sprint-log-10 board hygiene: meta-review + 12 worker scratchpads

Per CCA2A Layer-2 pattern (Mandatory Board-Hygiene Rule, CLAUDE.md):
worker scratchpads as the persistent A2A blackboard + Opus meta-review
as cross-spec reconciliation.

Worker scratchpads (W1-W12, 12 files, 11 new + 1 was partial):
each itemizes mandatory reads, design decisions, key delta vs parent
plan, and open questions surfaced for meta-review.

Meta-review (28 KB):
- Per-worker grades: W2/W3/W5/W8/W10/W11/W12 = A/A-; W1/W6/W7/W9 = B+;
  W4 = B- (shortest spec, AwareOp stub deferred)
- Sprint grade: B+ (substrate-level CausalEdge64 plan/code-gap finding
  is the central value-add; 3 hard blockers remain pre-sprint-11)
- 6 cross-spec inconsistencies (CSI-1..CSI-6) surfaced, 4 with proposed
  resolutions and severity flags
- 5 cross-cutting epiphanies (E-META-1..5):
  - E-META-1: specs-against-source > specs-against-plan
  - E-META-2: late-spec coordination gap (W7+W9 from main-thread
    surfaced bugs the fleet missed)
  - E-META-3: scratchpad discipline bimodal
  - E-META-4: 4 BindSpace columns + Sigma-tier band = AGI-as-glove API
  - E-META-5: diamond dep graph holds
- Sprint-11 spawn decision: NO until 5 pre-spawn fixes + 4 user
  ratifications (CSI-1 + OQ-1 + OQ-3 + OQ-5)
- Adjusted wave sequence: add Wave 0.5 PR-J1-INT4-32D-ATOMS before Wave 1

AGENT_ORCHESTRATION_LOG.md modification records the sprint-log-10
12-worker fleet run + main-thread W7/W9 backfill + meta-review.

https://claude.ai/code/session_01UwJuKqP828qyX1VkLgGJFS

Author: claude

.claude/board/AGENT_ORCHESTRATION_LOG.md

@@ -1220,3 +1220,13 @@ Detail:
   Plus W8 scope (full_pipeline.rs + bgz7_hydration_quality.rs) has ~5 unfixed sites.
 - test: /dev/vda at 100% (68 MB free); datafusion/parquet compile aborted.
 - Missing agent reports: W8, W10 (W10 code in working tree, uncommitted).
+
+## [W3 sprint-log-10] 2026-05-14 — PAL8 + NarsTables regression spec
+
+**Agent:** W3 (pal8-nars-regression, Sonnet, sprint-log-10 CCA2A)
+**Deliverables:** D-CE64-MB-2 + D-CE64-MB-3 spec
+**Output:** `.claude/specs/pr-ce64-mb-2-pal8-nars-regression.md` (32 KB)
+**Plans cited:** `causaledge64-mailbox-rename-soa-v1.md` §3
+**Key finding:** `edge.rs` has no unused bits in 51-63 (plan's "13 reserved bits" does not match impl — plasticity at 49-51, temporal at 52-63). W2 must resolve reclaim strategy before implementation. Tests written against functional accessor properties, not raw bit positions.
+**Status:** SPEC DRAFT complete. Tests: 6 gating + 1 ignored property test. CI extension: 3 new steps in rust-test.yml.
+W4 | 2026-05-14 | pr-ce64-mb-3-bindspace-efgh.md | ~14 KB | Plans: bindspace-columns-v1 §1-§5, causaledge64 §6-§7 | COMPLETE | Closes PR355#6 + FIX-5 + Phase2 | OQ: BindSpaceView placement (par-tile vs driver)

.claude/board/sprint-log-10/agents/agent-W1.md

@@ -0,0 +1,56 @@
+# Agent W1 — par-tile-crate — Sprint-10 Scratchpad
+
+**Worker:** W1 (par-tile-crate)
+**Timestamp:** 2026-05-14T12:35:38Z
+**Model:** claude-sonnet-4-6
+**Status:** COMPLETE
+
+## Deliverable
+
+Spec written to: `.claude/specs/pr-ce64-mb-1-par-tile-crate.md`
+File size: 37,134 bytes (~36 KB, 13 sections)
+
+## Plans Cited
+
+1. `.claude/plans/causaledge64-mailbox-rename-soa-v1.md` — §0 (zone framing), §4 (AttentionMask struct), §5 (MailboxSoA<N> struct + footprint + lifecycle), §6 (par-tile NEW substrate crate), §7 (PR-CE64-MB-1 scope ~1500 LOC), §11 OQ-5 (defer vendored-rayon)
+2. `.claude/board/LATEST_STATE.md` — confirmed ractor="0.14" from PR #366 S7-W3 (CallcenterSupervisor)
+3. `.claude/board/sprint-log-10/MANIFEST.md` — fleet coordination, W1 role
+4. `crates/lance-graph-supervisor/Cargo.toml` — exact ractor version match ("0.14", optional, features=["tokio_runtime"])
+5. `.claude/knowledge/A2Aworkarounds.md` — Workaround 1 (File Blackboard) coordination protocol
+
+## Key Delta vs Parent Plan
+
+Parent plan §6 listed par-tile as "NEW crate (to be specced)" with ~1500 LOC estimate. This spec materializes the full Cargo.toml, 7-module layout, all three Mailbox<T> backings (InMemoryMailbox/TokioMailbox/SupabaseSubMailbox), AttentionMask SoA with LRU eviction + wrap-around renormalization, MailboxSoA<N> with push_row/dispatch_cycle/emit_one/drop_row lifecycle, BindSpaceView via NonNull<u8> raw pointer (dep-isolation without importing cognitive-shader-driver), and dep-guard build.rs that panics on forbidden transitive deps.
+
+## Key Design Decisions (DELTA vs parent plan)
+
+1. **BindSpaceView dep isolation**: cognitive-shader-driver CANNOT be a par-tile dep (diamond invariant). Solved via `NonNull<u8>` + lifetime `'a` — callers (lance-graph-supervisor, bevy) hold Arc<BindSpace> and pass raw pointers.
+2. **EvictionSender enum**: `Std(std::sync::mpsc::Sender<EvictionEvent>)` + `#[cfg(feature = "tokio-backing")] Tokio(tokio::sync::broadcast::Sender<EvictionEvent>)` — keeps AttentionMask pure std by default.
+3. **LRU wrap at u32::MAX**: renormalize all LRU arrays (subtract u32::MAX/2, preserve relative ordering). Triggered at ~860 seconds of continuous operation at 1 cycle/5ms.
+4. **dep-guard build.rs**: `FORBIDDEN_DEPS = ["lance", "arrow", "datafusion", "blas", "mkl", "cognitive-shader-driver", "lance-graph-callcenter"]` — panics at compile time if any appear in transitive deps.
+5. **vendored-rayon as no-op stub**: feature exists but panics if enabled (OQ-5 deferred per §11).
+6. **Source LOC**: ~1,425 (within §7 1500 estimate); tests ~540 LOC separate.
+
+## Open Questions Surfaced for Meta-Review
+
+- **OQ-A**: Should causal-edge appear as a direct `[dependencies]` entry or should par-tile newtype CausalEdge64 rather than re-export? (Affects whether downstream crates get the type from par-tile or must also depend on causal-edge.)
+- **OQ-B**: Should vendored-rayon feature be a no-op stub (current spec) or omitted entirely until OQ-5 profiling shows the throughput cliff? (Stub risks confusion; omitting risks forgetting to re-add.)
+- **OQ-C**: BindSpaceView's `unsafe fn from_raw_parts` requires callers to prove the Arc<BindSpace> outlives the view. Should par-tile expose a safe constructor that takes `&Arc<BindSpace>` (requiring cognitive-shader-driver import, breaking diamond) or keep the unsafe API and document the contract?
+
+## Files-to-Touch Summary (from spec §11)
+
+| File | Action |
+|---|---|
+| `Cargo.toml` (workspace root) | Add `crates/par-tile` to members |
+| `crates/par-tile/Cargo.toml` | New |
+| `crates/par-tile/build.rs` | New (dep-guard) |
+| `crates/par-tile/src/lib.rs` | New |
+| `crates/par-tile/src/mailbox.rs` | New (Mailbox<T> trait + 3 backings) |
+| `crates/par-tile/src/attention.rs` | New (AttentionMask SoA) |
+| `crates/par-tile/src/soa.rs` | New (MailboxSoA<N>) |
+| `crates/par-tile/src/bind_space_view.rs` | New (BindSpaceView) |
+| `crates/par-tile/src/sigma_tier.rs` | New (SigmaTier enum + routing) |
+| `crates/par-tile/tests/` | New (5 test files) |
+| `crates/lance-graph-supervisor/Cargo.toml` | Add par-tile dep |
+| `crates/cognitive-shader-driver/Cargo.toml` | Add par-tile dep |
+

.claude/board/sprint-log-10/agents/agent-W10.md

@@ -0,0 +1,47 @@
+# agent-W10 scratchpad — sprint-log-10 CCA2A
+
+## Identity
+- Worker: W10 (pr-dep-graph)
+- Role: Cross-PR dependency graph + sequencing + parallel-landability + OQ gating
+- Output: `.claude/specs/sprint-10-pr-dep-graph.md`
+- Date: 2026-05-14
+
+## Mandatory reads completed
+1. `.claude/board/sprint-log-10/MANIFEST.md` — confirmed W10 scope
+2. `.claude/plans/causaledge64-mailbox-rename-soa-v1.md` §7 (PR sequencing) + §10 (blast radius) + §11 (OQs) + §15 (readiness checklist) — primary source
+3. `.claude/board/LATEST_STATE.md` — checked current contract inventory, active branches
+4. `.claude/board/AGENT_LOG.md` — file does not exist yet (first W10 agent in sprint-10)
+5. Worker spec files W1-W9 — not yet authored (sprint-10 workers running in parallel); this graph is based on parent plan §7 + crate-dep inference
+
+## Spec file
+- Path: `/home/user/lance-graph/.claude/specs/sprint-10-pr-dep-graph.md`
+- Size: 25183 bytes (~24.6 KB)
+- Sections: 11 (§1 scope, §2 PR inventory, §3 dep graph ASCII, §4 parallel-landability 6 waves, §5 OQ-to-PR gating table, §6 cross-spec consistency 6 checks, §7 risk matrix, §8 handover, §9 user gating recommendations, §10 files-to-touch, §11 cross-ref index)
+
+## Key findings
+
+### Dependency graph structure
+- 8 PRs total (7 from parent plan §7 + 1 ndarray prerequisite from W8)
+- 6 waves: Wave 0 (ndarray independent), Waves 1+2 (parallel: par-tile + causal-edge), Wave 3 (parallel pair: BindSpace + AriGraph), Wave 4 (convergence: MailboxSoA), Wave 5 (SigmaTierRouter), Wave 6 (bevy proof)
+- Maximum parallelism: 3 workers simultaneous (W1 + W2 + W3 in Waves 1+2)
+
+### Key delta vs parent plan §7
+- Parent plan shows linear sequence PR-CE64-MB-1 through MB-7. This spec surfaces the ACTUAL parallel structure: Waves 1+2 are parallel; Wave 3 is a parallel pair; Wave 6 (bevy) can start after Wave 4 without waiting for Wave 5.
+- Bevy plugin (PR-CE64-MB-7) does NOT hard-depend on SigmaTierRouter (PR-CE64-MB-6). Primary plugin path uses MailboxSoA directly. This is a meaningful unblocking: Wave 6 can proceed in parallel with Wave 5.
+
+### OQ gating table (primary contribution)
+- OQ-5 -> gates W1 (par-tile apex = Wave 1 foundation): must ratify before sprint-11 spawns
+- OQ-1 -> gates W7 (SigmaTierRouter = Wave 5, highest-risk PR): must ratify before MB-6 merges
+- OQ-3 -> gates W6 (MailboxSoA = Wave 4 convergence): must ratify before MB-5 merges
+- OQ-2, OQ-7, OQ-8 -> ratify at meta-review or during sprint-11 W4/W5 prep
+- OQ-4, OQ-6 -> documentation-only, defer
+
+### Top 3 open questions for meta-review
+1. **C-1 (accessor naming)**: W2's spec will define the actual accessor names for CausalEdge64 G/W/truth fields. Meta-reviewer MUST verify W6 and W7 specs use identical names. If they diverge, integration fails silently. This is the single highest-probability cross-spec integration bug.
+2. **C-5 (SigmaTier enum residence)**: Recommend moving SigmaTier from par-tile to lance-graph-contract to avoid import cycle risk. Meta-reviewer should check if W6 and W7 independently defined SigmaTier and force reconciliation.
+3. **C-2 (BindSpaceView lifetime)**: Parent plan §5 shows `BindSpaceView<'static>` in MailboxSoA rows, which requires Arc<BindSpace> interior. If W1 (par-tile) uses raw refs instead, W6 (MailboxSoA) cannot compile without unsound lifetime extension. This should be caught at meta-review before sprint-11 spawns.
+
+## Coordination
+- No other W1-W9 agents found to have shipped yet when W10 ran (sprint-log-10/agents/ was empty)
+- This graph is based on parent plan + crate-dep inference, not on reading other worker specs
+- Meta-reviewer should update this graph after reading all W1-W9 specs if any structural deps changed

.claude/board/sprint-log-10/agents/agent-W11.md

@@ -6,3 +6,15 @@
 **Mandatory reads completed:** MANIFEST.md (fleet table), parent plan §3/§7/§12, LATEST_STATE.md, AGENT_LOG.md (W12 only agent so far), scripts/miri-tests.sh (lance-graph), ndarray/scripts/miri-tests.sh, rust-test.yml + style.yml (CI structure), .claude/specs/ directory (no sprint-10 worker specs yet — W1-W9 not yet run)
 **Key observation:** W1-W9 worker specs do not yet exist (W12 is first runner). W11 must draft the test-plan spec from parent plan + miri scripts + CI structure alone, with placeholders for per-worker spec citations.
 **Drafting spec now.**
+
+## 2026-05-14T12:29 — W11 spec complete
+
+**Spec written:** `.claude/specs/sprint-10-test-plan.md` — 43,718 bytes, 775 lines
+**Plans cited:** causaledge64-mailbox-rename-soa-v1.md §3/§5/§6/§7/§8/§12; scripts/miri-tests.sh (lance-graph); ndarray/scripts/miri-tests.sh; rust-test.yml; style.yml; MANIFEST.md
+**Key delta:** Miri coverage growth target = ~760 current → ~1550 post-sprint-11 (3 mechanisms: W8 dispatch reroute +529 ndarray, new pure-Rust crates +116 lance-graph, expanded lance-graph Miri scope +145)
+**Open questions for meta-review:**
+  - OQ-T1: W1-W9 test count reconciliation (specs not yet authored)
+  - OQ-T2: proptest Miri runtime (PROPTEST_CASES=100 mitigation untested)
+  - OQ-T3: bevy headless approach (xvfb-run vs bevy/headless feature)
+  - OQ-T4: causal-edge Miri scope (confirm pure-Rust dep closure; W2 must verify)
+  - OQ-T5: ractor Miri guard in lance-graph-supervisor (W7 must add cfg(not(miri)) guards)

.claude/board/sprint-log-10/agents/agent-W12.md

@@ -6,3 +6,14 @@
 **Mandatory reads completed:** MANIFEST.md, §14+§15 of parent plan, sprint-5-9 roadmap, LATEST_STATE.md, EPIPHANIES.md (E-CE64-MB-1..10), CLAUDE.md board-hygiene rule, A2Aworkarounds.md, cca2a-sprint-prompt-template.md
 **No other worker scratchpads present** — W12 is first agent running.
 **Drafting spec now.**
+
+## 2026-05-14T12:35 — W12 spec complete (sonnet)
+
+**Commit:** (pending main-thread aggregation — not yet committed)
+**LOC delta:** +473 lines / ~28.5 KB at .claude/specs/sprint-10-execution-plan.md
+**Tests:** N/A (governance spec, no code)
+**Open questions:**
+- OQ-1: User ratification of Sigma-tier banding policy needed before sprint-11 Wave 5
+- OQ-3: User ratification of compartment plasticity update granularity needed before Wave 4
+- OQ-5: User ratification of rayon vendor decision (std::thread::scope first) needed before Wave 1
+- Meta-reviewer (M agent, Opus) needed to validate all 12 specs before sprint-11 spawns

.claude/board/sprint-log-10/agents/agent-W2.md

@@ -0,0 +1,19 @@
+## 2026-05-14T12:34 — W2 causaledge64-v2 (Sonnet)
+
+**Spec path:** `.claude/specs/pr-ce64-mb-2-causaledge64-v2.md`
+**Size:** 30,478 bytes (~30 KB, 13 sections)
+**Plans cited:** `causaledge64-mailbox-rename-soa-v1.md` §3 (layout extension), §2 (truth-band lens collapse), §7 (PR sequencing); `crates/causal-edge/src/edge.rs` (actual shipped layout); `crates/causal-edge/src/network.rs`; `crates/lance-graph-planner/src/cache/nars_engine.rs`
+
+**Key delta vs parent plan §3:**
+- CRITICAL: Plan §3 describes "current" layout with 13 reserved bits (bits 51-63). Actual shipped `edge.rs` uses ALL 64 bits — no reserved bits exist. This is a BLOCKER (OQ-LAYOUT-1).
+- Recommended Option C: drop `temporal(12)` → AriGraph SPO-G quad, `plasticity(3)` → MailboxSoA::plasticity_counters, `infer(3)` → AttentionMask::style_slots. Frees 18 bits for G(5)+W(6)+truth(2)+spare(5).
+- PAL8 serialization (4101-byte format referenced in plan) not found in `crates/causal-edge/` source — OQ-PAL8-FORMAT BLOCKER for W3.
+- `forward()` method reads bits 46-48 (`inference_type`) which becomes G slot under Option C — OQ-FORWARD-REFACTOR.
+- TrustTexture defined locally (zero external deps constraint preserved).
+- Feature flag `causal-edge-v2-layout` default ON; version bump 0.1.0 → 0.2.0.
+
+**Open questions for meta-review:**
+1. OQ-LAYOUT-1 (BLOCKER): Which of Options A-E does plan author ratify? Without answer, bit allocation is blocked.
+2. OQ-PAL8-FORMAT (BLOCKER for W3): PAL8 impl not in crates/causal-edge/ — where does it live? W3 cannot write round-trip tests without knowing.
+3. OQ-FORWARD-REFACTOR: Does `forward()` refactor land in PR-CE64-MB-2 scope or a follow-on PR?
+

.claude/board/sprint-log-10/agents/agent-W4.md

@@ -0,0 +1,13 @@
+## W4 run — 2026-05-14
+
+Status: COMPLETE
+Spec: .claude/specs/pr-ce64-mb-3-bindspace-efgh.md (~14 KB)
+
+Key findings:
+- Column H EntityTypeId already wired (PR #272): entity_type Box<[u16]> + push_typed + driver.rs:311. This PR adds TypeColumn wrapper + BindSpaceView accessor + FIX-5 test.
+- MergeMode::Superposition=2 already shipped in collapse_gate.rs. This PR documents semantics + Column E tagging.
+- Corrected footprint: 71,777+292=72,069 B/row (plan §1/§2 numbers predate cycle plane migration to Vsa16kF32).
+- AccumulatedOntology: global on ShaderDriver (NOT per-row), correctly per plan §11.
+- AwareOp ndarray impls D-F4/D-F5 out-of-scope; shipped as no-op stub [128u8;256].
+
+OQs: OQ-W4-1 BindSpaceView placement (par-tile vs driver, BLOCKER MB-5); OQ-W4-2 EdgeColumn 8-slot scope; OQ-W4-3 AwareOp stub.

.claude/board/sprint-log-10/agents/agent-W5.md

@@ -0,0 +1,47 @@
+# agent-W5 scratchpad — sprint-log-10
+
+## W5 run: 2026-05-14
+
+**Worker:** W5 (arigraph-spo-g)
+**Output:** `.claude/specs/pr-ce64-mb-4-arigraph-spo-g.md` (23 KB)
+
+### Mandatory reads completed
+
+1. `.claude/board/sprint-log-10/MANIFEST.md` — fleet table read; W5 row confirmed
+2. `.claude/plans/causaledge64-mailbox-rename-soa-v1.md` §6 + §7 — PR-CE64-MB-4 entry read
+3. `.claude/plans/ogit-g-context-bundle-v1.md` — D-OGIT-G-1 SPO-G u32 slot spec read in full
+4. `.claude/plans/oxigraph-arigraph-cognitive-shader-soa-merge-v1.md` §1-§9 — SPOW tetrahedron + Gaussian splat read
+5. `.claude/board/LATEST_STATE.md` — AriGraph inventory confirmed (4696 LOC, 7 modules)
+6. `.claude/board/AGENT_LOG.md` — not found (new board log file, no prior sprint-10 entries)
+7. `crates/lance-graph/src/graph/arigraph/` — mod structure confirmed (mod.rs, triplet_graph.rs, episodic.rs, retrieval.rs, sensorium.rs, orchestrator.rs, xai_client.rs, language.rs, spo_bridge.rs)
+8. `grep -rn "named_graph|ogit_context_id|SPO_G|witness_ref"` — zero hits (no partial SPO-G implementation)
+
+### Key findings from source inspection
+
+- Current `Triplet` struct: `{subject, object, relation, truth: TruthValue, timestamp: u64}` — no G field
+- `spo_bridge::promote_to_spo` takes `(triplet, gate, spo)` — no G parameter
+- `lance_cache.rs` SCHEMA_VERSION currently 2 — confirmed at line 62
+- `lance_cache_invalidate_*` + `schema_version_pinned` test pattern confirmed for migration guidance
+- `contract::hash::fnv1a` canonical per PR #307 — recommended for `witness_ref` derivation
+
+### Spec scope delivered
+
+1. SPO-G quad: Triplet extended with `g: u32`, `pearl_rung: u8`, `witness_ref: u64`
+2. Ghost-edge persistence: `ghost.rs` NEW with `GhostReason`, `GhostStore`, `GhostReactivationEvent`, `nars_revise_ghosts`
+3. Witness shapes: `witness.rs` NEW with `SpoWitness64` (Copy, 8B) + `SpoWitnessChain<N=32>` + `WitnessChainStore`
+4. Schema migration: SCHEMA_VERSION 2→3, backward-compat via default g=0
+5. Test plan: 7 tests covering round-trip, G-filter, witness packing, ghost persistence, reactivation, NARS decay, chain truncation
+6. Risk matrix: HIGH (Lance schema bump, promote_to_spo API break), MED (chain sizing, decay rate), LOW (rung encoding)
+
+### Open questions surfaced
+
+- OQ-W5-1: Lance persistence for ghost edges (recommend defer to PR-CE64-MB-4b)
+- OQ-W5-2: promote_to_spo API evolution (recommend new function, not builder)
+- OQ-W5-3: witness_ref derivation via contract::hash::fnv1a (confirm dependency already exists)
+
+### Delta vs parent plan
+
+Parent plan §6 had: "SPO-G quad mode + ghost-edge persistence + SpoWitnessChain" as 3-line summary.
+Spec adds: exact `Triplet` field layout (confirmed from source), `SpoWitness64` bit packing (mirrors CausalEdge64 layout), `GhostStore<'a>` lifetime-parameterized design, `nars_revise_ghosts` batch API, SCHEMA_VERSION bump target, `WitnessChainStore` HashMap design, context-separation law enforcement (G vs witness_ref).
+
+**Status: DONE**

.claude/board/sprint-log-10/agents/agent-W6.md

@@ -0,0 +1 @@
+W6 2026-05-14 DONE: pr-ce64-mb-5-mailbox-soa-attentionmask.md (47678 B) — AttentionMask SoA + AttentionMaskActor ractor + MailboxSoA<N> + BindSpaceView<'_> + sequence diagram + 9 tests + risk matrix. 3 new OQs: OQ-N (sizing), OQ-SHADOW (cycle-speed shadow table), OQ-BCAST-SIZE (broadcast capacity). Plans: causaledge64-mailbox-rename-soa-v1 §4+§5+§7.