diff --git a/.claude/board/AGENT_LOG.md b/.claude/board/AGENT_LOG.md
index e31fc4033..d93b22ef2 100644
--- a/.claude/board/AGENT_LOG.md
+++ b/.claude/board/AGENT_LOG.md
@@ -1,3 +1,9 @@
+## 2026-10-16 — PR-body harvest #1582–#1591 (general-purpose on Haiku)
+
+- One run: ten PR bodies and file lists, verbatim, into the session scratchpad.
+- Spot checks: #1588's two test names found in `tests/binder.rs`; #1587 and #1591 file counts equal `git diff --stat` (17, 18).
+- Misses: none. Written by the orchestrating main thread only.
+
 ## 2026-10-16 — board catch-up for merged PRs #1388–#1588 (Haiku + Sonnet grindwork)
 
 - Six Sonnet writers drafted 115 `PR_ARC_INVENTORY` entries from each PR's description and file list (brief: facts only, shas copied from the JSON). The main thread matched every header against the PR metadata mechanically and read samples; inserted by a script that refuses duplicates and a shorter file.
diff --git a/.claude/board/LATEST_STATE.md b/.claude/board/LATEST_STATE.md
index 8f0da59c2..7b5c830d4 100644
--- a/.claude/board/LATEST_STATE.md
+++ b/.claude/board/LATEST_STATE.md
@@ -1,3 +1,9 @@
+## 2026-10-16 — main after #1591 (merged `e53f2325`); ledger revision (c) in PR
+
+- **On main since #1581:** #1582 (doc drift, OpenCog oracle-only decision), #1584/#1585 (D-ASM follow-ups queued), #1586/#1587 (WCOJ map and probes; `seek_from`, `is_alpha_acyclic`), #1588 (CATS three-valued binding), #1589 (arc catch-up), #1590 (gather designs bannered as violations), #1591 (`deepnsm_v2::wire`: every CE64 field of a parsed triple, Gadamer gate first; plan D-CGW-0..14).
+- **In PR:** `LEDGER-REVISION-2026-10-16c.md`; arc entries for #1549 and #1581.
+- **Flagged:** bits 50..52 now carry two opposite encodings in code (pyramid 7 → 0 vs `with_novelty_wisdom` 1 → 7).
+
 ## 2026-10-16 — CE64 grammar wiring (branch `ccr-9f3b8075-74u47l-ce64-grammar-wiring`, in PR)
 
 - `deepnsm-v2::wire` writes every CausalEdge64 field of a parsed triple from a named source, with the Gadamer gate (`contract::revision`) deciding before anything is counted; W is the statement's Markov arc in a `WitnessTable<64>`; 43..45 is the Haspelmath dative-map function; loci carry Antecedent, TEKAMOLO, Quorum and Contradiction pointers. Probe: `lance-graph-quack/examples/wired_reasoning_probe.rs`. Plan: `ce64-grammar-wiring-v1.md` (D-CGW-0..12).
diff --git a/.claude/board/PR_ARC_INVENTORY.md b/.claude/board/PR_ARC_INVENTORY.md
index c36037185..ca0da5eae 100644
--- a/.claude/board/PR_ARC_INVENTORY.md
+++ b/.claude/board/PR_ARC_INVENTORY.md
@@ -1,3 +1,17 @@
+## 2026-10-16 — lance-graph PR #1581 (merged `95c0d104`, branch `ccr-b2e415d9-4jfvyk`, head `67d0698a`, 3 commits) — recovery ledger deduplicated against #1579; DOC-WIRE (2); N21
+
+- **Added:** `.claude/research/2026-10-recovery/LEDGER-UPDATE-2026-10-16b.md`; dated corrections on three docs that still read bits 59..60 as two readings (`causal-plane-inventory.md`, `TYPE_DUPLICATION_MAP.md`, `causaledge64-mailbox-rename-soa-v1.md`); entry `2026-10-16-ledger-update-dedup-1579.md`; arc entry for #1578.
+- **Locked:** none (docs only). Measured: #1570's five commands all exit 0.
+- **Deferred / open:** N17 gated on D-RA-2; D-ASM per-space codebook; DOC-WIRE (1), BAND-RETIRE, O30, O29.
+- **Confidence:** medium.
+
+## 2026-10-14 — lance-graph PR #1549 (merged `149839f8`, branch `research/oracle-fidelity-auditable-thoughts-20261014`, head `53131901`, 3 commits) — D-ORACLE-0: original-oracle fidelity and auditable thoughts
+
+- **Added:** `.claude/plans/d-oracle-fidelity-matrix-v1.md` (research protocol: run OpenNARS, ONA, OpenPsi, Hyperon and others one-to-one as frozen reference oracles over 100–150 pre-registered fixtures), an oracle reference ledger, and a handoff prompt.
+- **Locked:** none (research documentation; no oracle parity claimed).
+- **Deferred / open:** every oracle run; versions to be frozen before measurement.
+- **Confidence:** low (protocol only).
+
 ## 2026-10-16 — lance-graph PR #1590 (merged `2a6d502d`, branch `ccr-9f3b8075-74u47l-violation-banners`, head `d43a6ac5`, 1 commit) — board: mark the gather designs as violations; point at D-HOLO H14c/H14d
 
 - **Added:** capital-letter VIOLATION banners on plans `wcoj-closing-atoms-v1`, `quack-multitable-bind-v1`, `metta-datalog-assimilation-v1-council-v3`, the D-CMR-5 row, the joins line of `relalg-v1`; entries `wcoj-closing-atoms-council`, `wcoj-probes-1-3`, `coresearch-wcoj-leapfrog`, `asm-accepted-proposals`; STATUS rows D-QMB-0..5, D-ASM-11/11A/11B, D-CMR-5, D-RA-4; both 2026-10-16 INTEGRATION_PLANS entries; `TD-NO-BOUNDED-U32-GATHER-1`. Each names `d-holo-h14c-thought-first-v1.md` and `d-holo-h14d-logic-mask-read-v1.md` as the correct plans.
diff --git a/.claude/board/entries/2026-10-16-ledger-revision-1582-1591.md b/.claude/board/entries/2026-10-16-ledger-revision-1582-1591.md
new file mode 100644
index 000000000..10082a91f
--- /dev/null
+++ b/.claude/board/entries/2026-10-16-ledger-revision-1582-1591.md
@@ -0,0 +1,32 @@
+# 2026-10-16 — Recovery ledger revised against #1582–#1591
+
+**Status:** WORKING-MODEL (research document; board catch-up; no code changed).
+
+`.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16c.md` maps ten
+merged PRs onto the open list.
+
+- **The gather ban (#1590) reorders the list:**
+  - D2 (preimage op) and D-RA-4 are excluded.
+  - D-ASM needs a re-spec: its existence half and foreign fan-out are bannered.
+  - N19 (layout) is promoted: the join is the layout.
+  - Open: no DECISION record of the ruling, and the existing gather code is untouched.
+- **New collision (WIRE-50):** bits 50..52 have two encodings in code with opposite order.
+  - Pyramid, `maturation_ladder.rs`: 7 Astonishment → 0 Wisdom.
+  - `CausalEdge64::with_novelty_wisdom` (#1591): 0 = not on the ladder, 1 Staunen ..= 7 wisdom.
+  - D-CGW-1's gate keeps wire edges in the probe for now. Merged with D-MATURATION3-1.
+- **Closed:**
+  - C31 lance-graph doc drift;
+  - C32 OpenCog as oracle only;
+  - C33 CATS three-valued binding;
+  - C34 WCOJ pre-work measured;
+  - C35 G1/G2 loci;
+  - C36 Gadamer gate on parsed data;
+  - C37 the measured modulo-64 stamp failure, evidence for O30;
+  - C38 arc catch-up.
+- **Connections:**
+  - K9: the join is the layout.
+  - K10: one CE64 code table settles WIRE-50, DOC-WIRE (1), D-46 and D-CGW-1/9/10 together.
+  - K11: D-CGW-4 should call `Reaction::verdict`.
+  - K12: failing tests for the stamp failures.
+
+OPEN: WIRE-50, GATHER-DECISION, GATHER-CODE, D-ASM re-spec, O30, D-46, BAND-RETIRE, O29.
diff --git a/.claude/board/entries/README.md b/.claude/board/entries/README.md
index fe5a4e894..9d39394a6 100644
--- a/.claude/board/entries/README.md
+++ b/.claude/board/entries/README.md
@@ -25,7 +25,7 @@ index row, (3) no duplicate entry id. Checks 1 and 2 are deliberately
 opposite directions; the stranding this convention prevents shows up in
 exactly one of them, never both.
 
-360 entries, 2026-08-06 .. 2026-10-16.
+361 entries, 2026-08-06 .. 2026-10-16.
 
 | date | entry id | finding | file |
 |---|---|---|---|
@@ -36,6 +36,7 @@ exactly one of them, never both.
 | 2026-10-16 | `D-ASM-1` |  | [2026-10-16-metta-datalog-assimilation-council-v3.md](2026-10-16-metta-datalog-assimilation-council-v3.md) |
 | 2026-10-16 | `ledger-update-dedup-1579` |  | [2026-10-16-ledger-update-dedup-1579.md](2026-10-16-ledger-update-dedup-1579.md) |
 | 2026-10-16 | `ledger-revision-after-1576` |  | [2026-10-16-ledger-revision-after-1576.md](2026-10-16-ledger-revision-after-1576.md) |
+| 2026-10-16 | `ledger-revision-1582-1591` |  | [2026-10-16-ledger-revision-1582-1591.md](2026-10-16-ledger-revision-1582-1591.md) |
 | 2026-10-16 | `grammar-coordinates-annotation-wiring` |  | [2026-10-16-grammar-coordinates-annotation-wiring.md](2026-10-16-grammar-coordinates-annotation-wiring.md) |
 | 2026-10-16 | `D-HOLO-H14d` |  | [2026-10-16-d-holo-h14d-logic-mask-read.md](2026-10-16-d-holo-h14d-logic-mask-read.md) |
 | 2026-10-16 | `coresearch-wcoj-leapfrog` |  | [2026-10-16-coresearch-wcoj-leapfrog.md](2026-10-16-coresearch-wcoj-leapfrog.md) |
diff --git a/.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16c.md b/.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16c.md
new file mode 100644
index 000000000..f7916ed7f
--- /dev/null
+++ b/.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16c.md
@@ -0,0 +1,126 @@
+# Ledger revision, 2026-10-16 (c): #1582–#1591
+
+Maps the merges since the last mapping (`LEDGER-UPDATE-2026-10-16b.md`, #1581)
+onto the open list. That file and its predecessors are not edited; this one
+supersedes the rows it names. Board entry:
+`.claude/board/entries/2026-10-16-ledger-revision-1582-1591.md`.
+
+**Scope.** Ten merged PRs, #1582–#1591. Read from the merged PR bodies (Haiku
+harvest, verbatim), the code diffs of #1587, #1588 and #1591, and the board
+files each PR changed. #1583 and #1589 are board-only; #1590 is board and plans
+only.
+
+## 1. The ruling that reorders everything: gather is forbidden (#1590)
+
+#1590 put a VIOLATION banner on every plan, entry and STATUS row that designs a
+read at a position taken from another lane's values: gather, foreign lanes, and
+the preimage hop. The correct design it names is the D-HOLO pair
+(`d-holo-h14c-thought-first-v1.md`, `d-holo-h14d-logic-mask-read-v1.md`):
+`Thought(t)[T_a[i] : T_b[i]]`, both values of a pair at the same coordinate, so
+**the join is the layout**.
+
+Rows of this ledger the ruling touches:
+
+| row | before | after |
+|---|---|---|
+| D2 (one preimage op serving four callers, by reference from #1579) | OPEN, fruit H2 | **EXCLUDED.** D-CMR-5 is bannered; the op is not to be built. |
+| D-RA-4 (bounded recursion over D-CMR-5 preimage) | waits on D-RA-2 | **EXCLUDED** (bannered). D-RA-2 itself is not touched: it is about where single-table rules live, and only relalg's "joins" line is bannered. |
+| N15 → D-ASM | Queued (authorization) | **Re-spec needed.** The council-v3 §0 preimage hop, D-ASM-2's existence half (Semijoin / `EqU32Via`) and D-ASM-3's foreign fan-out (`Semijoin` × `GroupReduce{Count}`) are all bannered; D-ASM-11 / 11A / 11B too, although 11A/B were marked authorized the same day. D-ASM-1 (binder) and the Hyperon baseline measurement are untouched. |
+| N19 (layout costs, no A/B/C verdict) | OPEN, no decision depends on it | **Promoted.** "The join is the layout" makes the write layout the join strategy. H14d's measurement (bytes fall only when survivors are page-local) is now the cost model for every relation, not one probe's footnote. |
+
+What the ruling did NOT do, and is still open:
+
+- **No DECISION record.** The ruling exists only as banners. No entry carries DECISION / SCOPE / BASIS / REVISIT WHEN, so a session cannot read what is in scope (is a same-lane `seek` a gather? is a codebook lookup?) without re-deriving it.
+- **The code surface is untouched**, by design ("a separate change"): quack `Filter::Semijoin`, `Pred::EqU32Via`, `MaskOp::Gather`, and #1587's `SealedFacetLane::seek_from` / `gallop_partition_point` and relalg `is_alpha_acyclic`. The last two shipped with D-ASM-11 as their only planned consumer, which is now bannered.
+
+## 2. New closed ends
+
+| id | what closed | evidence |
+|---|---|---|
+| C31 | D8 doc drift, lance-graph part: `.cargo/config.toml` AVX2-oracle note (native for local, v4 / sapphirerapids for shipped), `CARGO-COMPUTE-SUBSTRATE.md` rev-pin template, TD-NDARRAY-SIMD-POPCNT-NATIVE item (b) withdrawn | #1582 |
+| C32 | OpenCog's role: a temporary validation oracle, harvested outside the workspace; AGPL with linking exception means an adapter does not change our licence, a port does | #1582, entry `2026-10-16-opencog-oracle-scope.md` (DECISION form) |
+| C33 | CATS optional fields bind three-valued through validity planes; reading the sentinel lane alone kept NULL rows | #1588; disable run red |
+| C34 | WCOJ pre-work measured: semijoin chain over-approximates on the cyclic F4 fixture; seek bound `m·(2·log₂(N/m)+4)` holds; GYO classifies 7 acyclic / 4 cyclic bodies; the exact cyclic mask needs one disjunct per value with today's ops; DuckDB core has no WCOJ | #1586, #1587 |
+| C35 | G1 (relative-clause antecedent → `Locus::Antecedent`) and G2 (TEKAMOLO → loci 0..3) | #1591 `wire.rs`, tests |
+| C36 | The Gadamer gate runs on real parsed data: every observation passes `GadamerRevision` before it is counted; a contest never resets a rung; echoes count nothing | #1591, four disable runs red |
+| C37 | O30 has a measured failure: `deepnsm_v2::belief::Stamp` folds sources modulo 64, so two verses 64 apart read as overlapping evidence | #1591 entry §7 |
+| C38 | Arc catch-up for 115 merged PRs | #1589 |
+
+## 3. Open items, revised
+
+Fruit: **L** low, **H** high, **Heavy**. Tier: who does the grindwork.
+
+| id | item | status after #1591 | fruit | tier |
+|---|---|---|---|---|
+| **WIRE-50** (new, highest) | Bits 50..52 now have two encodings in code with opposite order. The pyramid (`knowledge/maslov-pyramid-of-cognition.md`, `planner/tests/maturation_ladder.rs`): 7 Astonishment → 0 Wisdom. `CausalEdge64::with_novelty_wisdom` (#1591): 0 = not on the ladder, 1 Staunen ..= 7 wisdom. Under one global wire (R-b) a wire-written 7 reads as Astonishment. D-CGW-1's gate (wire edges stay in the probe) contains it today, but D-CGW-1 names only 43..45. | OPEN, decision | — | O |
+| **D-MATURATION3-1** (merged with WIRE-50) | Entry code (7 vs a separate "new"), `learn`'s rule, migration gate. #1591 is a concrete proposal for decision 3: code 0 = "not on the ladder". It costs Wisdom its code, which the pyramid gives to code 0. | OPEN, now one decision with two candidates in code | Heavy | O |
+| **DOC-WIRE (1)** (merged with D-CGW-1 / D-CGW-10) | One global meaning for 43..52 in `ce64-semantic-upper-half.md`. #1591 writes 43..45 as a Haspelmath dative-map function, beside the direction triad's `DIR_SHIFT = 43`. D-CGW-1 (Queued) is the same item: one code table in `causal-edge`, triad readers migrated. | OPEN, now has a queued owner | H | O decides, S migrates |
+| **D-46** (merged with D-CGW-9) | 46..49: D-CGW-9 proposes the ISA inference mantissa stays the law and activation leaves the edge. That answers D-46 in favour of the code. | OPEN, recommendation exists | — | O |
+| **EDGE-DOC** (new) | `CausalEdge64::novelty_wisdom`'s doc says "a writer picks one vocabulary per class and says which" — the per-class reading R-b withdrew (K1) and D-CGW-1 now forbids. | OPEN | L | spec-editor |
+| **GATHER-DECISION** (new) | Write the ruling as one DECISION entry (scope: which reads count as gather; whether same-lane seek and codebook lookups are in scope; REVISIT WHEN). | OPEN | L | O words it, S records |
+| **GATHER-CODE** (new) | The existing gather surface (`Filter::Semijoin`, `Pred::EqU32Via`, `MaskOp::Gather`) and #1587's orphans (`seek_from`, `gallop_partition_point`, `is_alpha_acyclic`). Each needs keep / gate / remove. `is_alpha_acyclic` may stay as a refusal test (refuse cyclic bodies). | OPEN | H | O decides per item |
+| **D-ASM** (re-spec) | See §1. What survives: D-ASM-1 binder, the Hyperon 26-script baseline, D-ASM-1b, -8, -10, -12, -13 where they do not route through a foreign lane. | OPEN, re-spec | Heavy | O, then /5plus3 |
+| **O25 + K2'** (evidence root) | Unchanged in shape; new concrete sites: `BeliefArena`'s modulo-64 stamp (C37) and AriGraph `revise_with_evidence`, which revises on a string match with no stamp (#1591 §1). | OPEN, needs O30 | H | O, then S |
+| **O30** | Evidential-base width. C37 is the measured cost of the current answer (64, folded). | OPEN, decision; now evidenced | — | O |
+| **K6** (derived packet, D-HL-5 H4) | #1591 writes 53..63 for parsed rows: W = the statement's previous state, topology from the clause path, certification from support balance, never `Causes`. That is the observation case; H4 asks the syllogism case. The two should agree on 59..63 vocabulary. | OPEN, one data point | H | S runs H4 |
+| **D-CGW-4 ↔ C22** (new connection) | `CausalCandidate` → `Causes` only with `IncreaseEligible ∧ Necessary` is exactly `RevisionVerdict::is_acceptable()`, whose first producer is `pearl.rs` `Reaction::verdict` (#1560). D-CGW-4 should call it, not re-derive it. | OPEN | H | S |
+| **ABL** | Ablation through the oracle families. #1582 fixed OpenCog's role; D-CGW-13 queues a cross-oracle readout on the same CE64 fields (Java OpenNARS, ONA, MeTTa, OpenCog, MORK); D-ORACLE-0 (#1549) pre-registers 100–150 fixtures. These are one harness. | OPEN | H | coresearch, then S |
+| **N16** (operand registry) | Related to D-CGW-8 (S/P/O identity as 8 × u8:u8 rails beside the u8 palette). | OPEN | Heavy | O |
+| **N17** | Unchanged: the absorption fix waits on D-RA-2. Not bannered. | OPEN | L tests / fix waits | S |
+| **N18** | Unchanged. | OPEN | H | S |
+| **N20** (qualia CLUT) | #1591 measured the palette's separation (nearest-neighbour L1 min 25, median 90) and showed a "3σ" claim needs the unmeasured A2B noise; D-CGW-14 (OpenPSI as a qualia device) is a candidate device profile. | OPEN | H | S |
+| **TD-ONTOLOGY-REGISTRY** (new, #1582) | `OntologyRegistry` resolves through string-keyed hash maps; review against codebook ids + masks, measured on one OGIT tenant. | OPEN, review | H | S |
+| **#1589 stale rows** | Its open-items entry lists K1, O25-as-policy, O26 and N1 as owed decisions; the ledger withdrew K1, made O25 a defect, downgraded O26 and narrowed N1. #1591's entry §4 repeats the N1 / rule-S7 reading of 50..52 that ruling R-c settled. | OPEN | L | spec-editor, dated correction lines |
+| **ARC-GAPS** | #1549 and #1581 had no arc entry. | **Closed in this PR** | — | — |
+| **STATUS D-CGW-0** | reads "In PR"; #1591 merged. | OPEN (owner session) | L | — |
+| O23, O29, R2–R4, O26, O7/O13/O15, BAND-RETIRE | Unchanged. | OPEN | — | O |
+
+## 4. Connecting the dots
+
+- **K9 — the join is the layout, and the layout is already measured.** The
+  gather ban (§1), N19's page-local cost (H14d), H14c's thought-first reads and
+  #1591's rows (each row already carries S, P, O, F/C and its loci at one
+  coordinate) are one design: whatever a query will pair must be written at
+  the same coordinate. That turns N19 from a cost note into the write-side
+  contract, and it is the replacement D-ASM needs for its existence half.
+- **K10 — one CE64 contract, written once.** WIRE-50, DOC-WIRE (1), D-46,
+  EDGE-DOC and D-CGW-1/9/10 are one change: a single code table in
+  `causal-edge` for 40..63, with `wire.rs`, `learn`, the pyramid tests and the
+  triad readers reading it. D-CGW-1's gate makes it safe to decide before any
+  wire edge leaves the probe.
+- **K11 — the docket is real now.** C22 built the verdict producer, C36 runs the
+  gate on parsed data, D-CGW-4 needs the verdict. Wiring D-CGW-4 to
+  `Reaction::verdict` closes the path from a sentence to an admitted `Causes`
+  with no new type.
+- **K12 — evidence roots have their failing cases.** C37 (modulo-64 stamps)
+  and AriGraph's stamp-less revision are the two concrete failures O30 and K2'
+  exist to prevent; both can become failing tests before the root type exists.
+- **K7 (from before) — ablation, oracles, D-ORACLE-0 and D-CGW-13 are one
+  harness.**
+
+## 5. Fruits
+
+- **Low:** GATHER-DECISION record; EDGE-DOC one-line fix; dated corrections on
+  #1589's and #1591's stale rows; failing tests for C37 and AriGraph's
+  stamp-less revision; N17's two reproducers as failing tests.
+- **High:** WIRE-50 + D-MATURATION3-1 decided and written into one code table
+  (K10); D-CGW-4 through `Reaction::verdict` (K11); GATHER-CODE triage; K6/H4;
+  the oracle harness (K7); N18; N20; the ontology registry review.
+- **Heavy:** D-ASM re-spec around layout joins (K9); O25 + K2' evidence root
+  after O30; N16; the write-side layout contract (N19 promoted).
+
+## 6. Decisions for the operator
+
+| # | question | recommendation | caveat |
+|---|---|---|---|
+| WIRE-50 | which order owns 50..52 | keep the pyramid (7 Astonishment → 0 Wisdom): it is the ruled reading (R-c) and has tests; give "not on the ladder" its own meaning only through D-MATURATION3-1 decision 3 | `wire.rs` and its probe re-pin; the probe's rule R2 reads rung order |
+| GATHER-DECISION | scope of "gather" | one DECISION entry; same-lane `seek` and frozen codebook lookups stated explicitly in or out | without it, every new plan re-asks |
+| GATHER-CODE | existing gather code | gate behind a named feature first, remove after D-ASM re-spec; keep `is_alpha_acyclic` as a refusal test | removal touches quack, mask-risc and lgj-abi |
+| D-46 | 46..49 | accept D-CGW-9: the mantissa is law, activation leaves the edge | activation needs a home (a lane beside the key) |
+| O30 | root width | unchanged: bounded root list keyed on (lane, row position), overflow refused | C37 shows folding is the wrong default |
+
+## 7. Agent ledger (Haiku harvest tier)
+
+| card | runs | items checked | wrong | missed |
+|---|---|---|---|---|
+| general-purpose on Haiku (PR bodies + file lists #1582–#1591) | 1 | #1588's two test names re-found in `tests/binder.rs`; #1587 / #1591 file counts equal `git diff --stat` (17, 18) | 0 | 0 |