Merge pull request #546 from AssetOverflow/docs/reconcile-current-state-2026-06-03

docs: reconcile current-state claims after GSM8K + sensorium progress

CLAUDE.md

@@ -58,6 +58,8 @@ Allowed sites:
 - `ingest/gate.py` for raw input injection.
 - `language_packs/compiler.py` and vocabulary construction.
 - `algebra/versor.py` for algebra-owned sandwich closure.
+- `sensorium/*/canonical.py` and pack-governed modality compiler construction
+  boundaries for pinned signal canonicalization and quantization.
 
 Forbidden sites:
 
@@ -104,14 +106,23 @@ runtime path.  Vault recall is exact and deterministic.
 - `calibration/*` — bounded replay-based calibration.
 - `docs/runtime_contracts.md` — response, telemetry, memory, identity, and testing contracts.
 
-### GSM8K math comprehension substrate (sealed; serving `6/44/0`, wrong=0 — moves only via ratified PRs)
+### GSM8K math comprehension substrate (sealed; serving `7/43/0`, wrong=0 — moves only via ratified PRs)
 
 - `core/reliability_gate/` — calibrated-learning ledger + gate (ADR-0175): `ClassTally` counts, `conservative_floor` (one-sided Wilson, N_MIN=10), θ ceilings.
-- `generate/derivation/` — the comprehension composer: `extract.py` (lexeme quantity extraction, EX-1/4/5 + function-word unit filter), `clauses.py` (GB-1 segmentation), `compose.py` (GB-2a list-sum + GB-3a clause-scoped referent guard), `accumulate.py` (GB-3b.1 single-referent gain/loss chaining), `multistep.py`/`search.py` (bounded search), `verify.py` (the wrong=0 self-verification gate: grounding ∧ cue ∧ unit ∧ completeness ∧ uniqueness).
+- `generate/derivation/` — the comprehension composer: `extract.py` (lexeme quantity extraction, EX-1/4/5 + function-word unit filter), `clauses.py` (GB-1 segmentation), `compose.py` (GB-2a list-sum + GB-3a clause-scoped referent guard), `accumulate.py` (GB-3b.1 single-referent gain/loss chaining), `goal_residual.py` (ADR-0207 R4 goal-residual production), `multistep.py`/`search.py` (bounded search), `verify.py` (the wrong=0 self-verification gate: grounding ∧ cue ∧ unit ∧ completeness ∧ uniqueness).
 - `generate/cue_precision/` — `(cue, op, unit_shape)` reliability ledger + trainer (ADR-0177 CP-1/CP-2a); inert (consulted by no serving/gate path yet).
-- `evals/gsm8k_math/` — `train_sample/` (real GSM8K, the capability metric), `practice/` (sealed attempt-and-eliminate lane + ADR-0163-F additive set), `confusers/` (ADR-0163-F2 discrimination probe — scored by `wrong→0` + pair-consistency, NOT flip-count).
+- `evals/gsm8k_math/` — `train_sample/` (real GSM8K dev sample, currently 7 correct / 43 refused / 0 wrong), `practice/` (sealed attempt-and-eliminate lane + ADR-0163-F additive set), `confusers/` (ADR-0163-F2 discrimination probe — scored by `wrong→0` + pair-consistency, NOT flip-count).
 - `scripts/verify_lane_shas.py`, `scripts/generate_claims.py --check` — the serving-frozen gate (pinned eval-lane SHAs + `CLAIMS.md`).
 
+### Sensorium / modality compiler substrate (parallel, afferent gates; no broad capability claim)
+
+- `sensorium/compiler/` — shared compiler law for content-addressed afferent compilation units, canonical deltas, local arenas, and trace-safe merge hashes.
+- `sensorium/audio/` + `sensorium/adapters/audio.py` — `audio_core_v1`, deterministic audio compiler substrate, gate closed by default.
+- `sensorium/vision/` + `sensorium/adapters/vision.py` — `vision_core_v1`, tile-first deterministic visual compiler substrate over synthetic eval fixtures, gate closed by default.
+- `sensorium/environment/` — ADR-0208 observation-frame contract for bundles of already-compiled afferent units; not late fusion and not a mutable world model.
+- `sensorium/sensorimotor/` + `sensorium/adapters/sensorimotor.py` — ADR-0209 afferent proprioception/contact/action-result evidence substrate; no decode path.
+- `sensorium/registry.py::decode*` + `AuthorityToken` / `EfferentGate` — ADR-0198 fail-closed efferent governance path. This is not a ratified motor decoder or actuator interface; no real action emission is claimed.
+
 ## Efficiency and Performance Doctrine
 
 Performance is part of correctness for this project because slow feedback hides
@@ -327,7 +338,7 @@ implicated rather than one.
    things the ADR says that didn't actually ship.
 
 2. **Before merging a stacked PR sequence into main.**  When 2+ PRs
-   stack (PR #420 stacked on #416, #423 stacked on #420), the
+   stack (PR #420 stacked on #416, PR #423 stacked on #420), the
    review-each-PR-individually pattern misses cross-PR consistency
    issues.  Audit the whole stack as one unit before any merge.
 

docs/claims_ledger.md

@@ -11,13 +11,16 @@ defended.
 > discipline applies to how we describe ourselves, not only to what the engine
 > outputs. Overclaiming to hit a deadline is failure, not success.
 
-- **Base commit:** `c058d96` (origin/main).
-- **Verification date:** 2026-06-02.
+- **Base commit:** `3a72d69` (origin/main before this reconciliation branch).
+- **Verification date:** 2026-06-03 / 2026-06-04 UTC merge window.
 - **Reproduce the machine-derived rows:**
   ```bash
-  uv run core capability ledger          # Tier-1 status (audit-passed ×3; NO expert)
-  uv run python scripts/verify_lane_shas.py   # Tier-2 pinned lane SHAs
-  uv run python scripts/generate_claims.py --check   # CLAIMS.md is current
+  uv run core capability ledger                  # Tier-1 status (audit-passed ×3; NO expert)
+  uv run python scripts/verify_lane_shas.py      # Tier-2 pinned lane SHAs
+  uv run python scripts/generate_claims.py --check   # CLAIMS.md is current, when regenerated from this ledger
+  uv run python -m pytest tests/test_vision_eval_gates.py -q
+  uv run python -m pytest tests/test_audio_compiler.py tests/test_audio_crdt_merge.py tests/test_audio_sensorium_mount.py -q
+  uv run python -m pytest tests/test_observation_frame_contract.py tests/test_sensorimotor_contract.py -q
   ```
 
 ---
@@ -32,7 +35,7 @@ material that blurs them is a credibility risk.
 |---|---|---|---|---|---|
 | **A** | **Sealed real GSM8K test** (1,319 cases, HuggingFace `openai/gsm8k` test split) | **0 correct / 0 wrong / 1,319 refused** | **Yes** | No | `ADR-0119.7`; ciphertext `evals/gsm8k_math/holdouts/v1/cases.jsonl.age` |
 | **B** | **CORE-authored synthetic "public" split** (150 cases, rule-based, written to exercise the grammar) | **150 correct / 0 wrong / 0 refused** (rate 1.0) | **No** | No | `evals/gsm8k_math/baselines/comparison_v1.json`, `frontier.json` |
-| **C** | **Real `train_sample`** (50-case dev sample of real GSM8K) | **6 correct / 44 refused / 0 wrong** | Yes (sample) | No | `evals/gsm8k_math/train_sample/v1/report.json` |
+| **C** | **Real `train_sample`** (50-case dev sample of real GSM8K) | **7 correct / 43 refused / 0 wrong** | Yes (sample) | No | `evals/gsm8k_math/train_sample/v1/report.json` |
 | **D** | **Expert-promotion composite gate** (CORE-authored lanes B1/B2/B3) | **185/185 + 14/14 + 40/40 + 50/50, wrong=0** | **No** | Yes (but reverted — §2) | `ADR-0131.4`; `core/capability/composite_math_gate.py` |
 
 **Per-number honest framing:**
@@ -62,12 +65,12 @@ material that blurs them is a credibility risk.
   frontier LLMs (Claude 3.5 Sonnet 96.4%, GPT-4 92.0%, Gemini 1.5 Pro 90.8%) are
   scored on *real* GSM8K. **Never cite B as "CORE scores 100% on GSM8K."** The
   qualitative differentiator is `wrong=0`, not the rate.
-- **(C) The dev-sample reality.** 6/44/0 on 50 real cases — the lane's
-  `exit_criterion` (`correct_min: 10`) is **NOT met**. A stricter candidate-graph
-  coverage probe reports **4/46/0** on the same 50 cases
-  (`train_sample_coverage_report.json`); both preserve `wrong=0`. The two readers
-  differ on 2 fast-path cases. The probe's movement (`3→4`) is precisely what
-  reverted the expert claim (§2).
+- **(C) The dev-sample reality.** `7/43/0` on 50 real cases — the lane's
+  `exit_criterion` (`correct_min: 10`) is **NOT met**. The latest lift is the
+  ADR-0207 §5 R4 goal-residual production (`generate/derivation/goal_residual.py`),
+  which moves train-sample case `0037` while preserving `wrong=0`. A stricter
+  candidate-graph coverage probe is a separate diagnostic lane and must not be
+  conflated with the serving train-sample report.
 - **(D) CORE-authored, and currently NOT conferring expert.** All four B-lanes are
   written and scored by CORE; none is external GSM8K. ADR-0131.4 replaced the
   original real-GSM8K gate (`correct_rate ≥ 0.60`) with this composite; ADR-0131.5
@@ -130,25 +133,32 @@ claim when its evidence moved.** Full record: `ADR-0200`.
 | **Refuse-rather-than-guess / `wrong = 0`** | typed refusal across all four GSM8K numbers; `verify.py` gate | A/B/C/D all `wrong=0` |
 | Exact CGA recall (no ANN / HNSW / cosine) | `vault/store.py`; `docs/Yellowpaper.md` recall section | `uv run core test --suite algebra` |
 | Safety pack: add-but-never-remove, fail-closed on load | `packs/safety/core_safety_axes_v1.json`; `ADR-0029` | `uv run core test --suite smoke` |
-| On-device, non-LLM, no sampling/gradient/tokenizer | architecture; `README.md` "Third Door" | — |
+| Sensorium compiler law: compiled afferent units are content-addressed, replayable, and merge by deterministic key | `sensorium/compiler/*`; audio/vision/sensorimotor tests | see §4 reproduce commands |
+| On-device, non-LLM, no sampling/gradient/tokenizer in the CORE runtime path | architecture; `README.md` "Third Door" | — |
 
 ---
 
-## 4. Multimodal status — text only is *capability*; the rest is substrate or proposal
+## 4. Multimodal status — distinguish capability from substrate
 
 The patent application's "multimodal" title is broader than what is demonstrated
-in-repo. **External materials must not imply working vision or motor.**
+as end-user capability in-repo. **External materials must not imply working
+open-domain vision, audio understanding, robotics, or autonomous action.** The
+current truth is: text is capability; audio/vision/sensorimotor are deterministic
+afferent substrates with gates and synthetic evals; motor/efferent output is
+fail-closed governance scaffolding only.
 
-| Modality | Status | Evidence |
-|---|---|---|
-| **Text** (English, Hebrew, Koine Greek) | **Active capability** | `sensorium/adapters/text.py`; language packs |
-| **Audio** | **Substrate landed, gate CLOSED (no capability claim)** | `sensorium/audio/*`; `make_audio_pack(gate_engaged=False)`; determinism + order-invariant CRDT merge + no-PCM-in-trace all gate-tested |
-| **Vision** (`ADR-0197`) | **Proposed only** — no code, no pack, no eval | `Modality.VISION` enum exists; no `sensorium/vision/` |
-| **Motor** (`ADR-0198`) | **Proposed only** — "design spike, no implementation" | registry has no `decode()` path; `persona/motor.py` is an *internal CGA screw motion*, not an actuator |
+| Surface | Status | Evidence | Honest claim boundary |
+|---|---|---|---|
+| **Text** (English, Hebrew, Koine Greek) | **Active capability** | `sensorium/adapters/text.py`; language packs | Working textual modality. |
+| **Audio** (`audio_core_v1`) | **Substrate landed, gate CLOSED by default** | `sensorium/audio/*`, `sensorium/adapters/audio.py`, `packs/audio/audio_core_v1/`, `tests/test_audio_*` | Deterministic compiler substrate; no broad audio-understanding capability claim. |
+| **Vision** (`vision_core_v1`) | **Substrate landed, gate CLOSED by default** | `sensorium/vision/*`, `sensorium/adapters/vision.py`, `packs/vision/vision_core_v1/`, `evals/vision_sensorium/`, `tests/test_vision_eval_gates.py` | Tile-first deterministic visual compiler over synthetic fixtures; no open-domain vision claim. |
+| **Environmental loop** (`ADR-0208`) | **Afferent observation-frame contract landed** | `sensorium/environment/frame.py`, `sensorium/environment/harness.py`, `tests/test_observation_frame_*` | Bundles already-compiled afferent units; not late fusion, not a mutable world model. |
+| **Sensorimotor feedback** (`ADR-0209`) | **Afferent substrate landed** | `sensorium/sensorimotor/*`, `sensorium/adapters/sensorimotor.py`, `tests/test_sensorimotor_*` | Proprioception/contact/action-result evidence only; no decoder, actuator driver, trajectory executor, or skill invocation. |
+| **Motor / efferent output** (`ADR-0198`) | **Partially implemented governance spike; action emission still refused unless verdict-governed** | `ModalityRegistry.decode*`, `AuthorityToken`, `DefaultEfferentGate`, `EfferentRefusal` | Registry path exists, but no motor decoder or real actuation capability is claimed. |
 
-Honest line: *text is a working modality; audio is a determinism-proven substrate
-with its capability gate deliberately closed; vision and motor are design
-proposals with no implementation.*
+Honest line: *text is a working modality; audio/vision/sensorimotor are
+replayable afferent compiler substrates with gates and eval fixtures; motor is a
+fail-closed efferent governance surface, not an actuator.*
 
 ---
 
@@ -159,7 +169,7 @@ The reviewer registry has exactly **one** signer: `shay-j`, `domains: ["*"]`,
 `audit_passed_claims` and the (now-quarantined) `math_expert_claims` entry is
 signed by the same identity. This is a real single-point-of-capture a partner may
 probe. The migration toward a multi-reviewer / threshold-signing registry is
-scoped as a Week-4 deliverable (`ADR-02xx`, design-only — no fake signers).
+scoped as a future design deliverable — no fake signers.
 
 ---
 
@@ -170,24 +180,29 @@ scoped as a Week-4 deliverable (`ADR-02xx`, design-only — no fake signers).
 - ❌ "CORE is externally validated on GSM8K." → The expert gate rests on
   **CORE-authored** lanes; the real external number is A = `0/0/1319`.
 - ❌ "CORE scores ~100% on GSM8K." → That is B, a CORE-authored synthetic split.
-- ❌ "CORE has vision / motor / working multimodal perception." → Audio is
-  substrate-only (gate closed); vision/motor are proposals.
+- ❌ "CORE has broad working multimodal perception." → Audio, vision, and
+  sensorimotor are afferent substrates/eval lanes, not demonstrated open-domain
+  capability.
+- ❌ "CORE can perform motor actions / robotics / vehicle control." → Efferent
+  output is fail-closed governance scaffolding only; no motor decoder/actuator
+  path is ratified.
 - ❌ "`audit-passed` means expert-level capability." → It is claim-shape
   compliance, not raw capability (`ADR-0113`).
 
 ---
 
-## 7. Reconciliation log (proposed — pending operator ratification)
+## 7. Reconciliation log
 
 These reconcile stale artifacts to the truth above. **History-describing**
-artifacts keep their content with a dated "valid-at … auto-reverted … current =
-audit-passed" note (keep the receipt; keep the mismatch-refusal firing).
-**Current-machine-state** artifacts reconcile to the truth. Tracked in `ADR-0200`.
+artifacts keep their content with dated context. **Current-machine-state**
+artifacts reconcile to the truth.
 
 | Artifact | Type | Action |
 |---|---|---|
-| `docs/reviewers.yaml` `math_expert_claims` | history (receipt) | keep entry; add quarantine note; do not re-sign |
+| `docs/claims_ledger.md` | current state | reconciled to `7/43/0`, afferent sensorium substrate landed, and fail-closed efferent governance scaffold |
+| `CLAUDE.md` GSM8K substrate header | current operator instruction | reconcile serving metric to `7/43/0`; keep wrong=0 governor |
+| `evals/gsm8k_math/train_sample/v1/README.md` | current eval summary | reconcile current report and note R4 goal-residual lift |
+| `docs/decisions/ADR-0207-gsm8k-substrate-ratification.md` | ratified decision with evolving execution status | keep ratification baseline; add current-state note after R4 lift |
+| `docs/reviewers.yaml` `math_expert_claims` | history (receipt) | keep entry; add quarantine note separately; do not re-sign |
 | `docs/decisions/ADR-0120-math-expert-ledger-flip.md` | history | header note: valid-at 2026-05-23, auto-reverted, current = audit-passed |
-| `evals/math_expert_claims/v1/expert_claims_math_v1_signed.json` | current state | regenerate → `promote_admitted: false` |
-| `README.md` §"Path to expert" + ledger lines | current state | reconcile narrative to built-attempted-reverted; verify test count |
-| `tests/test_mathlogic_expert_ledger_flip.py`, `tests/test_adr_0120_math_expert_promotion.py` | current state | flip 3 red "is-expert" assertions into fail-closed-revert assertions |
+| `evals/math_expert_claims/v1/expert_claims_math_v1_signed.json` | current state | future regeneration should preserve `promote_admitted: false` until a ratified expert path exists |

docs/decisions/ADR-0207-gsm8k-substrate-ratification.md

@@ -2,6 +2,7 @@
 
 **Status:** Accepted (ratified 2026-06-03)
 **Date:** 2026-06-03
+**Current execution state:** R4 goal-residual production landed after ratification; serving train-sample is now `7/43/0`, wrong=0.
 **Author:** Shay
 **Anchor:** [[thesis-decoding-not-generating]]
 **Consolidates (does not replace):** ADR-0164, ADR-0165, ADR-0174,
@@ -28,7 +29,9 @@ against the gates* (§6), not another design.
 
 ## 2. Verified findings (origin/main `2cb0922`)
 
-Every claim below was reproduced read-only against the tree on 2026-06-03.
+Every claim below was reproduced read-only against the tree on 2026-06-03. This section is the
+ratification baseline. Later execution notes may update metrics without rewriting the baseline
+finding.
 
 **252 ADRs.** The relevant design is on the books, not missing. Actual status lines (cited as
 they stand — these are *not* "design of record" labels until this ADR is ratified):
@@ -61,8 +64,8 @@ Held-hypothesis primitives (`lookback.reevaluate`, `contemplate`) are invoked li
    (imported by `math_candidate_graph.py` at lines 663, 685, 1004) and the overfitting vector.
 2. **The held-hypothesis primitives** — bolted into the candidate-graph flow (above).
 3. **The `derivation/` composer** — built and *already enriched* (see §3), but **disjoint
-   from serving**: `chat/` imports zero derivation modules, and the serving candidate-graph
-   path imports **exactly one** derivation symbol —
+   from serving** at the ratification baseline: `chat/` imports zero derivation modules, and
+   the serving candidate-graph path imports **exactly one** derivation symbol —
    `generate.derivation.product_bridge.resolve_promotable_product`
    (`math_candidate_graph.py:530`, the guarded product bridge behind the current lift). The
    composer proper (`extract`/`clauses`/`compose`/`accumulate`/`multistep`/`search`/`verify`)
@@ -74,10 +77,15 @@ ancestor of the prior HEAD `8327c6b`, −1,038 LOC); the primitives survived. So
 intended topology (**reader = the brain; injectors = emitters**) **was never realized — it
 fragmented.**
 
-**Empirical signature of the unfinished state:** train_sample **6/44/0** (reproduced live);
-sealed holdout **0/0/1319** (cited per `docs/claims_ledger.md` / ADR-0119.7 ciphertext — not
-CI-reproducible). The live regex path solves a few train surface forms and transfers to
-~nothing unseen. Safe failure (refuses, never wrong), but narrow-and-non-transferring.
+**Empirical signature of the unfinished baseline:** train_sample **6/44/0** (reproduced live at
+ratification); sealed holdout **0/0/1319** (cited per `docs/claims_ledger.md` / ADR-0119.7
+ciphertext — not CI-reproducible). The live regex path solves a few train surface forms and
+transfers to ~nothing unseen. Safe failure (refuses, never wrong), but narrow-and-non-transferring.
+
+**Current execution note (post-ratification).** ADR-0207 §5 R4 execution added
+`generate/derivation/goal_residual.py` and a guarded promotion bridge for residual-to-goal
+readings. This moves train_sample to **7/43/0**, still wrong=0. The lift is execution of this ADR,
+not a new architecture and not a relaxation of the §4 freeze.
 
 **Root cause of the wasted weeks is process, not architecture:** five `Proposed`/partial ADRs
 never ratified or driven to completion, with the deprecated fallback left live.
@@ -130,17 +138,18 @@ Execute these three levers — listed by role, not by sequence (lever 1 explains
 wire is the *last*, trivial step and composition the real first work):
 
 1. **WIRING — trivial, and already done for the safe shapes.** The wire pattern
-   (`product_bridge` at `:530`: `resolve(text) → CandidateGraphResult | fall-through`) is
-   proven, but it generalizes nothing: `product_bridge` promotes only via a two-token target
-   whitelist (`money`+`make/earn`, `weight`+`total/move`), hardcoded to 0003/0021. Extending
-   it is **not** the lowest-risk first step. Verified on the tree (2026-06-03): the general
-   composer (`resolve_pooled`) refuses **all ten** R1/R4/R5/R6 corpus positives — R1/R4 build
-   zero candidates, R5 builds only wrong ones — so wiring it yields **+0 correct**; and it
-   wrong-commits `0016 → 510`, so wiring it wholesale is a live `wrong=0` regression that
-   disagreement does **not** catch. Therefore WIRING and COMPOSITION are the **same task**:
-   the real first work is composer production for each target shape, each behind a structural
-   promotion gate (op-class/target proof via `extract_target` + `target_units`, not
-   disagreement alone). The wire is the last, trivial step once a shape builds-and-certifies.
+   (`resolve(text) → CandidateGraphResult | fall-through`) is proven, but it generalizes
+   nothing by itself. The initial `product_bridge` promoted only via narrow target proof for
+   0003/0021; the post-ratification `goal_residual` bridge promotes only goal-language +
+   residual-question + same-referent licensed-progress chains. Extending either bridge is
+   **not** the lowest-risk first step. Verified on the tree (2026-06-03): the general composer
+   (`resolve_pooled`) refuses **all ten** R1/R4/R5/R6 corpus positives — R1/R4 build zero
+   candidates, R5 builds only wrong ones — so wiring it wholesale yields **+0 correct** and
+   wrong-commits `0016 → 510`, a live `wrong=0` regression that disagreement does **not** catch.
+   Therefore WIRING and COMPOSITION are the **same task**: the real first work is composer
+   production for each target shape, each behind a structural promotion gate (op-class/target
+   proof via `extract_target` + `target_units`, not disagreement alone). The wire is the last,
+   trivial step once a shape builds-and-certifies.
 2. **COMPOSITION (ADR-0178) — the actual wall.** Complete the derivation composer's
    multi-step assembly: which quantities group, via which ops, in what order (the R4/R5/R6
    lever). This is the genuine research risk; extraction already feeds it.
@@ -159,7 +168,7 @@ generalizing). Never a per-shape branch.
 
 Every step must hold all of:
 
-1. train_sample stays **`wrong=0`** (6/44/0 or better — `correct` may only rise).
+1. train_sample stays **`wrong=0`** (current floor: 7/43/0; `correct` may only rise).
 2. The no-reference `<N> times` hazard (composition-capability-scope §9) stays **refused**.
 3. No partial graph admits while any source quantity is unbound (the completeness leg).
 4. Progress is measured on the composition-typed **validation sub-corpus**

evals/gsm8k_math/train_sample/v1/README.md

@@ -52,20 +52,28 @@ Each case in `cases.jsonl` preserves:
 
 ## Current Report
 
-`report.json` currently records **6 correct / 44 refused / 0 wrong**. The
+`report.json` currently records **7 correct / 43 refused / 0 wrong**. The
 ADR-0126 exit criterion remains unmet (`correct >= 10`, `wrong == 0`), so the
 runner still exits nonzero.
 
-The two most recent lifts are ADR-0195 product-promotion cases:
+The current correct set is:
 
 - `gsm8k-train-sample-v1-0003` — complete revenue product
   (`48 boxes x 24 erasers x $0.75`).
+- `gsm8k-train-sample-v1-0014` — fast-path serving case.
+- `gsm8k-train-sample-v1-0018` — fast-path serving case.
 - `gsm8k-train-sample-v1-0021` — complete total-work product
   (`15 pounds x 10 reps x 3 sets`).
+- `gsm8k-train-sample-v1-0024` — comparative product control retained green.
+- `gsm8k-train-sample-v1-0037` — ADR-0207 R4 goal-residual production
+  (`goal - Σprogress`) promoted through `generate/derivation/goal_residual.py`.
+- `gsm8k-train-sample-v1-0042` — fast-path serving case.
 
-Both are exposed through a guarded product bridge over the pooled derivation
-reader. The bridge does **not** promote the pooled reader wholesale; the eight
-known pooled-reader wrong commits remain refused.
+The product cases are exposed through a guarded product bridge over the pooled
+derivation reader. The R4 case is exposed through the guarded goal-residual
+bridge. Neither bridge promotes the pooled/general composer wholesale; known
+wrong commits remain refused, and the no-reference `<N> times` completeness
+hazard remains a hard refusal control.
 
 ## ADR-0164 Reader — Zero-Delta Diagnosis