From 138d42541bbaea1434d81b1aee57ba173c10b0e6 Mon Sep 17 00:00:00 2001 From: Claude Date: Fri, 23 Oct 2026 08:32:03 +0000 Subject: [PATCH 1/2] =?UTF-8?q?board:=20TD=20row=20=E2=80=94=20bgz-tensor?= =?UTF-8?q?=20integration=20tests=20never=20run=20in=20CI=20(Low-hanging?= =?UTF-8?q?=20fruit)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit CI runs bgz-tensor with --lib only, so the 9 lab probes under crates/bgz-tensor/tests (23 tests, incl. P-S2 behind TORQUE T-2026-10-23-1) are never gated. All pass on main 8f610a85. FRUITS.md regenerated. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01VncKoyG6JkpsoqhqTGTv3f --- .claude/board/FRUITS.md | 6 ++++-- .claude/board/TECH_DEBT.md | 20 ++++++++++++++++++++ 2 files changed, 24 insertions(+), 2 deletions(-) diff --git a/.claude/board/FRUITS.md b/.claude/board/FRUITS.md index ddc35533b..f1c3cb4c1 100644 --- a/.claude/board/FRUITS.md +++ b/.claude/board/FRUITS.md @@ -4,7 +4,7 @@ Generated by `.claude/tools/fruits_index.py --write`. Do not hand-edit: add or change a `**Fruit:**` line on the ledger row and regenerate. CI fails when this file is stale. -Open: 1. Closed: 0. +Open: 2. Closed: 0. ## discovery torque @@ -18,7 +18,9 @@ _none open_ ## engineering debt -_none open_ +| tier | row | status | ledger | +|---|---|---|---| +| Low-hanging | TD-BGZ-TENSOR-INTEGRATION-TESTS-UNGATED-1 (2026-10-23) — OPEN | OPEN | `TECH_DEBT.md` | ## blocker diff --git a/.claude/board/TECH_DEBT.md b/.claude/board/TECH_DEBT.md index 7faefd8d8..6e1847b23 100644 --- a/.claude/board/TECH_DEBT.md +++ b/.claude/board/TECH_DEBT.md @@ -1,3 +1,23 @@ +## TD-BGZ-TENSOR-INTEGRATION-TESTS-UNGATED-1 (2026-10-23) — OPEN + +CI runs `bgz-tensor` with `--lib` only (`.github/workflows/rust-test.yml:274`), +so nothing under `crates/bgz-tensor/tests/` runs in CI. The crate is +workspace-excluded, so no other step reaches it. + +On main `8f610a85` that directory holds 9 lab probes, 23 tests. Among them are +P-S2 (`p_s2_compose_chain_probe.rs`, #1726), the evidence behind TORQUE row +T-2026-10-23-1, and P-D3. Draft #1731 adds the EXO-129 probe in the same +directory, so its PR checks would pass without running it. + +**Receipt:** `cargo test --manifest-path crates/bgz-tensor/Cargo.toml --tests` +on main `8f610a85` with `CARGO_PROFILE_{DEV,TEST}_DEBUG=0 CARGO_INCREMENTAL=0`: +207 lib tests plus all 9 integration files (3+4+2+3+2+2+2+2+3 = 23 tests), 0 failed. +No test reads a file or an env var, so nothing blocks gating them. +**Fix:** drop `--lib` from that step (or add a second `--tests` step). The +step comment gives no reason for `--lib`. +**Fruit:** Low-hanging +**Status:** OPEN + ## TD-CANON-HIGH-AUDIT-RESIDUE-1 (2026-10-16, filed from #1430) — OPEN #1430 aligned the classid docs and two hand-written shifts with the canon-high From b208b6475c54c0936606d42c85811d88ff56e1cf Mon Sep 17 00:00:00 2001 From: Claude Date: Fri, 23 Oct 2026 08:35:36 +0000 Subject: [PATCH 2/2] =?UTF-8?q?board:=20two=20fruits=20=E2=80=94=20causal-?= =?UTF-8?q?edge=20Revision=20pools=20duplicates=20(P-E2),=20Fisher-z=20enc?= =?UTF-8?q?ode=20clamps=20silently=20(P-D3)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit TD-NARS-REVISION-UNGUARDED-CAUSAL-EDGE-ISA-1 (High-signal): isa.rs:209 revises without a stamp; P-E2 measures c 128->227 at k=8 duplicates. Same defect as TD-NARS-REVISION-UNGUARDED, a site that row does not list. ISS-FISHERZ-ENCODE-SATURATES-SILENTLY (Low-hanging): fisher_z.rs:67 clamps out-of-family cosines to the edge code with no flag (H5, never added). FRUITS.md regenerated. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01VncKoyG6JkpsoqhqTGTv3f --- .claude/board/FRUITS.md | 7 +++++-- .claude/board/ISSUES.md | 23 +++++++++++++++++++++++ .claude/board/TECH_DEBT.md | 24 ++++++++++++++++++++++++ 3 files changed, 52 insertions(+), 2 deletions(-) diff --git a/.claude/board/FRUITS.md b/.claude/board/FRUITS.md index f1c3cb4c1..ad37cbb0f 100644 --- a/.claude/board/FRUITS.md +++ b/.claude/board/FRUITS.md @@ -4,7 +4,7 @@ Generated by `.claude/tools/fruits_index.py --write`. Do not hand-edit: add or change a `**Fruit:**` line on the ledger row and regenerate. CI fails when this file is stale. -Open: 2. Closed: 0. +Open: 4. Closed: 0. ## discovery torque @@ -20,9 +20,12 @@ _none open_ | tier | row | status | ledger | |---|---|---|---| +| High-signal | TD-NARS-REVISION-UNGUARDED-CAUSAL-EDGE-ISA-1 — `Opcode::Revision` pools a duplicated source; a measured site of TD-NARS-REVISION-UNGUARDED (2026-10-23) | OPEN. **Basis:** VERIFIED-IN-CODE + MEASURED. | `TECH_DEBT.md` | | Low-hanging | TD-BGZ-TENSOR-INTEGRATION-TESTS-UNGATED-1 (2026-10-23) — OPEN | OPEN | `TECH_DEBT.md` | ## blocker -_none open_ +| tier | row | status | ledger | +|---|---|---|---| +| Low-hanging | ISS-FISHERZ-ENCODE-SATURATES-SILENTLY — out-of-family cosines encode to the edge code with no flag (2026-10-23) | OPEN. **Basis:** VERIFIED-IN-CODE (`crates/bgz-tensor/src/fisher_z.rs:67`, `i8_val.clamp(-128.0, 127.0) as i8`, read at `main` `8f610a85`) + MEASURED. | `ISSUES.md` | diff --git a/.claude/board/ISSUES.md b/.claude/board/ISSUES.md index 705dff450..2e1596bb5 100644 --- a/.claude/board/ISSUES.md +++ b/.claude/board/ISSUES.md @@ -1,3 +1,26 @@ +## ISS-FISHERZ-ENCODE-SATURATES-SILENTLY — out-of-family cosines encode to the edge code with no flag (2026-10-23) + +**Status:** OPEN. **Basis:** VERIFIED-IN-CODE (`crates/bgz-tensor/src/fisher_z.rs:67`, `i8_val.clamp(-128.0, 127.0) as i8`, read at `main` `8f610a85`) + MEASURED. +**Fruit:** Low-hanging + +- `FisherZTable::encode` clamps anything beyond the family's + `[z_min, z_max]` onto the edge code, and nothing reports it. An + out-of-family pair therefore reads as maximal in-family similarity. Below + `z_min` the value lands on −128, which `decode` maps one step past `z_min`. +- **Receipt:** P-D3, `crates/bgz-tensor/tests/p_d3_ood_support_probe.rs` + (`entries/2026-10-22-probes-pd4-pe2-pd3.md`). The clamp condition holds on + 0/120 in-range cosines and 403/403 saturating ones, and + `encode(1.0) == encode(0.99995)`. +- This is H5 of the H1–H5 falsifiers (EXO-50, session_01G22yT6 /tanh-01). H5 + was proposed and never added. +- **What closes it:** a checked encode that reports out-of-range values (a + count, or an `Option`) beside the existing one, with no change to the code + points. Add a can-fire test (a value past `z_max` is reported) and a + stay-silent test (in-family values are not). +- Note: until TD-BGZ-TENSOR-INTEGRATION-TESTS-UNGATED-1 is fixed, a test under + `bgz-tensor/tests/` does not run in CI. Put this one in the crate's lib + tests. + ## ISS-ADMISSION-PREP-BINDING — may preparation translate global identities into focus slots before a fold? (2026-10-20) **Status:** RULED 2026-10-20 (ADMISSION option B, `entries/2026-10-20-six-decisions.md` §1): same-table, version-witnessed key-range seeks over an attested sorted key are admitted in preparation; a position derived from another column or table is not, whatever the phase. Frozen codebook lookups are admitted under the attestation the plan defines (`d-relalg-unified-v1.md` § Decisions). D-MPS-8 stays conditionally blocked (a value-to-relationship-table read position is option C); D-RPR-1 and D-RST-4/6 are admitted only where they classify under B. Was: OPEN (operator ruling). **Basis:** VERIFIED-IN-CODE (`crates/mork-pathmap-probe/src/{rmap,join,chain}.rs`) and MEASURED (D-RMAP-1A §8, Schnellverschluss §7b, D-RPR §11.10). diff --git a/.claude/board/TECH_DEBT.md b/.claude/board/TECH_DEBT.md index 6e1847b23..886736500 100644 --- a/.claude/board/TECH_DEBT.md +++ b/.claude/board/TECH_DEBT.md @@ -1,3 +1,27 @@ +## TD-NARS-REVISION-UNGUARDED-CAUSAL-EDGE-ISA-1 — `Opcode::Revision` pools a duplicated source; a measured site of TD-NARS-REVISION-UNGUARDED (2026-10-23) + +**Status:** OPEN. **Basis:** VERIFIED-IN-CODE + MEASURED. +**Fruit:** High-signal + +- `crates/causal-edge/src/isa.rs:209` `truth_code` (and `:197` `truth`) runs + `Opcode::Revision` through `truth::revision` on four scalars. There is no + stamp, so nothing can see that the two operands share evidence. +- **Receipt:** P-E2, `crates/causal-edge/tests/p_e2_stamp_gate_probe.rs` + (`entries/2026-10-22-probes-pd4-pe2-pd3.md`). Revising one source with + itself k times starting from c = 128 gives c = 170 / 204 / 227 at + k = 2 / 4 / 8, with posterior sd ratio 0.837 / 0.658 / 0.491. +- The test-local ONA-shape gate already exists in that file and is + disable-verified. It checks stamp overlap first, and on overlap keeps the + stronger belief. With it, duplicates give Δc = 0 and an sd ratio of 1.000, + while disjoint sources still pool. +- This is the same defect as TD-NARS-REVISION-UNGUARDED (2026-07-23, ndarray + `nars.rs` and planner `nars_engine.rs`). That row does not list this site. +- **Open:** where the stamp lives (G5 in the entry). The guard belongs at the + belief store, not in the pure truth function. +- **What closes it:** pick the stamp's home, then call `Revision` only through + a gate that refuses or chooses on overlap. Move P-E2's three tests onto the + production path; each must go red when its guard is removed. + ## TD-BGZ-TENSOR-INTEGRATION-TESTS-UNGATED-1 (2026-10-23) — OPEN CI runs `bgz-tensor` with `--lib` only (`.github/workflows/rust-test.yml:274`),