Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 19 additions & 1 deletion .claude/board/LATEST_STATE.md
Original file line number Diff line number Diff line change
@@ -1,4 +1,22 @@
## 2026-08-20 — branch `claude/lance-graph-stage-3-recovery-2wrdbd` — DisMech compact evidence vocabulary + citation sidecar
## 2026-08-21 — #975 MERGED (4a43698) — two measured corrections + a retracted absence

- **`DismechTopology` doc + this file corrected**: the label-KNOWN 3,978 are
NOT the oracle population. **2,512 over 549 diseases** carry a mediator;
1,466 (36.9%) name nothing; **92** `INDIRECT_UNKNOWN_INTERMEDIATES` name
mediators despite their label. 2,512 is the population REQUIRING GROUNDING —
45 of 3,095 distinct mediator strings are exact node references (1.5%).
- **`deepnsm-v2` academic carve under-fills**: 20,845 rows are 18,559 distinct
surface forms, carve 90.6% full, basins 73..79 empty. Falsifier added,
disable-verified.
- **AriGraph retraction**: `crates/lance-graph/src/graph/arigraph/` is **15
modules, ~327 KB** (`ppr`/`bm25`/`rrf`/`community`/`markov_soa`/`episodic`/
`witness_corpus`/`retrieval`/…). It is IMPLEMENTED and UNWIRED — the state
`E-ARIGRAPH-IS-AN-ISLAND` already recorded. Absent ⇒ build the organs;
unwired ⇒ close the seam. Do not rebuild these under a new name.
- **Open, blocking any gold set (operator decision):** the third bucket for
the 1,466 label-only edges and the 92 contradictory ones.

## 2026-08-20 — #974 MERGED (8a93423) — DisMech compact evidence vocabulary + citation sidecar

### Current Contract Inventory — 1 new zero-dep module

Expand Down
101 changes: 101 additions & 0 deletions .claude/board/PR_ARC_INVENTORY.md
Original file line number Diff line number Diff line change
@@ -1,3 +1,104 @@
## 2026-08-21 — lance-graph #975 (MERGED, merge commit 4a43698, head 5577728) — two measured corrections: the DisMech oracle population, and the academic carve fill; plus a retracted absence

- **Added — the oracle-population correction (`5fac79d` + `5577728`),
`lance-graph-contract`:** `DismechTopology::source_knows_intermediates()`'s
doc and the `LATEST_STATE` summary both claimed the 3,978 label-KNOWN edges
ARE the oracle population. Measured over the 2,100-file corpus: **1,466
(36.9%) carry no `intermediate_mechanisms` at all** — the source asserts
mediators exist and does not name them. The usable population is **2,512
over 549 diseases**, and it is the population *requiring grounding*, not a
usable oracle: only **45 of 3,095** distinct mediator strings are exact node
references (1.5%); label matching grounds 1,834 (59.3%) and leaves 1,261
(40.7%) ungrounded prose. **No bits move, no API changes** — the function
answers a question about the LABEL and is correct as implemented.
- **Added — the academic-carve finding (`3fa7acb`), `deepnsm-v2`:**
`plans/causal-rung-standing-wave-v1.md:465` claimed the 20k academic
vocabulary "fits the palette256² pair carve NATIVELY". It under-fills:
`academic_20k.csv` is 20,845 ROWS but **18,559 distinct surface forms**
(2,286 duplicates — COCA lists a word once per part of speech), and
`from_frequency_ranked` admits by surface form. Carve fill **90.6%**; last id
at basin 72 slot 126; **basins 73..79 empty.** Companion falsifier added
(`duplicates_consume_ranks_but_not_slots_so_the_carve_underfills`),
disable-verified.
- **Added — the AriGraph retraction (`ffc5218`):** a prior claim that "no
`ppr` / `bm25` / `rrf` / `community` function exists" was FALSE.
`crates/lance-graph/src/graph/arigraph/` is **15 modules, ~327 KB** with
exactly those organs. Cause: `fn .*ppr` matches **`approx`**, and `head -20`
pushed the real hits off the end.
- **Locked:**
- **Absent ≠ unwired, and the remedies are opposite.** The board's own
`E-ARIGRAPH-IS-AN-ISLAND` already had the correct state — every module
ships and tests green, the chain is open at the joints. Absent ⇒ build the
organs. Unwired ⇒ build nothing, close the seam. The retracted claim
recommended the expensive one.
- **A measurement a committed parser can make must not be made by an ad-hoc
script.** The 2,489/1,489 figures in `5fac79d` came from a Python
line-scanner and were wrong by 23 edges; `dismech-rs`'s
`dismech_oracle_census` over `graph::build_causal_graph` gives 2,512/1,466
in under a second, and a pyyaml structural parse agrees on every figure.
- **92 `INDIRECT_UNKNOWN_INTERMEDIATES` edges DO name mediators** — the
source contradicting its own label. Neither oracle nor restraint control.
- **Deferred:** the third bucket for the 1,466 label-only edges and the 92
contradictory ones (operator decision, blocks any gold set); grounding
against the baked ontology identity space (measured complementary, not
superior — see below).
- **Docs:** `EPIPHANIES.md` ×3 —
`E-DISMECH-KNOWN-INTERMEDIATES-ARE-PROSE-NOT-IDENTITIES-1` (with its
same-day ⊕ correction), `E-ACADEMIC-CARVE-UNDERFILLS-ROWS-ARE-NOT-WORDS-1`,
`E-ABBREVIATION-GREP-MANUFACTURED-AN-ABSENCE-1`; `LATEST_STATE.md`;
`plans/causal-rung-standing-wave-v1.md` dated correction with the original
claim retained.
- **Review:** codex 2×P1, both CONFIRMED and both fixed in `5577728` — the
impossible subset/corpus counts (a subset cannot exceed the whole; mine was
the wrong number, the subset's 2,497 was right all along), and "usable
oracle" being the wrong noun for an ungrounded population. CodeRabbit did
not review: the repo is under its 10-star auto-review threshold. Bugbot:
usage limit.
- **Confidence:** High for every count (two independent structural parsers
agree, and the census line reproduces `E-DISMECH-CORPUS-CENSUS-1` exactly).
Medium for the grounding ratios — the rungs are heuristic label matching and
the 40.7% residue is reported, never dropped.

## 2026-08-20 — lance-graph #974 (MERGED, merge commit 8a93423, head b38fe21) — DisMech compact evidence vocabulary + typed citation sidecar

> Arc entry written 2026-08-21 (late — the obligation was missed at merge and
> is recorded here rather than silently skipped; the Termination clause does
> NOT apply, this PR shipped types).

- **Added (`0a99c35`), `lance-graph-contract::dismech_evidence`** (new
zero-dep module): `DismechTopology` (the four measured `causal_link_type`
states, 2 bits, `from_source`/`as_source`/`to_bits_2`/`from_bits_2`,
`source_knows_intermediates()` / `mediator_unresolved()` /
`topology_unresolved()`); `Supports` (4, 2 b); `EvidenceSource` (5, 3 b);
`CitationNamespace` (PMID/DOI/ORPHA/NCT/CGGV/URL); `CitationKey`
(`Identified{namespace,id}` | `ContentAddressed(ContentId)`);
`BibliographyRecord{key, title: ContentId}`.
- **Added (`e4a3cba` + `26c554e`):** S3.0 closed as NOT-NEEDED — the proposed
address type withdrawn, the retraction of #973's overclaim KEPT. A withdrawn
proposal whose retraction survives is the honest shape.
- **Added (`f08750f`):** `I-STRINGS-ARE-CAM-INDEX-ONLY-1` (operator ruling).
- **Locked:**
- **SOURCE-SIDE ONLY — deliberately does NOT reference `CausalEdge64`.** The
durable causal overlay must not become a pile of hot reasoning registers
(operator ruling); `DismechTopology → CE64 bits 59..60` happens at
HYDRATION, in the consumer, as a 1:1 read of `to_bits_2()`.
- **Every parse FAILS CLOSED.** `UNKNOWN` is a value the corpus asserts 408
times, so minting it from a parse failure would forge an assertion the
source never made — pinned by
`unrecognised_topology_fails_closed_and_never_becomes_unknown`.
- **Citation identity never derives from the title** — pinned by
`reference_identity_survives_a_title_rewording`, the falsifier that matters
for an LLM-generated corpus. Where no stable identifier exists,
`CitationKey::ContentAddressed` says so rather than synthesising one.
- **Superseded by #975:** the doc claim that `source_knows_intermediates()`
"separates the oracle population from the restraint control". The predicate
is right; the population is 2,512, not 3,978.
- **Docs:** `LATEST_STATE.md` Contract Inventory.
- **Review:** codex 2×P1, both fixed in `b38fe21`, plus a stale HP board claim
corrected in the same commit.
- **Confidence:** High — 1178/1178 at merge, every enum round-tripped
exhaustively.

## 2026-08-20 — lance-graph #971 (MERGED, merge commit 2cbe62d, head d627f5c) — four NARS kernels carved behind a maturity gate · Stage-2.5 census · Stage-2.6a V3 invariance · CE64 ⇄ V3 losslessness

- **Added — Stage 2 (`ed2fe8b`), `lance-graph-contract`:** `MaturityPolicy
Expand Down