANSWERED and were parseable — it says nothing about a member that was absent or unparsed, which the attendance fields on the same line cover separately. A pass reading `1/2 present · stance_agreement: consensus` is one voice agreeing with itself, and the two fields must be read together; neither is a summary of the other."
+    ],
+    "stance_agreement_rate": {
+        "definition": "share of passes whose seats reached consensus: `stance_agreement: 'consensus'` over post_run/command=run lines AT SCHEMA >= 5. Three values, not a boolean, and the third is why this metric can exist honestly: stance tallying is opt-in and off by default, so most passes never ask the question — `not_tallied` records that, where a boolean would have been forced to write `false` and read as disagreement. Written by tally_stances' own verdict (`consensus !== null` vs `split`), never re-derived from the tally numbers here, the same one-definition discipline solo_conclusion_rate follows by deferring to isSoloConcluded.",
+        "instrument": "quorum_result lines, phase=post_run, command=run, field stance_agreement — replayed by `./scripts-run src/scripts/council_attendance_metrics`, which prints this rate next to its denominator and its excluded count",
+        "threshold": "none committed before data — baseline first, threshold at review date. Nothing is gated on this value and no pass is held, delayed or failed by it.",
+        "exclude_pre_v5_lines_or_the_rate_is_invented": "A v4-and-earlier line carries no stance_agreement field and NO honest value can be inferred for it: that pass may have run a tally whose verdict was never recorded, or none at all, and the line cannot tell you which. So the denominator is v5+ lines only. Defaulting the missing field to `not_tallied` manufactures a measurement out of an absence — on the committed span fixture it moves the rate from 1/3 to 1/6, a plausible number with nothing behind it. `attendance_metrics.test.ts` asserts both the exclusion and its mirror (dropping the v4 stratum must not move the rate), and the sabotage that defaults instead of excluding fails both.",
+        "denominator_is_not_the_attendance_denominator": "agreement_eligible is a SUBSET of passes, so this rate and the four above are computed over different populations and must never be presented as parts of one table without their own denominators printed. The replay prints both counts on the line for exactly that reason."
+    }
 }

package/src/config/routing-coverage-seed.json

@@ -0,0 +1,29 @@
+{
+    "_comment": "Routing-coverage ratchet seeds (road-to-routing-assurance 0.3). Coverage ratio = corpus cases / routed units, per scope, seeded at the MEASURED current value on 2026-08-25. CI fails only on a decrease -- the COUNT-ratchet disposition of lint_trigger_precision.ts. TWO scopes and never one aggregate: a blended figure would read ~0.46 and describe neither surface. Lowering a seed is a defect, not a fix: a ratio falls when a corpus case is removed OR when units are added without cases, and both are things to repair rather than to record.",
+    "schema_version": 1,
+    "registered_at": "2026-08-25",
+    "owner": "maintainer",
+    "review_by": "2027-08-25",
+    "measurement": {
+        "rules": "intersection of tests/eval/routing-matrix/*.yaml stems with the tier_1+tier_2 ids in dist/router.json, over that id count. Intersection rather than the file count, so a fixture naming no routed rule cannot raise the ratio.",
+        "skills": "src/skills/*/evals/triggers.json over src/skills/*/SKILL.md."
+    },
+    "seed": {
+        "rules": 0.8952,
+        "skills": 0.301
+    },
+    "seed_provenance": {
+        "rules": "94 covered of 105 routed rules (tier-1 21/24, tier-2 73/81) on 2026-08-25.",
+        "skills": "76 of 299 skills carry evals/triggers.json on 2026-08-25.",
+        "finding": "The gap between the two seeds IS the roadmap defect D1, quantified. The rules surface is ~90% covered by a deterministic corpus that can fail a PR; the skills surface -- the one production routes on -- is 25%, covered only by a harness that is advisory only, never gating (rule_trigger_eval.ts:4)."
+    },
+    "history": [
+        {
+            "date": "2026-08-25",
+            "scope": "skills",
+            "from": 0.2542,
+            "to": 0.301,
+            "reason": "road-to-routing-assurance Phase 2.2 authored 14 high-tier corpus files (76 -> 90 of 299). The seed rises so the gain cannot be given back silently; it may never be LOWERED."
+        }
+    ]
+}

package/src/config/rule-activation-census.json

@@ -1,8 +1,8 @@
 {
-    "_comment": "Rule-activation ratchet baseline (road-to-mixed-trigger-activation-cost step 4.1). Two axes: the scoped/mixed ID SETS are pinned by identity, and the unconditional token total may only walk DOWN. Raising a number here to clear a failing check is the config-weakening move this repo blocks by construction \u2014 every raise needs a baseline_history entry whose reason is a real sentence. Regenerate with `check_rule_activation_census --write-baseline`; never hand-edit the counts.",
+    "_comment": "Rule-activation ratchet baseline (road-to-mixed-trigger-activation-cost step 4.1). Two axes: the scoped/mixed ID SETS are pinned by identity, and the unconditional token total may only walk DOWN. Raising a number here to clear a failing check is the config-weakening move this repo blocks by construction — every raise needs a baseline_history entry whose reason is a real sentence. Regenerate with `check_rule_activation_census --write-baseline`; never hand-edit the counts.",
     "measured_at_commit": "unrecorded",
     "tokens_exact": true,
-    "unconditional_tokens": 111012,
+    "unconditional_tokens": 113699,
     "scoped_ids": [
         "design-review-after-ui-write",
         "roadmap-progress-sync",
@@ -39,11 +39,15 @@
         },
         {
             "date": "2026-08-20",
-            "reason": "Re-anchored WITH an authorized addition, not to clear a red: the new `recurring-criticism` rule plus the reconciling edits to `self-repair-loop` and `decision-revisit-gate`. The cost is structural, not sloppy authoring \u2014 the rule fires on a recurrence phrase in the USER PROMPT, and on Claude the emitter writes no `paths:` at all once any non-path trigger is present, so a prompt-triggered rule is unconditional by construction. There is no scoped variant of this obligation to choose instead: a path-scoped version would never see the prompt that triggers it, which is the whole mechanism. The body was already cut once for this reason (~45 percent, depth moved to guideline:agent-infra/recurring-criticism-mechanics, which is read on demand and costs nothing at session start); what remains is the Iron Law, the three-outcome table that is the anti-sycophancy guard, and the honest-enforcement paragraph. Trimming further would remove the reasoning the rule itself marks as the clause most likely to be trimmed. Delta 3575 tokens against a 2000 allowance, standing cost per session, stated rather than hidden."
+            "reason": "Re-anchored WITH an authorized addition, not to clear a red: the new `recurring-criticism` rule plus the reconciling edits to `self-repair-loop` and `decision-revisit-gate`. The cost is structural, not sloppy authoring — the rule fires on a recurrence phrase in the USER PROMPT, and on Claude the emitter writes no `paths:` at all once any non-path trigger is present, so a prompt-triggered rule is unconditional by construction. There is no scoped variant of this obligation to choose instead: a path-scoped version would never see the prompt that triggers it, which is the whole mechanism. The body was already cut once for this reason (~45 percent, depth moved to guideline:agent-infra/recurring-criticism-mechanics, which is read on demand and costs nothing at session start); what remains is the Iron Law, the three-outcome table that is the anti-sycophancy guard, and the honest-enforcement paragraph. Trimming further would remove the reasoning the rule itself marks as the clause most likely to be trimmed. Delta 3575 tokens against a 2000 allowance, standing cost per session, stated rather than hidden."
         },
         {
             "date": "2026-08-20",
             "reason": "Raised 103,265 -> 106,386 (+3,121) for one new rule, fix-what-you-see, added on maintainer instruction after a session identified a red check's cause, named the file's author, and handed it back unfixed. The rule permits exactly two dispositions -- fix it, or ship a tracked follow-up roadmap in the same change -- and removes ownership and note-only as escapes from active-remediation's ladder. WHAT THIS RE-ANCHOR EXPOSES, and it is the reason the entry is this long: the rule ships type auto rather than always because check_always_budget has two characters of headroom and its cap may only move down. That switch cleared the always gate and did NOT reduce the cost, because this census counts a rule as standing payload unless it is path-SCOPED, and a red CI check has no file pattern to scope on. So the two gates disagree about what a keyword-triggered rule costs, and the auto downgrade moved the same 3,121 tokens from a budget that refuses them into one that only asks for a sentence. Stated rather than exploited: the honest reading is that this rule costs standing payload whatever its type says, and the always-vs-auto choice bought delivery breadth, not weight."
+        },
+        {
+            "date": "2026-08-23",
+            "reason": "Re-anchored for one new rule, playbook-precedence, which ships with ADR-244's sixth context type and states that a repository's OWN configured playbook outranks a shipped skill for the same task. WHY THE DELTA IS THIS LARGE, stated rather than implied: the rule is 4,398 chars of source, but it declares seven triggers and every one of them is a keyword or phrase, so the emitter writes NO `paths:` block at all and the host loads it in every session -- the same mixed-trigger cost `ui-audit-gate` and `design-review-after-ui-write` each paid before being re-scoped to paths in road-to-mixed-trigger-activation-cost Phase 2. That is the honest reading of this number: it is not the price of the obligation, it is the price of expressing the obligation with keywords. A playbook lives at a known location, so a path-scoped form is available and would return most of these tokens; it is not cut here because the trigger set is the rule author's design decision and re-cutting it inside a merge-queue drain would change routing behaviour nobody reviewed. Recorded as the next candidate for that Phase-2 treatment rather than silently absorbed."
         }
     ]
 }

package/src/config/rule-enforcement-baseline.json

@@ -19,7 +19,6 @@
   "content-quoting-floor.md",
   "copilot-routing.md",
   "cross-source-consistency.md",
-  "decision-revisit-gate.md",
   "delegation-policy.md",
   "design-fidelity.md",
   "devcontainer-routing.md",

package/src/config/rule-stub-ceilings.json

@@ -1,94 +1,115 @@
 {
-    "_comment": "Per-rule stub ceilings (road-to-rule-stub-projection Phase 2). One exact-BPE token ceiling per rule carrying a migration pointer. SHRINK-ONLY: a rule may fall below its ceiling freely, and may not rise above it. Raising a ceiling to clear a failing check is the config-weakening move this repo blocks by construction \u2014 every raise needs a `history` entry whose `reason` is a real sentence. The ceilings are a floor-plus-residue SNAPSHOT taken at adoption, not a target: a rule at its ceiling is not thereby correct, only not worse. Regenerate with `check_rule_stub_ceiling --write-baseline`; never hand-edit the counts.",
-    "measured_at_commit": "05eeb5ca3",
-    "tokens_exact": true,
-    "ceilings": {
-        "active-remediation": 1601,
-        "analysis-skill-routing": 76,
-        "architecture": 955,
-        "artifact-drafting-protocol": 680,
-        "augment-edit-discipline": 137,
-        "autonomous-execution": 1633,
-        "brand-consistency": 133,
-        "broken-access-control": 743,
-        "cli-output-handling": 83,
-        "code-comment-discipline": 692,
-        "commit-conventions": 136,
-        "context-hygiene": 2470,
-        "copilot-routing": 128,
-        "design-fidelity": 2245,
-        "devcontainer-routing": 128,
-        "docker-commands": 79,
-        "domain-adoption-policy": 497,
-        "framework-neutrality-in-generic-skills": 621,
-        "git-history-discipline": 1205,
-        "improve-before-implement": 845,
-        "laravel-routing": 126,
-        "laravel-translations": 82,
-        "legal-safety-floor": 1401,
-        "linked-projects-onboarding-gate": 172,
-        "minimal-safe-diff": 893,
-        "missing-tool-handling": 103,
-        "model-recommendation": 156,
-        "no-roadmap-references": 644,
-        "package-ci-checks": 74,
-        "persona-governance": 330,
-        "php-coding": 80,
-        "provider-lifecycle-discipline": 472,
-        "reviewer-awareness": 131,
-        "roadmap-ci-steps-policy": 564,
-        "roadmap-progress-sync": 2479,
-        "rule-type-governance": 87,
-        "skill-improvement-trigger": 81,
-        "skill-quality": 78,
-        "slash-command-routing-policy": 77,
-        "symfony-routing": 129,
-        "ui-audit-gate": 1280,
-        "untrusted-input-defense": 934,
-        "upstream-proposal": 76
+  "_comment": "Per-rule stub ceilings (road-to-rule-stub-projection Phase 2). One exact-BPE token ceiling per rule carrying a migration pointer. SHRINK-ONLY: a rule may fall below its ceiling freely, and may not rise above it. Raising a ceiling to clear a failing check is the config-weakening move this repo blocks by construction \u2014 every raise needs a `history` entry whose `reason` is a real sentence. The ceilings are a floor-plus-residue SNAPSHOT taken at adoption, not a target: a rule at its ceiling is not thereby correct, only not worse. Regenerate with `check_rule_stub_ceiling --write-baseline`; never hand-edit the counts.",
+  "measured_at_commit": "178b89235",
+  "tokens_exact": true,
+  "ceilings": {
+    "active-remediation": 1601,
+    "analysis-skill-routing": 76,
+    "architecture": 955,
+    "artifact-drafting-protocol": 680,
+    "augment-edit-discipline": 137,
+    "autonomous-execution": 1633,
+    "brand-consistency": 133,
+    "broken-access-control": 743,
+    "cli-output-handling": 83,
+    "code-comment-discipline": 692,
+    "commit-conventions": 136,
+    "context-hygiene": 2470,
+    "copilot-routing": 128,
+    "design-fidelity": 2540,
+    "devcontainer-routing": 128,
+    "docker-commands": 79,
+    "domain-adoption-policy": 497,
+    "framework-neutrality-in-generic-skills": 636,
+    "git-history-discipline": 1205,
+    "improve-before-implement": 845,
+    "laravel-routing": 126,
+    "laravel-translations": 82,
+    "legal-safety-floor": 1401,
+    "linked-projects-onboarding-gate": 172,
+    "minimal-safe-diff": 893,
+    "missing-tool-handling": 103,
+    "model-recommendation": 156,
+    "no-roadmap-references": 644,
+    "package-ci-checks": 74,
+    "persona-governance": 330,
+    "php-coding": 80,
+    "provider-lifecycle-discipline": 472,
+    "reviewer-awareness": 131,
+    "roadmap-ci-steps-policy": 564,
+    "roadmap-progress-sync": 2479,
+    "rule-type-governance": 87,
+    "skill-improvement-trigger": 81,
+    "skill-quality": 78,
+    "slash-command-routing-policy": 77,
+    "symfony-routing": 129,
+    "ui-audit-gate": 1491,
+    "untrusted-input-defense": 934,
+    "upstream-proposal": 76
+  },
+  "history": [
+    {
+      "date": "2026-08-19",
+      "rule": "roadmap-progress-sync",
+      "from": 1424,
+      "to": 2117,
+      "reason": "Iron Law 3 gained the preservation test that decides WHO resolves a deferred item \u2014 a routing table plus the two council refinements that make it binding (a follow-up naming a roadmap that does not exist yet fails closed to the user; a council resolution is recorded at the item or it did not happen) and the residual limit both authoring seats named. This is obligation surface, not explanation: the table decides a route and the two refinements are the conditions under which the route is legal, so a reader who reaches only the pointer target has already made the decision. It could not move to the pointer target either \u2014 roadmap-progress-mechanics.md sits at the 16,000-char check_depth_budget ceiling in this same change, and pushing prose there reds a count ratchet whose whole job is refusing a fifth over-ceiling file. The section was condensed once before this raise (2201 -> 2117) by turning four prose paragraphs into bullets; further cutting would delete substance, which preservation-guard forbids."
     },
-    "history": [
-        {
-            "date": "2026-08-19",
-            "rule": "roadmap-progress-sync",
-            "from": 1424,
-            "to": 2117,
-            "reason": "Iron Law 3 gained the preservation test that decides WHO resolves a deferred item \u2014 a routing table plus the two council refinements that make it binding (a follow-up naming a roadmap that does not exist yet fails closed to the user; a council resolution is recorded at the item or it did not happen) and the residual limit both authoring seats named. This is obligation surface, not explanation: the table decides a route and the two refinements are the conditions under which the route is legal, so a reader who reaches only the pointer target has already made the decision. It could not move to the pointer target either \u2014 roadmap-progress-mechanics.md sits at the 16,000-char check_depth_budget ceiling in this same change, and pushing prose there reds a count ratchet whose whole job is refusing a fifth over-ceiling file. The section was condensed once before this raise (2201 -> 2117) by turning four prose paragraphs into bullets; further cutting would delete substance, which preservation-guard forbids."
-        },
-        {
-            "date": "2026-08-20",
-            "rule": "context-hygiene",
-            "from": 2447,
-            "to": 2470,
-            "reason": "One sentence, +23 tokens, and it is obligation surface rather than explanation: for CI, do not hand-write the wait loop. The mechanics went to the pointer target where this rule's own migration note already sends them (context-hygiene-mechanics gains the exit codes, the four not-settled states and the measured incident) \u2014 that move is what took the growth from +179 to +23, and the residue is the pointer itself, which cannot live in the guideline because a rule that does not name the helper is a rule nobody follows to it. The incident it encodes: a hand-written `until ! gh pr checks N | grep -q pending` exited when the API failed, because the error text contains no `pending`, and the session reported a settle that had not happened and retracted it in the next reply. The rule's neighbouring waiter law already says one waiter per condition; what it did not say is that the exit condition is the part that gets written wrong, which is why the helper is named here rather than described."
-        },
-        {
-            "date": "2026-08-20",
-            "rule": "roadmap-progress-sync",
-            "from": 2117,
-            "to": 2185,
-            "reason": "One sentence in Iron Law 1, +68 tokens, and it is obligation surface rather than explanation: `roadmap:progress` now passes `--archive`, so the law's operational meaning changed \u2014 a complete roadmap is moved by the same run instead of reported to a human, and the two exclusions a reader must know before trusting that (`--check` never archives, the PostToolUse hook never archives) are exactly the part that cannot be deferred to a pointer. The prose was NOT moved to docs/guidelines/agent-infra/roadmap-progress-mechanics.md, which is where this rule's migration note sends its mechanics, because that file has roughly a hundred characters of headroom under the 16,000-char depth ceiling and check_depth_budget is shrink-only with four files already over it \u2014 a FIFTH over-ceiling file reds, so moving 400 characters there would have traded this gate for that one. The full contract, the council record and its recorded dissent live with the code in src/agent-src/scripts/archival_sweep.ts, which is the third home and the one with room."
-        },
-        {
-            "date": "2026-08-20",
-            "rule": "artifact-drafting-protocol",
-            "from": 611,
-            "to": 680,
-            "reason": "Three lines plus a pointer extension, +69 tokens, for a new contract field that governs what a run may author without asking: an artifact discovered AFTER an accepted execution contract halts by default, and `late_artifacts: auto-research` is the only declared value that continues, capped at three per run. The five-step auto-research procedure, the reasoning for the default and its reversibility argument all went to the pointer target (artifact-drafting-protocol-mechanics gains a Late artifacts section) rather than here, which is what held the growth to a bullet. The residue is obligation surface and cannot live in the guideline: this rule's own batch-mode bullet already tells the reader that a mid-run artifact triggers the full interactive protocol, so a reader who never opens the guideline would conclude that a declared value cannot change it. Named default rather than described behaviour, because the default direction is the decision -- AI council 2/2 (anthropic + openai) chose `halt` over `auto-research` on the autonomy-defaults-sheet fork of road-to-user-out-of-the-loop, record agents/evidence/council/drain-blocker-dispositions-a.md."
-        },
-        {
-            "date": "2026-08-20",
-            "rule": "roadmap-progress-sync",
-            "from": 2185,
-            "to": 2479,
-            "reason": "One subsection, +294 tokens, for the `deferred_policy` contract field: a declared policy removes the synchronous options ROUND at closure without moving any row of the preservation test's routing table. It is obligation surface and it cannot live anywhere else. Not in the pointer target -- roadmap-progress-mechanics.md sits at 16,039 chars against check_depth_budget's 16,000-char ceiling, and that gate is a count ratchet whose next failure is a FIFTH over-ceiling file, so prose pushed there is the one move that turns a warn into a red. Not in the contract context either: Iron Law 3 is the gate being amended, and a reader who learns from roadmap-execution-contract that a field exists but never reads this rule would not know that `cancel-with-memo` still stops for the user. The three-row table is the delta in full; the reasoning, both alternatives and the reversibility argument went to the contract context's own decision comment. Condensed once before this raise (2509 -> 2411) by cutting a paragraph that restated the preservation test. AI council 2/2 (anthropic + openai) on the autonomy-defaults-sheet fork of road-to-user-out-of-the-loop, record agents/evidence/council/drain-blocker-dispositions-a.md. RE-BASED on merge with origin/main, 2026-08-20: this entry was authored against a base of 2117 and raised to 2411. main had independently raised the same ceiling 2117 -> 2185 for its own reason, so the merged tree measures 2479 and the honest arithmetic for THIS raise is 2185 -> 2479 -- the same +294 for the same subsection, re-based rather than re-argued. Recorded rather than silently regenerated because the gate's own failure text names a history-versus-ceilings disagreement as the signature of a hand-edited count, and a merge that quietly adopted one side's numbers would produce exactly that signature with nothing saying why."
-        },
-        {
-            "date": "2026-08-22",
-            "rule": "untrusted-input-defense",
-            "from": 845,
-            "to": 934,
-            "reason": "Six lines, +89 tokens, correcting a claim the section itself had made false. It said no content-scanning backstop existed and that a future one would be the first candidate to change enforced_by; injection_scan_hook.ts now exists, so the first half is false, and the second was always wrong because a warn-only hook cannot refuse and therefore cannot enforce. Every sentence left carries one of those claims. The measurement half -- slot, 15 channels, 99.00 percent recall, 0.85 percent false positives, the manifest keys -- did move to the pointer target, which is what cut the overshoot from +194 to +89; what stays is the rule's own enforcement disclosure, and a stub ceiling that forbids correcting a false statement about the rule's own enforcement is the wrong constraint applied to the right file."
-        }
-    ]
+    {
+      "date": "2026-08-20",
+      "rule": "context-hygiene",
+      "from": 2447,
+      "to": 2470,
+      "reason": "One sentence, +23 tokens, and it is obligation surface rather than explanation: for CI, do not hand-write the wait loop. The mechanics went to the pointer target where this rule's own migration note already sends them (context-hygiene-mechanics gains the exit codes, the four not-settled states and the measured incident) \u2014 that move is what took the growth from +179 to +23, and the residue is the pointer itself, which cannot live in the guideline because a rule that does not name the helper is a rule nobody follows to it. The incident it encodes: a hand-written `until ! gh pr checks N | grep -q pending` exited when the API failed, because the error text contains no `pending`, and the session reported a settle that had not happened and retracted it in the next reply. The rule's neighbouring waiter law already says one waiter per condition; what it did not say is that the exit condition is the part that gets written wrong, which is why the helper is named here rather than described."
+    },
+    {
+      "date": "2026-08-20",
+      "rule": "roadmap-progress-sync",
+      "from": 2117,
+      "to": 2185,
+      "reason": "One sentence in Iron Law 1, +68 tokens, and it is obligation surface rather than explanation: `roadmap:progress` now passes `--archive`, so the law's operational meaning changed \u2014 a complete roadmap is moved by the same run instead of reported to a human, and the two exclusions a reader must know before trusting that (`--check` never archives, the PostToolUse hook never archives) are exactly the part that cannot be deferred to a pointer. The prose was NOT moved to docs/guidelines/agent-infra/roadmap-progress-mechanics.md, which is where this rule's migration note sends its mechanics, because that file has roughly a hundred characters of headroom under the 16,000-char depth ceiling and check_depth_budget is shrink-only with four files already over it \u2014 a FIFTH over-ceiling file reds, so moving 400 characters there would have traded this gate for that one. The full contract, the council record and its recorded dissent live with the code in src/agent-src/scripts/archival_sweep.ts, which is the third home and the one with room."
+    },
+    {
+      "date": "2026-08-20",
+      "rule": "artifact-drafting-protocol",
+      "from": 611,
+      "to": 680,
+      "reason": "Three lines plus a pointer extension, +69 tokens, for a new contract field that governs what a run may author without asking: an artifact discovered AFTER an accepted execution contract halts by default, and `late_artifacts: auto-research` is the only declared value that continues, capped at three per run. The five-step auto-research procedure, the reasoning for the default and its reversibility argument all went to the pointer target (artifact-drafting-protocol-mechanics gains a Late artifacts section) rather than here, which is what held the growth to a bullet. The residue is obligation surface and cannot live in the guideline: this rule's own batch-mode bullet already tells the reader that a mid-run artifact triggers the full interactive protocol, so a reader who never opens the guideline would conclude that a declared value cannot change it. Named default rather than described behaviour, because the default direction is the decision -- AI council 2/2 (anthropic + openai) chose `halt` over `auto-research` on the autonomy-defaults-sheet fork of road-to-user-out-of-the-loop, record agents/evidence/council/drain-blocker-dispositions-a.md."
+    },
+    {
+      "date": "2026-08-20",
+      "rule": "roadmap-progress-sync",
+      "from": 2185,
+      "to": 2479,
+      "reason": "One subsection, +294 tokens, for the `deferred_policy` contract field: a declared policy removes the synchronous options ROUND at closure without moving any row of the preservation test's routing table. It is obligation surface and it cannot live anywhere else. Not in the pointer target -- roadmap-progress-mechanics.md sits at 16,039 chars against check_depth_budget's 16,000-char ceiling, and that gate is a count ratchet whose next failure is a FIFTH over-ceiling file, so prose pushed there is the one move that turns a warn into a red. Not in the contract context either: Iron Law 3 is the gate being amended, and a reader who learns from roadmap-execution-contract that a field exists but never reads this rule would not know that `cancel-with-memo` still stops for the user. The three-row table is the delta in full; the reasoning, both alternatives and the reversibility argument went to the contract context's own decision comment. Condensed once before this raise (2509 -> 2411) by cutting a paragraph that restated the preservation test. AI council 2/2 (anthropic + openai) on the autonomy-defaults-sheet fork of road-to-user-out-of-the-loop, record agents/evidence/council/drain-blocker-dispositions-a.md. RE-BASED on merge with origin/main, 2026-08-20: this entry was authored against a base of 2117 and raised to 2411. main had independently raised the same ceiling 2117 -> 2185 for its own reason, so the merged tree measures 2479 and the honest arithmetic for THIS raise is 2185 -> 2479 -- the same +294 for the same subsection, re-based rather than re-argued. Recorded rather than silently regenerated because the gate's own failure text names a history-versus-ceilings disagreement as the signature of a hand-edited count, and a merge that quietly adopted one side's numbers would produce exactly that signature with nothing saying why."
+    },
+    {
+      "date": "2026-08-22",
+      "rule": "untrusted-input-defense",
+      "from": 845,
+      "to": 934,
+      "reason": "Six lines, +89 tokens, correcting a claim the section itself had made false. It said no content-scanning backstop existed and that a future one would be the first candidate to change enforced_by; injection_scan_hook.ts now exists, so the first half is false, and the second was always wrong because a warn-only hook cannot refuse and therefore cannot enforce. Every sentence left carries one of those claims. The measurement half -- slot, 15 channels, 99.00 percent recall, 0.85 percent false positives, the manifest keys -- did move to the pointer target, which is what cut the overshoot from +194 to +89; what stays is the rule's own enforcement disclosure, and a stub ceiling that forbids correcting a false statement about the rule's own enforcement is the wrong constraint applied to the right file."
+    },
+    {
+      "date": "2026-08-23",
+      "rule": "design-fidelity",
+      "from": 2245,
+      "to": 2540,
+      "reason": "One subsection, +295 tokens, for the second axis: artefact MATURITY is a property of the handed-over artefact, instruction MANDATE is a property of the prose, and `strict` on a self-declared wireframe is a structure mandate rather than a pixel mandate. It is an Iron Law plus the two discriminator clauses that decide it -- obligation surface, which this repository's rule/mechanics split puts in the rule by construction. The lookup half ALREADY went to the pointer target in the same change: the maturity-to-spec table and the pinned `daf-wireframe-near-miss` row live in design-fidelity-mechanics.md \u00a7 Artefact maturity, and the rule links them. Pushing the remaining prose there too is the one move that is not available -- that migration is exactly what took the mechanics file from 15,265 to 16,524 chars past check_depth_budget's 16,000 ceiling, and paying for it required splitting the asset & imagery block out into its own guideline in this same commit series. A second push would re-red the ratchet the split just cleared. Raised rather than condensed because the two clauses that would compress are the ones that carry the decision: 'the discriminator reads the ARTEFACT, never the prose' and 'an undeclared artefact is treated as finished' -- drop either and the rule authorises the redesign it exists to prevent."
+    },
+    {
+      "date": "2026-08-23",
+      "rule": "ui-audit-gate",
+      "from": 1280,
+      "to": 1491,
+      "reason": "road-to-frontend-power E2.2 moved this rule's `enforced_by:` from `none` to `hook:design-pass`, because `agent-config ui:audit` now writes the artefact the rule previously had nothing to point at. +127 tokens, and every one of them is the SCOPE of that change rather than a description of the mechanism: the check is FRESHNESS not existence (a stale artefact looks like evidence, which is worse than a missing one), the carrier is default-OFF and warns before it blocks, and the greenfield asymmetry is NOT closed -- `_lib/ui_surface.ts` is a path predicate, measured recall 20/23 with all three misses being a UI request that has produced no UI file yet, so this gate cannot fire on the first write of a new surface. Condensed first, from +279 to +127, by cutting the narrative framing and the changed-in-both-directions paragraph. The three limits are what remains and they are the reason the raise is legitimate: a rule that newly claims a carrier and does not state where the carrier stops is a stronger claim than the tree supports, which is the exact over-claim this repository has already had to correct twice in the hook-architecture host table. Not pushed to the pointer target: `existing-ui-audit` is a SKILL describing how to run an audit, and the boundary of a gate's enforcement is an obligation-surface fact -- putting it there would file the limit where nobody reading the gate will find it. Amended in the same change, 1407 -> 1491 (+84): landing the carrier revealed that `design-pass-stop` is `severity: advisory` and the dispatcher enforces advisory as a CEILING, so an EXIT_BLOCK from it is downgraded and then mapped to exit 0. Claiming a block in this rule while the transport discards it would be the inert-refusal defect `run-continuation` already recorded once. Limit 2 therefore states what actually happens -- the stop verdict is REPORTED as would-block, not enforced -- and names what would make it real: an entry in concern_severity.test.ts's BLOCKING_ALLOWLIST, which that file calls a security-relevant decision and which would be the third turn-END refusal in the tree. That is a stronger reason to spend the tokens than the original raise had: a rule whose enforced_by names a carrier, and which does not say the carrier cannot refuse, is the over-claim this whole section exists to prevent."
+    },
+    {
+      "date": "2026-08-23",
+      "rule": "framework-neutrality-in-generic-skills",
+      "from": 621,
+      "to": 636,
+      "reason": "Two entries on the exempt list, +15 tokens: `^monorepo-workspace$` and `^workspace-link$`, the two skills road-to-monorepo-scope-and-detection adds. The list is not lookup material and cannot go to the pointer target: it is the rule's own ACTIVATION condition -- the gate decides from it whether an artefact is generic at all, so a rule whose exempt list lives elsewhere cannot answer its own first question. Every prior addition to that list moved this number for the same reason, and the alternative the gate offers (move prose to the pointer target) has nothing to move here: the two entries ARE the change. Raised rather than condensed because 15 tokens of pattern cannot be condensed without deleting one of the two patterns, which would make a framework-specific skill fail a rule it is exempt from."
+    }
+  ]
 }

package/src/config/suggestion-capture.json

@@ -0,0 +1,28 @@
+{
+    "_comment": "Record schema for the suggestion-block capture instrument (road-to-suggestion-block-capture Phase 1.3). Registered BEFORE the hook exists, so the field set is a pre-commitment rather than a description of whatever the code ended up writing. COUNTS-ONLY BY CONSTRUCTION: there is no field able to hold free text — no prompt, no option label, no command name beyond the evidence class. That is the discipline dispatch-economy-metrics.json:27 fixes as 'a count derived from the result object, never its content', and it is the reason a privacy review here is a schema read rather than a scrubber audit. Every budget-shaped config in this tree carries owner and review_by per the budget-ownership lint.",
+    "schema_version": 1,
+    "registered_at": "2026-08-24",
+    "owner": "maintainer",
+    "review_by": "2027-08-24",
+    "sink": "agents/runtime/state/audit/suggestion-capture.jsonl",
+    "record_fields": {
+        "ts": "ISO-8601 UTC second precision. Not a duration and not a correlation key.",
+        "block_emitted": "boolean — the previous assistant turn matched the render signature.",
+        "options_count": "integer — how many numbered options the block carried. A COUNT of options, never their labels.",
+        "evidence_class": "enum: render-signature | latch-consumed | latch-stale | no-latch. WHY the classification was reachable, not what the turn said.",
+        "turn_classification": "enum: option_n | as_is | other | stale_block."
+    },
+    "classification_enum": ["option_n", "as_is", "other", "stale_block"],
+    "evidence_class_enum": ["render-signature", "latch-consumed", "latch-stale", "no-latch"],
+    "field_ban": {
+        "rule": "No record field may hold a prompt, an option label, a command name, a file path, or any other free-form string. The enums above are closed; a value outside them is a defect in the writer, not a new category.",
+        "why_a_ban_and_not_a_scrubber": "A scrubber can fail and leave content in the sink. A type with no field able to carry content cannot. Same reasoning as the telemetry engagement event, which was deliberately built with no payload/notes/extra key.",
+        "checked_by": "tests/scripts/suggestion_capture.test.ts asserts the written line's key set equals record_fields exactly, so an added field fails a test rather than shipping quietly."
+    },
+    "option_n_note": "`option_n` records THAT a numbered option was chosen and, via options_count, out of how many. It deliberately does NOT record WHICH number: the block's option order is content, and the capture-rate question this instrument exists to answer does not need it.",
+    "soak_window": {
+        "length_days": 14,
+        "fixed_at": "2026-08-24",
+        "why_fixed_before_the_window": "A window whose length is chosen after the numbers are in is a window chosen to produce a number. Fixed here, before any capture code ran, per the same pre-registration discipline hook-latency-budget.json records for itself."
+    }
+}