EW64↔CE64 decisions resolved (5-agent council) + RawEdge/counterfactual shipped (D-ATOM-4)

.claude/board/AGENT_LOG.md

@@ -1,3 +1,43 @@
+## [Main thread / Opus, autoattended] 5-agent dev council → D-ATOM-4/RawEdge shipped (① Heel-compose REFUTED, ② RawEdge built, ③ deferred)
+
+**Branch:** claude/jolly-cori-clnf9. **Cargo:** contract lib **550 green** (+5 counterfactual); default clippy clean. User: "use the 5 agent development council." Convened R1–R5 (Opus, read-only); consolidated + auto-resolved + built.
+
+**Verdict:** **①** DROP Heel-compose — R4(critic) + my own full-file reads: `Heel.plasticity` COOLS (`revise_truth`), not on the EW64 SoA, wrong edge encoding → phantom join (`E-BASIN-NOT-EDGE-PLASTICITY`, the 4th-strike object conflation). Coarse strength = MRU slot-order (shipped); per-edge Hebbian = per-plane `PlasticityState` (gated). **②** SHIPPED RawEdge: wired orphaned `counterfactual` mod (R5 P0), `RawEdge(i8)` not u64 (R5 P0, `size_of==1` structural guarantee), impl `EpisodicEdge`, filled `deposit_counterfactual` v2 (−6 on split), +3 latent scaffold fixes (SplitPoles Eq-with-f32; 2 v3-stub unused-params). Closes the counterfactual seam (not the prefetch loop — R4). **③** deferred (firewall-placement).
+
+**Process:** `E-READ-NOT-GREP` (user) — review agents must READ full files not grep/head/tail (fragments produced the 4 wrong framings). Baked into the agent-brief template (pattern §Rule 7). Spec §9; council consolidation `board/reviews/ew64-decisions-council.md`. Next: update #449 + subscribe stays active.
+
+---
+
+## [Main thread / Opus, autoattended] ① RESOLVED-IN-PRINCIPLE — per-plane clinical model verified REAL in causal-edge/src; coarse-first→per-plane-later
+
+Grep'd causal-edge/src (read, not built — owning the meta-lesson). Per-plane independence CONFIRMED: plasticity.rs freeze_s/heat_s + "established clinical pattern" (:16); edge.rs:713 live freeze_s; :750 pathological-plane count; lib.rs:46/52. Layout confirmed: plasticity 50-52 (PLAST_SHIFT=50), mantissa 46-49; v1/v2 minefield live (edge.rs 49 vs 50). RESOLUTION: ① is a build ORDER not either/or — (1) coarse NOW (per-basin Heel.plasticity × EW64 MRU, both offline; = #2 coarse + #1 compose), (2) per-plane PlasticityState = real, already-built, GATED clinical layer (phase 2). RawEdge = consensus first-build; sense-candidate = firewall-placement slice. Captured spec §8. Holding for @jan.
+
+---
+
+## [Main thread / Opus, autoattended] ① plasticity GROUNDED in high_heel.rs (owned the meta-flag); feedback #2 captured; HOLDING
+
+Read `high_heel.rs:135–187` directly (owning #2's meta-flag — ① was narrated from the board 3 turns running). CONFIRMED: `Heel::plasticity()` = a **per-basin u8** (0=frozen..3=hot), ONE per `HighHeelBGZ` (≤240 edges), **already shipped in contract** (offline). So ① is NOT "Heel-scalar vs PlasticityState" (different objects) — it's **GRANULARITY**: per-basin u8 (coarse, exists) vs per-edge-plane 3-bit (fine, gated). **Synthesis:** compose the EXISTING per-basin `Heel.plasticity` × the shipped `EpisodicEdges64` MRU slot-order — no new field (reconciles #1 "don't store" + #2 "u8 already bought"); default coarse, go per-plane only if S/P/O harden independently (clinical-patterns hint, unverified). #2 also: sense-candidate = a firewall-PLACEMENT question, not a menu pick; RawEdge mantissa-only = both-session consensus. Captured spec §7. Decisions remain @jan's — holding.
+
+---
+
+## [Main thread / Opus, autoattended] other-session feedback #1 captured — 3 decisions grounded vs causal-edge/layout.rs; HOLDING for @jan
+
+**Captured** (spec §6, #449) session-#1's grounded resolutions (verified `causal-edge/src/layout.rs`: per-plane plasticity 50–52, mantissa i4 46–49, Heel = 128-byte container): **①** per-plane (50–52) NOT Heel scalar — and DON'T store a graded weight; compose strength from MRU-slot × signed-mantissa × per-plane (RISC, avoids drift). **②** `RawEdge` mantissa-only as a TYPE (structural one-writer-per-field, like `MailboxSoaView`), not a convention. **③** sense-candidates = reuse proposer layer (VSA16k/aerial `TopKDistance`) as ⟨f,c⟩ proposals, top-k upstream, substrate sees only resolved opaque edge; lowest priority.
+
+**Held, not acted:** the decisions are @jan's (reserved as "THREE DECISIONS for @jan"; feedback explicitly "no action"; the ① compose-don't-store reframe is architecturally significant). Build queue now clarified: ② RawEdge type + the ①-compose `strength` fn are buildable-now (contract, offline); the plasticity WRITE stays gated. Awaiting @jan's pick. Also corrected §2's imprecise "PLAST_SHIFT 49 vs 50" → plasticity 50–52, mantissa 46–49.
+
+---
+
+## [Main thread / Opus, autoattended] episodic-witness64-ce64-prefetch SPEC — consolidates shipped hot tier + gated phases + 3 user decisions
+
+**Branch:** claude/jolly-cori-clnf9 (synced; #447 + #448 merged). Both overnight slices landed: the white-matter HOT TIER is complete in main — D-EW64-2 (promote/MRU), D-EW64-3 (coldest/contains), D-EW64-4 (DemotionSink + promote_into). Safe-unblocked queue EXHAUSTED.
+
+**Shipped (this turn):** `.claude/specs/episodic-witness64-ce64-prefetch.md` — the queued seam spec. Phase A SHIPPED; Phase B (plasticity-write co-fire) GATED, Phase C (surreal/LanceDB-LIVE wingman) GATED on OQ-11.6, Phase D (EpisodicWitness64 SoA column) GATED offline, Phase E (comprehension↔arcuate ±5 wire) NEEDS-DESIGN. Frames the **3 decisions for @jan**: (1) plasticity model — `Heel` scalar vs `PlasticityState` per-plane; (2) `RawEdge` mantissa-only scope (D-EW64-5); (3) sense-candidate source for the comprehension wire.
+
+**Holding** code construction for those 3 decisions (no gated/minefield work unattended). PR + subscribe next; this is the morning handover artifact.
+
+---
+
 ## [Main thread / Opus, autoattended] D-EW64-3/4 review LAND + CodeRabbit contains nit applied (#448)
 
 Opus review agent: **LAND** — no P0/P1 (exhaustively verified: coldest == eviction victim for every word, no holes, promote_into word == promote().0 + sink gets exactly the eviction; firewall + API clean; 545 green). 2 optional editorial P2s NOT applied (don't block). CodeRabbit: 1 nit (💤 low value) — `contains` → `self.iter().any(|x| x == e)` (more idiomatic, reuses iter; equivalent) — APPLIED. episodic_edges tests still green; default clippy clean. #448 CI re-runs on this push.

.claude/board/EPIPHANIES.md

@@ -1,3 +1,15 @@
+## 2026-06-01 — E-READ-NOT-GREP — judgment-critical review agents must READ full files, not grep/sed/head/tail; fragments invalidate judgment
+
+**Status:** IRON RULE (process; user-stated 2026-06-01). Across the EW64 council + the 3 prior relayed sessions, every wrong framing came from narrating off grep/head fragments rather than reading the type. The council's R3 even found a grep-induced mis-citation (`edge.rs:750 concern_level` reads `direction()`, not `PlasticityState`). **RULE:** when a review/council/grounding agent's VERDICT depends on a type's semantics, its brief MUST instruct it to READ the relevant files in full (the Read tool), NOT grep/sed/head/tail. A fragment seen out of context produces a confident-but-wrong judgment — **grep is for LOCATING, reading is for JUDGING.** Baked into the agent-brief template (`autoattended-multiagent-pattern.md` §Rule 7). Cross-ref: `E-BASIN-NOT-EDGE-PLASTICITY` (the conflation grep-fragments produced).
+
+---
+
+## 2026-06-01 — E-BASIN-NOT-EDGE-PLASTICITY — the 4th-strike object conflation: per-basin `Heel.plasticity` (a NARS-confidence COOLING knob, not on the EW64 hot path) is NOT coarse edge-plasticity; the Hebbian edge weight is per-plane `PlasticityState` (gated) or the MRU slot-order (shipped)
+
+**Status:** FINDING (5-agent council-resolved + orchestrator source-verified, 2026-06-01). R4(critic) + my own full-file reads killed the "compose `Heel.plasticity` × MRU" resolution that spec §6/§8 (relayed from 2 prior sessions) walked into. **Verified by reading the source:** (a) `MailboxSoaView` (the EW64 hot path) has **NO** Heel/plasticity column (only `energy`/`edges_raw`/`meta_raw`/`entity_type`) — Heel is unreachable from the EW64 edges; (b) `Heel.plasticity`'s only writer is `revise_truth()`, which **COOLS** as NARS confidence rises (`high_heel.rs:252` "Cool plasticity as confidence rises"; `is_frozen = plasticity==0 && conf>0.8`) — **opposite polarity** to Hebbian fire→hot; (c) `HighHeelBGZ.edges` are `CausalEdge64` u64s, EW64 slots are `EdgeRef` — different encodings, no index map. So "compose Heel × MRU" is a **phantom join of anti-correlated signals on the wrong edge set.** **RESOLUTION:** coarse strength = the MRU slot-order (#447, shipped); the real per-edge Hebbian weight = per-plane `PlasticityState` co-fire (GATED, phase B); no Heel, no new field. **The 4th strike** (after CausalEdge64-lens / per-plane-axis / Heel-vs-PlasticityState): same-word-different-**OBJECT** — "plasticity" names BOTH a cold-path basin cooling knob AND a hot-path per-edge Hebbian state; they don't compose. Cross-ref: D-ATOM-4/RawEdge (shipped), spec §9, `E-READ-NOT-GREP`.
+
+---
+
 ## 2026-06-01 — E-EW64-STRENGTH CORRECTION — "W15 0..3 plasticity" is `high_heel::Heel` (128-byte container field), NOT the 64-bit `CausalEdge64`; the 64-bit edge's plasticity is the 3-bit-per-plane `PlasticityState`
 
 **Status:** CORRECTION (factual, Plan-agent-grounded against source 2026-06-01). Refines `E-EW64-STRENGTH-IS-CE64-PLASTICITY`'s mechanism claim; does NOT change D-EW64-2 (MRU slot-order strength stores NO plasticity — it stands regardless).

.claude/board/LATEST_STATE.md

@@ -498,3 +498,5 @@ PR sequence: #360 → #361 → post-#360 substrate-sweep (this PR).
 - **`crates/lance-graph-arm-discovery`** (NEW, **excluded** standalone zero-dep crate; build via `cargo test --manifest-path crates/lance-graph-arm-discovery/Cargo.toml`). The **Aerial+** Rust transcode (Karabulut 2025, 2504.19354v1) — the upstream runtime-data proposer leg of `streaming-arm-nars-discovery-v1`. Public surface: `encode::{FeatureSpec, Dataset}`, `rule::{Item, CandidateRule, Proposer}`, `translator::{arm_to_nars, NarsTruth, CandidateTriple, FeedProjector, DebugProjector, NARS_PERSONALITY_K}`, `ndjson::to_ndjson`, and (feature `aerial`, default-on) `aerial::{Rng, AerialAutoencoder, AerialParams, AerialProposer, extract_rules, ExtractParams}`. 35/35 tests, clippy `-D warnings` clean. Emits the `{"s","p","o","f","c"}` ndjson the SPO store loader reads; `(f,c)` == `TruthValue::new(f,c)` == `ruff_spo_triplet::Triple{f,c}`. Determinism boundary: nondeterministic AE is seeded + feature-gated + emits *proposals* only. Synergy map: `.claude/knowledge/aerial-arm-ruff-spo-codegen-synergies.md`. Status board: D-ARM-13 (Shipped on branch) + D-ARM-SYN-1/2/3 (Queued). **Not** in `lance-graph-contract` yet — `rule`/`translator` are the local seam until D-ARM-1/2 land the shared carriers.
 
 > **2026-06-01 — PR-in-flight (autoattended)** (D-EW64-3/4): `lance_graph_contract::episodic_edges` gains `EpisodicEdges64::{coldest, contains, promote_into}` + the `DemotionSink` trait. `coldest()` = the eviction victim (symmetric to `strongest()`); `contains()` = family-discriminating membership; `promote_into(e, sink)` = `promote` routing the evicted (coldest) edge to a `DemotionSink` — the hot→cold connectome exit. `DemotionSink` impls (surreal/LanceDB-LIVE "wingman", `E-SUBSTRATE-IS-THE-SCHEDULER`) are deferred + GATED on OQ-11.6. Zero-dep; contract lib 545 green; default clippy clean; `episodic_edges.rs` pedantic+nursery clean.
+
+> **2026-06-01 — Shipped (autoattended, 5-agent council)** (D-ATOM-4/RawEdge): `contract::counterfactual` wired into `lib.rs` (was orphaned); `RawEdge(i8)` mantissa-only **structural** impl of `EpisodicEdge` (`size_of==1` — a u64 newtype could read plasticity 50–52); `deposit_counterfactual` v2 filled (−6 on split). Closes the counterfactual seam (NOT the prefetch loop). +3 latent scaffold fixes. 550 contract lib green, clippy clean. The council REFUTED the prior "compose `Heel.plasticity` × MRU" ① resolution (`E-BASIN-NOT-EDGE-PLASTICITY`): coarse strength = MRU slot-order (shipped); per-edge Hebbian = per-plane `PlasticityState` (gated).

.claude/board/reviews/ew64-decisions-council.md

@@ -0,0 +1,13 @@
+# 5-agent dev council — EW64↔CE64 decisions (2026-06-01)
+
+User: "use the 5 agent development council." Five Opus read-only reviewers; orchestrator adjudicated against full-file source reads (not the board).
+
+| reviewer | ① plasticity | ② RawEdge | ③ sense-candidate |
+|---|---|---|---|
+| R1 architectural | CONFIRM *pattern* (compose-not-store; method-on-carrier caveat) | CONFIRM (projection newtype) | CONFIRM (firewall-placement) |
+| R2 iron-rule | CONFIRM firewall-safe (Heel ≠ CE64 reclaim zone) + mandatory field-isolation test | CONFIRM (impl-carrier for existing trait) | CONFIRM (arcuate Binary16K invariant) |
+| R3 type-accuracy | CONFIRM types; flagged `edge.rs:750` mis-citation | CONFIRM (trait BLOCKED, no RawEdge dup) | CONFIRM |
+| **R4 critic** | **HOLD — drop Heel half** (no SoA home; `revise_truth` cools; wrong edge type) | LAND-w-rev (write-scoped; counterfactual seam ≠ prefetch loop) | LAND (defer) |
+| R5 build-order | (compose buildable but) | LAND-w-rev (i8 not u64; wire orphaned module — 2 P0s) | defer |
+
+**Adjudication (orchestrator, source-verified):** R4 wins ① — `MailboxSoaView` has no Heel column; `high_heel.rs:252` `revise_truth` COOLS; `HighHeelBGZ.edges` are `CausalEdge64` ≠ `EdgeRef`. R1/R3 endorsed the pattern/types, not the join. **① = MRU-is-coarse-strength (shipped) + per-plane gated; drop Heel.** **② RawEdge(i8) SHIPPED** (D-ATOM-4) with R5's two P0s applied. **③ defer.** Findings → `E-BASIN-NOT-EDGE-PLASTICITY`, `E-READ-NOT-GREP`; spec §9.

.claude/knowledge/autoattended-multiagent-pattern.md

@@ -337,3 +337,7 @@ This workspace's existing knowledge docs and conventions that overlap with the p
 - **Append-only zones in lance-graph:** `crates/lance-graph-contract/src/lib.rs` (re-export list), `crates/lance-graph/Cargo.toml` (`[features]` table), `crates/lance-graph/src/graph/mod.rs` (graph backend registry). Workers touching these must declare them in their slice brief.
 - **Harvest tools:** lance-graph already has some — `docs/UNIFIED_INVENTORY.md`, `docs/TYPE_DUPLICATION_MAP.md`, `docs/SEMIRING_ALGEBRA_SURFACE.md`. These map to §7's "reference indexes" role and should be re-harvested after any wave that adds a new type or semiring.
 - **5th savant for SIMD invariants:** lance-graph (and any consumer of the ndarray fork) extends the project-agnostic 4-savant taxonomy in §3 with a workspace-specific `simd-savant` whose card lives at `.claude/agents/simd-savant.md`. Its scope is **the SIMD source-of-truth invariant — all SIMD must come from `ndarray::simd` via the polyfill (`simd.rs` + `simd_ops.rs` > `simd_{type}.rs` per-arch)**. It runs PRE-SPAWN (verifies worker brief routes through the polyfill), DURING-IMPL (commit-level grep for raw `_mm*` / `vld1q_*` outside `ndarray/src/simd_*`), and PRE-MERGE (gate on any PR with SIMD code). 8-entry AP-SIMD-N anti-pattern catalogue covers the common violations. Hand-offs: UB/OOB → PP-13; missing primitive → file `TD-NDARRAY-SIMD-<NAME>` and route to ndarray maintainer; spec-drift → PP-16; cross-crate aliasing → PP-15. Workspace-specific because SIMD source-of-truth depends on having a polyfill repo to be the source — not all projects do, so this savant is an adapter, not a transferable §3 slot.
+
+### Rule 7 — judgment agents READ, don't grep (added 2026-06-01)
+
+When a review/council/grounding agent's **verdict** depends on a type's semantics, its brief MUST instruct it to **READ the relevant files in full** (the Read tool), NOT grep/sed/head/tail. A fragment out of context produces a confident-but-wrong judgment — **grep is for LOCATING, reading is for JUDGING.** Belegt: 4 successive wrong plasticity framings (CausalEdge64-lens / per-plane-axis / Heel-vs-PlasticityState / Heel-compose) all came from narrating off grep fragments; the EW64 council's R3 found a grep-induced mis-citation (`edge.rs:750`). Cross-ref EPIPHANIES `E-READ-NOT-GREP`.