board: population-law cross-check — #1311 witness vs real-data fold experiment

.claude/board/INTEGRATION_PLANS.md

@@ -1,3 +1,7 @@
+## 2026-10-03 (2) — population-law-crosscheck-v1 — which execution law survives two independent witnesses → `.claude/plans/population-law-crosscheck-v1.md`
+
+**Status:** ANALYSIS (D-PLX-0..1). No code. Revises three #1311 conclusions (second fan-out, G4, `values()` modelling).
+
 ## 2026-10-03 — frontend-parity-witness-v1 — SQL, Cypher, Gremlin, SurrealQL on one population algebra → `.claude/plans/frontend-parity-witness-v1.md`
 
 **Status:** MEASURED for SQL ↔ Gremlin (test-only witness in `lance-graph-quack/tests/gremlin_parity.rs`); analysis for Cypher (#1306 plan) and SurrealQL (AST). D-FPW-0..4. No production code.

.claude/board/STATUS_BOARD.md

@@ -1,3 +1,12 @@
+## D-PLX — Population-law cross-check (2026-10-03)
+
+Plan: `.claude/plans/population-law-crosscheck-v1.md`. Entry: `entries/2026-10-03-population-law-crosscheck.md`.
+
+| D-id | scope | status | gate / falsifier |
+|---|---|---|---|
+| **D-PLX-0** | cross-check of #1311 against the real-data fold experiment; reclassify the five gaps | Shipped | plan §A–§M |
+| **D-PLX-1** | test-only falsifier: a K-space fold sink read back per row through an fk (lane kind vs result handle) | Queued | plan §L |
+
 ## D-FPW — Frontend parity witness (2026-10-03)
 
 Plan: `.claude/plans/frontend-parity-witness-v1.md`. Entry: `entries/2026-10-03-frontend-parity-witness.md`.

.claude/board/entries/2026-10-03-population-law-crosscheck.md

@@ -0,0 +1,27 @@
+# Population-law cross-check: frontend witness (#1311) against a real-data fold experiment (2026-10-03)
+
+**Status:** ANALYSIS + VERIFIED-IN-CODE; no code changed. Plan:
+`.claude/plans/population-law-crosscheck-v1.md`. The experiment's headline counts were
+re-derived from `crates/deepnsm/word_frequency/academic_20k.csv` (20,845 / 20,842 /
+18,559; three n=2 keys).
+
+- **Law that survives both:** execute over the population that already carries the
+  multiplicity; reach other data only by functional reads; switch anchors instead of
+  building target populations; a phase boundary is real only when a later pass needs,
+  per row, a fold over several rows of an earlier pass.
+- **#1311 revised:** a second fan-out is not itself a barrier. Re-anchoring onto a
+  population that references the current anchor costs read depth (G1). The barrier is a
+  carried fan-in aggregate (A → R1 → R2 over edge tables).
+- **I→S:** the temporary key column is an API artifact (the key is a function of the
+  result's coordinate; sinks keep no producing-key metadata; no coordinate-derived key
+  form exists). The presence fold's dependency on a complete I is real, unless the
+  source has an ordered resident projection.
+- **Result as operand:** mask results already feed a later phase over the same rows
+  (Quack `GroupPlan`). `i64`/record sinks are finalized in the K-space in host code by
+  design (Quack HAVING). Missing: reading a produced K-space result per row from a
+  later N-row pass. No `LaneRef::I64`; `PowerSums` exists only in the ndarray facade.
+- **#1311 gaps:** G1 confirmed general; G2 split (re-anchor vs phase boundary); G3 and
+  G5 simple missing operators; G4 mis-specified witness (by source, a missing
+  `masked_sum_i32_via`).
+- **Next:** one test-only falsifier (a fold sink read back per row through an fk) to
+  decide lane kind vs typed result handle.

.claude/board/entries/README.md

@@ -25,10 +25,11 @@ 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.
 
-183 entries, 2026-08-06 .. 2026-10-03.
+184 entries, 2026-08-06 .. 2026-10-03.
 
 | date | entry id | finding | file |
 |---|---|---|---|
+| 2026-10-03 | `population-law-crosscheck` |  | [2026-10-03-population-law-crosscheck.md](2026-10-03-population-law-crosscheck.md) |
 | 2026-10-03 | `known-is-not-alpha-premise-gate` |  | [2026-10-03-known-is-not-alpha-premise-gate.md](2026-10-03-known-is-not-alpha-premise-gate.md) |
 | 2026-10-03 | `frontend-parity-witness` |  | [2026-10-03-frontend-parity-witness.md](2026-10-03-frontend-parity-witness.md) |
 | 2026-10-03 | `dir-sim-soa-quack` | Directory simulation on SoA + Quack: one-edge mutation 853 B at 1k and 100k users | [2026-10-03-dir-sim-soa-quack.md](2026-10-03-dir-sim-soa-quack.md) |

.claude/plans/frontend-parity-witness-v1.md

@@ -19,6 +19,19 @@ Every traversal that does not lower is refused with a named reason. Two of the
 refusals are real substrate gaps, and both were already named independently by
 #1306 (Cypher) and #1308/#1310 (chained hops). No new primitive is proposed here.
 
+> **⊘ Corrections (2026-10-03, `population-law-crosscheck-v1.md`).** Three conclusions
+> below are revised, not deleted:
+> - **§2 / §6 G2 / §8:** "a second fan-out cannot keep the invariant" is wrong as a rule.
+>   A fan-out onto a population that holds an fk to the current anchor re-anchors again
+>   (at the cost of read depth, G1). The barrier is carrying a fan-in aggregate to the
+>   next population, not the hop count.
+> - **§6 G4:** the witness used `Country` (`u32`), so even the local sum cannot lower.
+>   It does not isolate a foreign-value sum. By source no such operator exists, but it
+>   is not measured here.
+> - **§3:** the adapter and the oracle both leave the cursor on the vertex after
+>   `values(f)`, so `values(f)` followed by a non-reducing step returns vertices in both.
+>   No test exercises that shape; adapter/oracle agreement there would prove nothing.
+
 ## §1 — What exists (A)
 
 | layer | in code? | where |