diff --git a/.claude/board/LATEST_STATE.md b/.claude/board/LATEST_STATE.md
index be7e133a0..fd351afd6 100644
--- a/.claude/board/LATEST_STATE.md
+++ b/.claude/board/LATEST_STATE.md
@@ -1,3 +1,36 @@
+## 2026-09-05 — #1188 MERGED (431371b3): D-BLW-5 measured — the observer-effect instrument reads SILENT at the κ floor
+
+| PR | merge | content |
+|---|---|---|
+| **#1188** | `431371b3` | `tests/d_blw_5_observer.rs` (1689 lines, `cycle-driver`) — the observer-effect probe with the belief-arena reader, resumed on operator direction and MEASURED; supervisor → `jc` dev-dep; four epiphanies (incl. the operator ruling on loco/r2il); `TD-SUPERVISOR-CLIPPY-RED-ON-BASE-1`; D-BLW-5b queued. |
+
+**Contract inventory net delta: none** — one test binary and one dev-dep. No new
+contract type, no SoA column, no widening: the probe CONSUMES `contract::shape_rank`
+(`ShapeRankPayload` / `RemeasureLedger` / `RemeasureKey`, shipped #1181) and the
+existing `BeliefArena`; the inventory delta recorded on the branch section below
+stands unchanged by the merge.
+
+**What the merge fixes in the board's picture.** D-BLW-5's `**PAUSED by operator
+2026-08-05**` wording — in `STATUS_BOARD.md` under the `D-BLW-5` row — is
+superseded by the `> **Update 2026-09-05 (D-BLW-5, first measurement):**` line
+directly beneath it: the loop is RESUMED and RUN, not paused. Read the row through
+that line; the PAUSED text is kept per append-only, not because it is current.
+
+**The one-line result, and the reason it is not a null.** All seven pre-registered
+gates pass with their disable twins; O4/O5 are SILENT at the pinned κ floor 0.10
+while reader B saturates in proportion to injected typicality on T and F−, with
+P / N / CTRL bit-identical. κ is invariant to a saturating reader — so the silence
+is a property of the OBSERVABLE, not evidence that the loop does nothing. That is
+what D-BLW-5b exists to separate, by moving the observable onto the reader
+marginals (`rate_b`, φ) with floors pinned in a spec BEFORE the run.
+
+**Also merged here:** the operator ruling
+`E-PLANNING-MIGRATES-TO-LOCO-R2IL-DATAFUSION-IS-GRACE-PERIOD-1` (recorded in the
+INVENTORY DELTA section at the head of this file) and the independent D-OIF-1
+census that reached #1185's ruling A by a separate route; its epiphany carries a
+dated correction line ceding the narrower remove cone to #1185.
+
+---
 ## 2026-09-05 — OPERATOR RULING: planning → ogar-loco / ogar-r2il, DataFusion in grace period — INVENTORY DELTA
 
 - **Ruling:** E-PLANNING-MIGRATES-TO-LOCO-R2IL-DATAFUSION-IS-GRACE-PERIOD-1. Every DataFusion-hosted surface in this inventory (`datafusion_planner`, `sql_query`, Python `SessionContext`, `graph_table`, `rls.rs`, `query`/`query-lite`, the `datafusion-dispatch`/`datafusion-plan` forward-stubs) is now **grace-period: maintained, not extended**. New behaviour lands in loco/r2il programs or on Lance directly.
diff --git a/.claude/board/PR_ARC_INVENTORY.md b/.claude/board/PR_ARC_INVENTORY.md
index 7848f386a..ad31fe39e 100644
--- a/.claude/board/PR_ARC_INVENTORY.md
+++ b/.claude/board/PR_ARC_INVENTORY.md
@@ -1,3 +1,13 @@
+## 2026-09-05 — lance-graph PR #1188 (merged `431371b3`, branch `claude/pr-294-ragged-path-validation-170zcy`) — D-BLW-5 measured: the observer-effect instrument is alive and reads SILENT
+
+- **Added:** `crates/lance-graph-supervisor/tests/d_blw_5_observer.rs` (1689 lines, `#[cfg(feature = "cycle-driver")]`) — the D-BLW-5 observer-effect probe with the belief-arena reader, resumed on operator direction; `jc` as a supervisor **dev-dep** (the ORCHESTRATOR-RATIFIED exception from the 08-05 design note — dev-only, no runtime edge); `EPIPHANIES` `E-BLW5-FIRST-MEASUREMENT-1`, `E-NARS-EXPECTATION-CHOICE-PREFERS-IGNORANCE-TO-A-CONFIDENT-NEGATIVE-1`, `E-PLANNING-MIGRATES-TO-LOCO-R2IL-DATAFUSION-IS-GRACE-PERIOD-1` (operator-ruled) and `E-THE-UNFINISHED-UDF-WAS-NOT-THE-DEBT-1`; `TECH_DEBT` `TD-SUPERVISOR-CLIPPY-RED-ON-BASE-1`; `STATUS_BOARD` D-BLW-5 update line + D-BLW-5b row; plan `cycle-loop-closure-driver-v1` §12.9b; `.claude/knowledge/observer-effect-tfpn-doctrine.md` §7; three `exec-runs/` tag-files (build spec + Sonnet worker + the D-OIF-1 census).
+- **Measured:** `shape₀ × rank₀` (`ShapeRankPayload`, sealed per arm in `RemeasureLedger` at V0) injected into `BeliefArena` as 16 `prior Inh band_k` statements plus a per-subject `subject Inh prior` at `f = 1, c = typicality`; propagation `rcr_abduce` + `close_transitive` iterated to a fixed point (`MAX_REASON_ROUNDS = 16`); arms T / F+ / F− / P / N / CTRL over 40 owners 1:1. **All seven pre-registered gates pass with their twins**; O4/O5 read **SILENT** at the pinned κ floor 0.10, while reader B saturates *in proportion to injected typicality* on T and F− and P/N/CTRL come back **bit-identical**. CTRL idempotence and the DROP gate hold.
+- **Locked:** κ is invariant to a saturating reader — that is WHY O4/O5 are silent, and it is a property of the instrument, not a null result about the loop. D-BLW-5b (same arms/payload/arena, observable moved onto the reader marginals `rate_b` / φ with floors pinned in a spec BEFORE the run) is the successor, queued.
+- **Corrected:** the NARS `expectation() = c·(f − 0.5) + 0.5` CHOICE discards a *confident negative* in favour of a vacuous path — the first build encoded typicality as a frequency below 0.5 and the arena preferred ignorance (c = 3.5e-11). Fix: `f = 1, c = typicality`. Recorded as its own epiphany because it is a substrate property, not a probe bug.
+- **Also (operator ruling, rode this PR):** *"Every planning is in migration to ogar-loco and ogar-r2il, especially datafusion is out of the picture, what exists gets a grace period, nothing new will migrate to it."* `CLAUDE.md` carries the ⊘ note; every DataFusion-hosted surface is maintained-not-extended. The D-OIF-1 census banked here reached ruling A independently of #1185; its entry carries a dated correction line ceding the narrower remove cone (MedCare's gated constructor exists) to #1185's `E-THE-UNFINISHED-FUNCTION-WAS-NOT-THE-DEBT-1`.
+- **Deferred:** the D-BLW-5b floor spec; the E-NXG-22 `out_of_support` bit; `TD-SUPERVISOR-CLIPPY-RED-ON-BASE-1` (`cycle_driver.rs` 8× `needless_pass_by_ref_mut` + `probe_ignition_64k.rs:435`, all pre-existing on base — the new test is clippy-clean on its own target).
+- **Docs:** merged after `origin/main` was brought in twice (lance 11 / lancedb 0.38 via #1190, then #1185), by merge — no history rewrite. **Confidence:** High on every gate (each pre-registered, each with a disable twin); Medium on the interpretation that κ-silence is instrument-invariance rather than absence of effect — that is exactly what D-BLW-5b falsifies.
+
 ## 2026-09-05 — lance-graph PR #1185 (merged `f8b629a3`, branch `claude/open-ideas-fetch-v1`) — D-OIF-1 ruled SUPERSEDED; the cross-repo reachability law
 
 - **Added:** `.claude/plans/open-ideas-fetch-v1.md` (D-OIF-0..7, §2 = the D-OIF-1 retirement ruling with tree verdict / obligation table / remove-grace-retain cone / layer audit / stale-claim list); `EPIPHANIES` `E-A-COLUMN-OF-INDICES-INTO-A-CODEBOOK-THAT-DOES-NOT-EXIST-1` and `E-THE-UNFINISHED-FUNCTION-WAS-NOT-THE-DEBT-1`; `ISSUES` `ISS-PIN-RULING-PROSE-DRIFTS-BEHIND-THE-MANIFEST`; four `IDEAS.md` entries + three in-place status flips; `STATUS_BOARD` D-OIF-0..7 + three `-DEC` rows.