Handover: alpha-channel plan session (post-#978)

.claude/board/EPIPHANIES.md

@@ -1,3 +1,101 @@
+## 2026-08-21 — E-HHTL-IS-MINTED-IN-THE-ARTIFACT-NOBODY-CITES-1 — "zero on every baked row in both production bakes" is precise about the two it names and silent about the third, where the five OBO namespaces are 100% minted
+
+**Status:** FINDING (measured directly on the pinned `.soa` bytes, SHA-256
+verified against `MedCare-rs/data/config/bakes.tsv`). **Confidence:** High —
+every number is a count over 512-byte rows, tiers read at bytes 4..10.
+
+The board headline (`INTEGRATION_PLANS.md` ARC-B entry) and
+`EPIPHANIES.md:899` both state HHTL is **"zero on every baked row in both
+production bakes"**, citing `ogar-obo` (68,797 rows) and MedCare's
+`join-map.md` (68,797 rows). Both citations are correct. **Both describe the
+same artifact set of two.** A third pinned artifact exists:
+
+| artifact | rows | HHTL != 0 |
+|---|---:|---:|
+| `obo-core.soa` | 68,797 | **0 (0.00%)** |
+| `spine.soa` | 7,641 | **0 (0.00%)** |
+| `all-lanes.soa` | 770,360 | **164,031 (21.29%)** |
+
+Per lane in `all-lanes.soa`: **MONDO `0x0301` 32,095 rows 100% · HPO `0x0302`
+19,836 100% · UBERON `0x0303` 14,975 100% · PATO `0x0304` 1,887 100% ·
+ICD-10-GM `0x030F` 16,905 100% · OMIM `0x0319` 18,712 100% · OPS `0x0311`
+98.9% · Orphanet `0x0317` 67.6%**; LOINC / CUI / RxNorm / FMA ~0%.
+
+**Why this matters and is not a footnote:** the five 100%-minted namespaces are
+exactly the ones a DisMech overlay grounds against. The generalized claim
+("HHTL is dormant, the gap is mint + read") licenses a session to skip HHTL
+entirely; the measured claim says HHTL is available **if a reader names
+`all-lanes.soa`** and unavailable **if it names `obo-core.soa`**. So *which
+artifact a consumer reads* is a first-class design decision, not a detail — and
+a ladder level claiming "+ HHTL topology" must state its artifact or it is
+measuring nothing.
+
+**The generalization error is the transferable part.** Two independent
+citations agreeing is evidence about the *rows they counted*, not about the
+artifact set. Two sources counting the same 68,797 rows is ONE measurement
+reported twice. Before promoting "on every row" from "on these rows", enumerate
+the artifacts, not the citations.
+
+**⊘ CORRECTION 2026-08-21 (same day, operator-prompted) — the entry above
+measured ONE of TWO readings.** Operator: *"Obo HHTL ist meines Wissens mit
+zipper bereits indirekt hydriert."* Correct. The census above counted the
+**cascade tiers** (bytes 4..10); `rails::HhtlMode::of_row` PREFERS the
+**RailHead** reading and uses Cascade only as the fallback. Measured on the
+Zipper rail registers (`rails.rs:130-147`): MONDO **32,094/32,095**, HPO
+**19,835/19,836**, UBERON **14,973/14,975** (plus 8,525 `part_of`, the only
+lane with mereology), PATO 1,886/1,887 — median logical-DN depths 6/7/8/5, and
+264 rows deep enough to use the continuation slab. So the OBO hierarchy IS
+hydrated, indirectly, exactly as stated. `obo-core.soa`/`spine.soa` remain zero
+on BOTH readings. New in this correction: cascade and rail are **independent** —
+Orphanet (14,063 cascade) and OMIM (18,712) carry **zero** rail DN, so a
+prefix-containment consumer silently gets depth 0 there. **The lesson compounds
+the entry's own:** it is not enough to enumerate the artifacts — a claim about
+a field must also name which READING of it was counted, when the accessor
+picks between two registers. Full table: plan §8a ⊘ correction.
+
+Cross-ref: `.claude/plans/dismech-causality-v3-v1.md` §8a; ARC-B
+`docs/architecture/ARC-B-OWNERSHIP-AND-ADDRESSING-REASSESSMENT.md:23` (regraded
+in place: its conclusion holds for `obo-core`/`spine`, needs the `all-lanes`
+qualifier); `EPIPHANIES.md:899`.
+
+## 2026-08-21 — E-THE-ORACLE-POPULATION-IS-64-PERCENT-AND-A-GATE-HARDCODES-THE-OTHER-36-1 — a third of the "known intermediates" name no intermediate, and the gate that would have caught it asserts the wrong number
+
+**Status:** FINDING (measured three independent ways on `/workspace/dismech`
+@`557e15436`, 1,968 disorder files). **Confidence:** High — the three methods
+agree exactly.
+
+The DisMech supervision story rests on `INDIRECT_KNOWN_INTERMEDIATES` being a
+population where "the source names the mediators, so they can be hidden and
+recovery measured" (`E-DISMECH-CORPUS-CENSUS-1`). Measured, **only 2,449 of
+3,825 (64.0%) such edges actually name one.** 1,376 (36.0%) carry the label and
+an absent or empty mediator list.
+
+Three methods, agreeing: (a) a line-oriented per-edge walk — **2,449** edges
+over **534** disorder files, 3,714 mediator strings; (b) key-occurrence count —
+`intermediate_mechanisms:` appears **2,525** times with **0** inline empty
+lists, and 2,525 = 2,449 + 74 + 1 + 1, exactly the per-bucket split;
+(c) `contract::dismech_evidence.rs:155-176` at the narrower
+`pathophysiology[].downstream[]` scope — **1,347 of 3,844 (35.0%)** empty.
+
+**Consequences.** The held-out corpus is **2,449 edges over 534 diseases**, so
+a 20% split is ~490 edges / ~107 diseases, not the 774 a 3,869 denominator
+implies. And **74 `INDIRECT_UNKNOWN_INTERMEDIATES` edges DO name mediators** —
+a source contradiction the four-label taxonomy does not anticipate; they must
+be removed from any restraint control or they will read as hallucinated
+closure by the benchmark's own definition.
+
+**The gate that should have caught this hardcodes the wrong number.**
+`MedCare-rs/.claude/plans/dismech-missing-links-v1.md` Gate W1.1 asserts
+`known_links.tsv == 3.869` — unsatisfiable on any corpus revision. Its own
+instruction is *"stoppen und melden, nicht die Zahl anpassen"*, so this entry
+is the report, not an edit.
+
+**Transferable:** a label that asserts a property and a field that carries it
+are two different measurements. Counting the label is not counting the data —
+and a supervision corpus sized from the label is oversized by exactly the rows
+where the source labelled but did not fill.
+
+Cross-ref: `.claude/plans/dismech-causality-v3-v1.md` §3a; `E-DISMECH-CORPUS-CENSUS-1`.
 ## 2026-08-21 — E-ABBREVIATION-GREP-MANUFACTURED-AN-ABSENCE-1 — I reported a shipped 15-module subsystem as non-existent because `fn .*ppr` matches `approx`, and a `head` limit hid the real hits
 
 **Status:** FINDING (self-inflicted, caught by the operator pointing at

.claude/board/INTEGRATION_PLANS.md

@@ -1,3 +1,66 @@
+## 2026-08-21 — ALPHA-CHANNEL RUNG OVERLAY (scraping, not a new arc)
+
+`.claude/plans/alpha-channel-rung-overlay-v1.md` — the design for the ONE
+empty row of the thinking table (`hhtl-thinking-tables-le-contract-v1.md`
+§2.3, row **Rung ladder**: *"(unassigned) · unminted, undesigned"*), scraped
+from an operator brainstorm the same day: the Photoshop **alpha channel** as
+an ephemeral layer over the ontology carrying the residue of a search, rung
+levels 2-10 as layers, second-order thought at the same address in a separate
+thinking table, a mask over the activities, focus of attention.
+
+**Headline: six of the nine pieces already exist or are already planned, and
+the plan mints no new type.** `PhaseCensus` (kanban_actor) is the activity
+mask, shipped. The five tactics + `ReasoningGap` are the elimination search,
+shipped (V1). Rung layers are dialectic **V3**; 64k is dialectic **V4** and
+stays behind its own gate (*"only after V0-V3 green at small scale"*). The
+residue carrier may be the shipped `attention_mask.rs` -- unaudited, so
+D-ACR-0 is an audit before anything is built on it.
+
+**The one genuinely missing primitive is `RowFocusMask`** (S3.1b): named on
+this board, absent from every crate (grep hits only `STATUS_BOARD.md` and one
+handover). D-ACR-1 is that primitive and everything else queues behind it.
+
+**Two things the plan refuses.** No new address type -- S3.0/PR #973 was closed
+at exactly this spot (*"CLOSED -- NOT NEEDED (use `IdentityQuad` / `ClassAddr`
+/ V3 rail)"*, the empty column being HYDRATION not ADDRESS), and this overlay
+is hydration over existing addresses. No CE64 bit -- 59..63 is
+`TRUTH_SHIFT`+`SPARE_SHIFT`, the band set ONLY by an explicit
+`with_reasoning_band()` call, and an overlay must not become a fifth
+derivation path into it.
+
+**The contamination boundary is not new machinery.** *"Damit der Graph nicht
+von Patienten kontaminiert wird"* is the one-writer-per-mailbox rule already
+ratified: the overlay's owner is the session mailbox, so a patient-derived
+write to an ontology row has no owner that could perform it. The invariant is
+one-directional -- the overlay reads the graph, the graph never reads the
+overlay -- which is also what makes the residue safe to discard whole.
+
+Graded a **pruner, never a proof**, the same grade `ONTOLOGY_BAKE_STATE.md`
+already gives HHTL.
+
+## 2026-08-21 — DISMECH × CAUSALITY-V3 REBASE REPORT (report first, no code)
+
+`.claude/plans/dismech-causality-v3-v1.md` — the operator-required §26
+deliverable that gates implementation of the DisMech/Causality-V3 arc. Twelve
+sections, every number carrying the command or `file:line` that produced it;
+anything not personally measured is labelled *claimed, unverified*.
+
+**Headline:** the expensive substrate really is largely built — AriGraph
+(8,750 LOC / 187 tests), CLAM+CHAODA (4,900 / 77), Aerial+ ARM discovery,
+HHTL, the OCR+DOM→doc.v1 convergence — but it is **read-rich and write-empty at
+four independent layers** (CE64 bits 53-63, `CausalEdgeV3`,
+`CausalWitnessFacet`, `dismech_evidence`), all with zero production writers.
+Three §0 claims are stale, two are wrong: the oracle population is **64%** of
+its label count (`E-THE-ORACLE-POPULATION-IS-64-PERCENT-...-1`), and HHTL is
+**100% minted** for the five OBO namespaces in the artifact the board does not
+cite (`E-HHTL-IS-MINTED-IN-THE-ARTIFACT-NOBODY-CITES-1`).
+
+**Sequence:** D-CV3-0..2 (pin corpus, freeze splits, Level-0 scorer) must be
+green BEFORE D-CV3-3 mints a `HoleV3` tenant — a carrier minted before its
+benchmark becomes the fifth EXISTS-UNCALLED entry. `HoleV3` is `ValueTenant =
+16` (15 is reserved for `BoardAggregates`), never CE64, which has **zero free
+bits**.
+
 ## 2026-08-19 — ARC-B OWNERSHIP + ADDRESSING REASSESSMENT (supersedes parts of the plan wave)
 
 Not a plan — the **assessment that corrects the plan wave**, per two

.claude/board/STATUS_BOARD.md

@@ -28,6 +28,38 @@ relations (MedCare-rs commitment #10) — an OGAR/lance-graph classid-mint
 capacity question for the operator.
 
 ---
+
+### D-CV3-* — DisMech × Causality-V3 (plan `dismech-causality-v3-v1.md`, 2026-08-21)
+
+| D-id | Scope | Repo | Status | Falsifier |
+|---|---|---|---|---|
+| D-CV3-0 | Pin corpus (`/workspace/dismech` @`557e15436`) + emit 3 frozen TSVs from a typed parse; no new types | MedCare-rs | **Queued** | fresh container reproduces 2,449 / 4,076 / 361 exactly |
+| D-CV3-1 | Splits A (random edge) + B (disease-held-out, 534 groups) as committed artifacts | MedCare-rs | **Queued** — gates on D-CV3-0 | group-disjointness; held-out share 15-25% |
+| D-CV3-2 | Level-0 scorer: Recall@K + MRR + abstention, structural only | MedCare-rs | **Queued** — gates on D-CV3-1 | non-trivial on BOTH arms; can-fire + can-stay-silent pair |
+| D-CV3-3 | `HoleV3` as `ValueTenant = 16`; `awareness_state` orthogonal to `unknown_kind`; NOT in CE64 (0 free bits) | lance-graph | **Queued** — gates on D-CV3-2 green | field-isolation matrix; `ENVELOPE_LAYOUT_VERSION` unchanged; two-axis independence round-trip |
+| D-CV3-4 | Producer: `dismech_evidence` -> hole rows (its first caller) | lance-graph | **Queued** — gates on D-CV3-3 | populated rows 0 -> 4,076 + 361, else a 5th EXISTS-UNCALLED carrier |
+| D-CV3-5 | `Communities` x `EpisodicBasins` cross-validation (the only available unknown-unknown detector) | lance-graph | **Queued** | fires on a synthetic bridge AND stays silent on a coherent graph |
+| D-CV3-6 | Call `reciprocal_rank_fusion` in `OsintRetriever::retrieve` (cheapest real integration) | lance-graph | **Queued** — gated on G0 | fused ranking differs from BFS-only on >=1 real query |
+
+
+### D-ACR-* — Alpha-channel rung overlay (plan `alpha-channel-rung-overlay-v1.md`, 2026-08-21)
+
+Fills `hhtl-thinking-tables-le-contract-v1.md` §2.3's empty **Rung ladder**
+row. Mints no type. Six of nine brainstorm pieces already had homes; these are
+the rest.
+
+| D-id | Scope | Status | Falsifier |
+|---|---|---|---|
+| D-ACR-0 | Audit `attention_mask.rs`/`attention_mask_actor.rs`: residue carrier, or a name collision? Report only | **Next** | names a caller, or records EXISTS-UNCALLED |
+| D-ACR-1 | `RowFocusMask` — the one missing primitive (S3.1b names it; no crate contains it) | Queued — gates on D-ACR-0 | can-fire AND can-stay-silent on non-trivial input |
+| D-ACR-3 | The one-way invariant as a test: no ontology-owned write traces to a patient-tagged read through ANY call path (corrected from write-authorization-only after CodeRabbit found a session-derived value can flow to the ontology owner via a shared parameter/return, then be written as the owner's own act) | Queued — gates on D-ACR-1 | a write whose call graph includes a session-tagged read is the bug, even if the write itself is authored by the ontology owner |
+| D-ACR-2 | Mint the Rung-ladder rail | Queued — gates on operator mint decision (HTT §8 Q3) | `rail_carving` gains its first non-default consumer |
+| D-ACR-4 | Second-order row at the same address, separate table | Queued | a rung-2 read reconstructs where rung-1 looked, on a fixture with an independent answer |
+| D-ACR-5 | 64k lowering | **BLOCKED** — dialectic V4's own gate (V0–V3 green at small scale) | — |
+
+**Not claimed:** that the residue improves recall or finds needles. Graded a
+pruner, never a proof.
+
 ## preparation-arc plan wave — 2026-08-19 (operator: "integration plans for all open arcs")
 
 Five plans, each PROPOSED (no code — the reset charter's audit-first order