docs: #000037 §20+§17.1 refresh + #000025 Phase 1f closure + cookbook harvest section

Three coordinated doc landings tying together fox's evening #000037 → #000025 closed-loop work (commits f625cac through ff1752c + 8999b55): == #000037 §20 Status — stale prose refresh == §20 said "open · awaiting go/no-go" but file header + TICKETS.md both say "in progress" with phases 0/1/1.b/2 all landed. Refresh §20 to reflect actual state with per-phase commit anchors: - Phase 0 (doc): landed; David review applied per §21 - Phase 1 (pure-function controller): f625cac (arborist/substrate/prometheus.py + test_prometheus.py) - Phase 1.b (gap-close): f9f5ae4 (§14 row 4 Hermes-saturation guard, §13 step 11 falsification-fixture proposal, §15 entropy + memory gates weight-tunable, ESCALATE > QUARANTINE > REJECT priority cascade) - Phase 2 (advisory audit writes): a786d6d (prometheus_audit.py + controller_events sibling table; does NOT enter audit_events.event_hash preimage) - §12 Trigger 2 fired 2026-05-10: divergence variance ratio 0.575 > 0.5 with N=37 — Phase 1 opening is now empirically gate-satisfied per 8999b55 - Phase 3: deliberately NOT landed; replaced with read-only dry-run simulator surfacing five design findings (see §22) - Closed-loop signal: 40 corpus-derived 5F fixtures harvested per ff1752c == #000037 §17.1 — "78 atomic claim-pack records" → 92 == §17.1 said "#000031 Phase 2 has 78 atomic claim-pack records that max out at ANCHOR-WARRANTED". #000031 closed at 92 records (78 was an interim count during Phase 2). Refresh with the journey (78 → 92) + AUTOCOUNT-tagged via db-where metric so future drift fires immediately. Note the resolution context: all 92 now resolve via 74 citation-aliases + 13 term-aliases under #000031 Phase 2.5 + B-1 + B-2. The unconscious sweep drains the ANCHOR-WARRANTED → EVIDENCE-WARRANTED promotion backlog (derivations.proof_blob rows still need computation even on resolved chains). == #000025 §11 Status — Phase 1f closure note == #000025 file header lists Phase 1a/1b.2/1c/1d/1e but the §11 Status section was frozen at "Open · awaiting go/no-go" — a two-versions-old prose snapshot. Refresh with the per-phase landing trail + add a Phase 1f section for the corpus-derived falsification harvest that ff1752c shipped: - Phase 1f closes the controller → 5F battery loop fox designed in #000037 §3 ("Divergence → candidate falsification fixture") - bench/scripts/harvest_falsification_proposals.py reads qa.db, stratifies top-20-by-cache_key per audit_mode (HYBRID + UNGROUNDED), writes the 41-line JSONL pack (1 _meta + 40 fixtures, AUTOCOUNT-tagged via fixture-rows) - Every fixture row carries _harvest_meta with cache_key, witness_divergence at harvest time, audit_mode_at_harvest, harvest_threshold, source_ticket: "#000037 §13 step 11" - 5F battery exercises them every test run; test_5f_falsification_harvested_pack_runs_clean asserts error_detection_rate == 1.0 by construction (every harvested row IS a falsification) - Self-amplifying — coverage grows with corpus, not with hand-curation Listed open items (§10.11 / 10.13 / 10.14) preserved verbatim from file header so the index claim "still open" stays in sync. == Cookbook: new "Adjacent: live-corpus → bench-fixture harvest" section == New section in warrant-substrate-cookbook.md between "Re-running the substrate build" and "References" documenting the harvest pattern. Three discipline patterns reused from the textbook substrate noted explicitly: 1. Attribution metadata on every derived artifact (same shape as derivations.proof_blob carrying inclusion proofs back to source chunks) 2. Determinism via sort-and-cap (same shape as citation-alias cascade's "top 5 AND-join then top 3 OR-join" stratification) 3. Pin the metadata contract in tests (same shape as the cookbook appendix's discipline pins — silent regression becomes loud test failure) Plus the reusable recipe for any controller emitting Proposal records: define the dataclass, write a harvester filtering + stratifying, pin metadata in tests, wire a make target. References section gets four new entries pointing to #000037, #000025, the harvest script, and the fixture pack. == Drift caught + refreshed during this commit == While editing the cookbook, the AUTOCOUNT regression test (from fc5ba50 / 6c6defb) caught two stale counts from fox's in-flight prometheus work: - test_prometheus.py: 36 → 42 (fox's uncommitted +6 for Phase 1.c sweep weight profile) - test_prometheus_audit.py: 14 → 17 (fox's uncommitted +3) Refreshed both inline + in the test/code-density table (prometheus row test LOC also bumped 804 → 955 to match wc -l). This is the harness firing exactly as designed — fox's uncommitted tests changed live state and my doc claims went stale within minutes. The test message named the file + line + claimed-vs-live, refresh was a 60-second turnaround. == Verification == $ pytest tests/test_doc_counts.py 3 passed in 2.92s $ pytest tests/ -q 2337 passed, 37 skipped in 107.43s Hygiene: only docs/ paths staged. fox's in-flight changes to arborist/qa/runner.py + arborist/substrate/prometheus.py + tests/test_prometheus.py + tests/test_prometheus_audit.py remain in their working tree, untouched by this commit.
2026-05-10 18:04:24 -04:00 · 2026-05-10 18:04:24 -04:00 · 3b301267ff
commit 3b301267ff
parent 8999b55c78
3 changed files with 183 additions and 20 deletions
--- a/docs/tickets/ticket-000025-5f-battery.md
+++ b/docs/tickets/ticket-000025-5f-battery.md
@ -578,10 +578,43 @@ against real workload), §10.14 (threshold handoff to #000012).

 ## 11. Status

-**Open · awaiting go/no-go.** Larger surface than #000023 / #000024
-(five sub-batteries vs three / four remaining). Recommended
-sequencing: ship Phase 1a in one push (seed fixtures + non-stub
-runners), Phase 1b expands.
+**As of 2026-05-10: in progress.**
+
+- **Phase 1a** (Function + Finetuning seed): landed.
+- **Phase 1b.2** (Falsification curated fixtures): landed.
+- **Phase 1c** (5F synthetic side expanded 10 → 30): landed
+  2026-05-09 per #000025 §10.11 cross-reference.
+- **Phase 1d** (synthetic side 30 → 50): landed 2026-05-09.
+- **Phase 1e** (falsification motif coverage 50 → 62 covering
+  every documented failure-motif tag): landed 2026-05-10, closes
+  §10.12.
+- **Phase 1f** (corpus-derived falsification harvest): landed
+  2026-05-10 in commit `ff1752c`. Closes the controller → battery
+  loop fox designed in #000037 §3: Phase 1 controller emits
+  `FalsificationFixtureProposal` records from high-divergence
+  `providence_cache` rows (witness_divergence ≥ 0.5);
+  `bench/scripts/harvest_falsification_proposals.py` stratifies
+  top-20-by-cache_key per audit_mode and writes
+  `bench/fixtures/5f/falsification-harvested-v1.jsonl`
+  (<!--AUTOCOUNT:fixture-rows:bench/fixtures/5f/falsification-harvested-v1.jsonl-->41<!--/AUTOCOUNT-->
+  rows = 1 `_meta` header + 40 fixtures, stratified 20 HYBRID +
+  20 UNGROUNDED). The 5F battery exercises them every test run;
+  `test_5f_falsification_harvested_pack_runs_clean` asserts
+  `error_detection_rate == 1.0` (every harvested row IS a
+  falsification by construction). Self-amplifying coverage —
+  the pack grows with the corpus + controller, not with
+  hand-curation. Every fixture row carries `_harvest_meta` with
+  source `cache_key`, `witness_divergence` at harvest time,
+  `audit_mode_at_harvest`, `harvest_threshold`, and
+  `source_ticket: "#000037 §13 step 11"` for full traceability.
+  Regenerate via `make bench-5f-harvest`.
+
+**Still open** (carried in file header):
+- §10.11 — real shard finetuning chains
+- §10.13 — Feedback Loop latency/efficiency calibrated to real workload
+- §10.14 — threshold handoff to #000012
+
+Larger surface than #000023 / #000024 (five sub-batteries vs three / four).

 The synthesis statement that justifies 5F's existence as a distinct
 axis:
--- a/docs/tickets/ticket-000037-prometheus-sigma-recursive-falsification-controller.md
+++ b/docs/tickets/ticket-000037-prometheus-sigma-recursive-falsification-controller.md
@ -999,10 +999,15 @@ These are the right guardrails for the pure-function layer.

 ### 17.1 Warrant-promotion integration (#000031)

-#000031 Phase 2 has 78 atomic claim-pack records that max out at
-`ANCHOR-WARRANTED` until cited textbooks are ingested and
-`derivations.proof_blob` rows are computed. The unconscious sweep
-(§3.1, Target B) drains exactly this backlog. Controller scoring
+#000031 closed at <!--AUTOCOUNT:db-where:documents?source_type=claim_pack-->92<!--/AUTOCOUNT-->
+atomic claim-pack records (started at 78, reached 92 by 2026-05-09
+under Phase 2.5). All 92 records now resolve to at least one
+textbook substrate via the 74 citation-aliases (#000041) + 13
+term-aliases (#000042) shipped under #000031 Phase 2.5 + B-1 + B-2.
+The unconscious sweep (§3.1, Target B) drains the residual
+ANCHOR-WARRANTED → EVIDENCE-WARRANTED promotion backlog —
+`derivations.proof_blob` rows still need to be computed even on
+records that have resolved warrant chains. Controller scoring
 includes `WarrantPromotionGain_i` (§4.1) so a branch that turns
 many anchor-only claims into evidence-warranted claims is
 high-value even if it doesn't move bench scores.
@ -1103,16 +1108,58 @@ the decision to the Phase 2 ticket.

 ## 20. Status

-Phase 0 (this ticket): open · awaiting go/no-go · David review
-2026-05-10 applied.
+**As of 2026-05-10:** in progress.

-When fox approves: this ticket flips to `in progress · Phase 0
-landed in commit <sha>` with no code change — the doc itself is
-the deliverable.
+- **Phase 0** (this ticket): landed; David review (`response_ticket-
+  000037-...txt`, 2026-05-10 06:37 Asia/Kuala_Lumpur) applied per §21.
+- **Phase 1** (pure-function controller): landed in `f625cac` —
+  `arborist/substrate/prometheus.py` (893 LOC) implementing §13
+  steps 1-9 + 12, §5 numerically-stable softmax fork-selection, §6
+  eight-class hard-veto order, §7 Kelly-bounded allocation + four
+  safety guards, §7.1 EMA-smoothed difficulty update, §14 exception-
+  matrix dispatch, §15 three weight profiles. Test surface:
+  `tests/test_prometheus.py` (36 tests including the §16.2
+  named-contract pins).
+- **Phase 1.b** (gap-close pass): landed in `f9f5ae4` — §14 row 4
+  Hermes-saturation guard (`utilization >= budget` → DEFERRED +
+  HERMES_SATURATED advisory event), §13 step 11 falsification-fixture
+  proposal emission, §15 entropy + memory gates moved to
+  weight-tunable `ControllerWeights` fields, veto-class cascade
+  hardening (ESCALATE > QUARANTINE > REJECT priority pins).
+- **Phase 2** (advisory audit writes): landed in `a786d6d` —
+  `arborist/substrate/prometheus_audit.py` (200 LOC) + 54 LOC store
+  integration. `controller_events` sibling table with three event
+  kinds (`controller_decision`, `controller_difficulty`,
+  `controller_budget_allocation`), idempotent on
+  `(event_kind, body_hash)` UNIQUE constraint. **Does not enter
+  `audit_events.event_hash` preimage** — audit chain semantics
+  unaffected by advisory writes. Test surface:
+  `tests/test_prometheus_audit.py` (14 tests).
+- **§12 Trigger 2 fired 2026-05-10**: divergence variance ratio
+  0.575 > 0.5 with N=37 samples — Phase 1 opening is now backed
+  by empirical gate satisfaction, not speculative implementation.
+- **Phase 3** (sleep-sweep scheduler): **deliberately not landed**.
+  Replaced with a read-only dry-run simulator
+  (`bench/scripts/prometheus_sigma_sweep_dryrun.py`) driving the
+  Phase 1 controller against real shard data; surfaces five design
+  findings (chunk-size dominates Kelly; capital_cost split needed;
+  τ_qa per audit_mode; Target B canonical-shape headline; quarantined-
+  row veto exercises end-to-end). See §22 for details. The dry-run
+  is the calibration substrate the eventual Phase 3 implementation
+  will reference.
+- **Closed-loop signal**: 40 corpus-derived 5F fixtures harvested
+  from Phase 1 `FalsificationFixtureProposal` records via
+  `bench/scripts/harvest_falsification_proposals.py` →
+  `bench/fixtures/5f/falsification-harvested-v1.jsonl` (commit
+  `ff1752c`). Stratified 20 HYBRID + 20 UNGROUNDED, top-by-cache_key
+  for determinism. The 5F battery exercises them every test run
+  with `error_detection_rate == 1.0` (every harvested row IS a
+  falsification by construction).

-Phase 1 opens when §12 trigger fires. Phase 2 (advisory audit
-writes) and Phase 3 (unconscious sweep) gate sequentially on
-Phase 1 having landed plus their own §12 trigger renewal.
+Phase 3 (the actual scheduler) gates on a renewed §12 trigger
+plus a Phase 3 ticket. Until then the closed loop runs at
+test-suite cadence: Phase 1 emits proposals → harvester writes
+fixtures → 5F battery exercises → coverage grows with the corpus.

 ---

--- a/docs/warrant-substrate-cookbook.md
+++ b/docs/warrant-substrate-cookbook.md
@ -432,6 +432,80 @@ arborist warrant-resolve --use-aliases --write
 #       + 13-term-alias state (counts as of 2026-05-10)
 ```

+## Adjacent: live-corpus → bench-fixture harvest
+
+The textbook substrate above gives the cookbook a static surface
+— 18 hand-curated, license-validated, deterministically ingested
+sources. A parallel pattern landed 2026-05-10 in `ff1752c` under
+#000037 → #000025 that does the opposite: **grow bench fixtures
+from live corpus state, not hand-curation**. Worth documenting
+here because the discipline patterns (attribution metadata,
+determinism via stratification, fail-loud regen) carry across
+both surfaces.
+
+### Pattern
+
+```
+#000037 Phase 1 controller (arborist/substrate/prometheus.py)
+  emits FalsificationFixtureProposal records when
+  witness_divergence (n_unverified / n_quotes) >= 0.5
+    ↓
+bench/scripts/harvest_falsification_proposals.py reads qa.db,
+  filters live providence_cache rows, stratifies top-20 per
+  audit_mode (HYBRID + UNGROUNDED), writes JSONL
+    ↓
+bench/fixtures/5f/falsification-harvested-v1.jsonl
+  (<!--AUTOCOUNT:fixture-rows:bench/fixtures/5f/falsification-harvested-v1.jsonl-->41<!--/AUTOCOUNT--> rows = 1 _meta header + 40 fixtures)
+    ↓
+tests/test_bench_batteries.py:test_5f_falsification_harvested_pack_*
+  exercises every test run; error_detection_rate == 1.0
+  by construction (every harvested row IS a falsification)
+```
+
+### Three discipline patterns reused from the textbook substrate
+
+1. **Attribution metadata on every derived artifact** — same
+   discipline as `derivations.proof_blob` carrying inclusion
+   proofs back to source chunks. Each harvested fixture row
+   carries `_harvest_meta` with `cache_key`,
+   `witness_divergence` at harvest time, `audit_mode_at_harvest`,
+   `harvested_from`, `harvest_threshold`, and `source_ticket:
+   "#000037 §13 step 11"`. Regeneration provenance auditable.
+
+2. **Determinism via sort-and-cap** — same discipline as
+   citation-alias cascade tuning's "top 5 AND-join then top 3
+   OR-join" structure. The harvester sorts top-by-cache_key per
+   audit_mode bucket + caps at 20 per mode. Same `qa.db` state →
+   byte-identical fixture file across regenerations.
+
+3. **Pin the metadata contract in tests** —
+   `test_5f_falsification_harvested_pack_runs_clean` asserts
+   every row carries the meta. Same shape as the cookbook
+   appendix's discipline pins: a future regeneration that drops
+   attribution fires a test failure, not a silent regression.
+
+### Where to use this pattern
+
+Anywhere a controller emits Proposal records that should become
+real fixtures. The recipe:
+
+- Define the Proposal dataclass on the controller side (see
+  `arborist/substrate/prometheus.py:FalsificationFixtureProposal`).
+- Write a harvester script under `bench/scripts/` that reads
+  the source state (`qa.db`, `~/.arborist/shards/*.db`),
+  filters + stratifies, and emits a JSONL fixture pack.
+- Pin the metadata contract in `tests/test_bench_batteries.py`
+  (or the equivalent battery test file).
+- Wire `make <battery>-harvest` so regen is one command.
+
+The 18-textbook substrate is the static foundation; the
+harvested-fixture pattern is how that foundation gets exercised
+against current corpus state on every test run. Both live in
+this cookbook because the discipline rules — attribution,
+determinism, fail-loud — are identical.
+
+---
+
 ## References

 - `#000031` — surface-ingest cited textbooks
@ -441,6 +515,15 @@ arborist warrant-resolve --use-aliases --write
  proprietary candidates: Hilbert-Ackermann 1928, Mendelson,
  Enderton, Jech, Goldstein, Barendregt, Stanley, Brualdi, Knuth;
  the substrate substitution pattern is the workaround)
+- `#000037` — Prometheus-Σ recursive falsification controller;
+  source of the `FalsificationFixtureProposal` records the
+  harvest pattern consumes
+- `#000025` — 5F battery; sink for the harvested falsification
+  fixtures (§11 Phase 1f)
+- `bench/scripts/harvest_falsification_proposals.py` — harvester
+  script (266 LOC) reading qa.db + emitting stratified fixtures
+- `bench/fixtures/5f/falsification-harvested-v1.jsonl` —
+  current harvested pack (regenerate via `make bench-5f-harvest`)
 - `bench/results/full-warrant-resolution-2026-05-10.md` — the
  bench journal that closed 84% → 100%
 - `bench/fixtures/textbooks/manifest-v1.jsonl` — license-validated
@ -542,7 +625,7 @@ than waiting for bench-time STRICT-rate drift to surface it.
 - `tests/test_pi_star_protocol_and_registry.py` — <!--AUTOCOUNT:tests:tests/test_pi_star_protocol_and_registry.py-->21<!--/AUTOCOUNT--> tests for
  `PiStar` Protocol contract + registry mutation discipline
  (no public unregister; `name@version` content-pinned).
- `tests/test_prometheus.py` — <!--AUTOCOUNT:tests:tests/test_prometheus.py-->36<!--/AUTOCOUNT--> tests for the Prometheus-Σ
+- `tests/test_prometheus.py` — <!--AUTOCOUNT:tests:tests/test_prometheus.py-->42<!--/AUTOCOUNT--> tests for the Prometheus-Σ
  recursive-falsification controller (#000037 Phase 1, commits
  `f625cac` + `f9f5ae4`). Verifier-style discipline (NOT
  calculator pattern — it's a state machine, not closed-form
@ -551,7 +634,7 @@ than waiting for bench-time STRICT-rate drift to surface it.
  Hermes-saturation guard, §13 step 11 falsification-fixture
  proposal, §15 weight-tunable entropy + memory gates,
  ESCALATE > QUARANTINE > REJECT priority ordering.
- `tests/test_prometheus_audit.py` — <!--AUTOCOUNT:tests:tests/test_prometheus_audit.py-->14<!--/AUTOCOUNT--> tests for the
+- `tests/test_prometheus_audit.py` — <!--AUTOCOUNT:tests:tests/test_prometheus_audit.py-->17<!--/AUTOCOUNT--> tests for the
  `controller_events` sibling table (#000037 Phase 2, commit
  `a786d6d`). Pins migration idempotency, no-chain-mutation
  invariant (advisory writes never enter `audit_events.event_hash`
@ -594,8 +677,8 @@ than waiting for bench-time STRICT-rate drift to surface it.
 | qa/prompts.py | 153 | 207 (<!--AUTOCOUNT:tests:tests/test_qa_prompts.py-->20<!--/AUTOCOUNT--> tests) | 1.35 |
 | concepts/extract.py | 288 | 278 (<!--AUTOCOUNT:tests:tests/test_concepts_extract.py-->20<!--/AUTOCOUNT--> tests) | 0.97 |
 | textbooks_manifest.py | 236 | 463 (<!--AUTOCOUNT:tests:tests/test_textbooks_manifest.py-->43<!--/AUTOCOUNT--> tests) | 1.96 |
-| substrate/prometheus.py | 893 | 804 (<!--AUTOCOUNT:tests:tests/test_prometheus.py-->36<!--/AUTOCOUNT--> tests) | 0.90 |
-| substrate/prometheus_audit.py | 200 | 388 (<!--AUTOCOUNT:tests:tests/test_prometheus_audit.py-->14<!--/AUTOCOUNT--> tests) | 1.94 |
+| substrate/prometheus.py | 893 | 955 (<!--AUTOCOUNT:tests:tests/test_prometheus.py-->42<!--/AUTOCOUNT--> tests) | 1.07 |
+| substrate/prometheus_audit.py | 200 | 477 (<!--AUTOCOUNT:tests:tests/test_prometheus_audit.py-->17<!--/AUTOCOUNT--> tests) | 2.39 |

 Pattern: small contract-defining modules (weights, prompts,
 warrant_chain) get 1.4-3.6× test LOC because they're foundation;