contract: qualia_profile — ICC-style named-qualia table (219 calibrated, D-QPROF-0)

.claude/board/ISSUES.md

@@ -1,3 +1,11 @@
+## ISS-QUALIA-A2B-CLUT-UNMEASURED — no measured transform from convergence 17D to the calibrated PCS 17D (2026-10-16)
+
+**Status:** OPEN. **Basis:** VERIFIED-IN-CODE (`contract::qualia_profile::SHARED_AXES` maps 5 of 17 axes by name; `device_to_pcs_shared` returns `measured = false` for the other 12). HYPOTHESIS: same-named axes mean the same thing.
+
+- The workspace has two 17-axis "qualia" vectors on different axes: `qualia::AXIS_LABELS` (convergence observables, computed) and `qualia_profile::PCS_AXES` (phenomenological, hand-calibrated, 219 patches). Only valence / arousal / tension / warmth / clarity coincide by name.
+- `engine_bridge::classification_distance` compares device telemetry to six Ekman archetypes in a 6-axis device subspace — observer vocabulary applied to mechanism signal, no PCS in between — and after D-CSV-5b the stored value is a constant `1.0`.
+- **What closes it:** the chart probe. Drive each of the 219 named patches through the thinking engine (Jina lens → convergence 17D), record the pairs, fit the A2B CLUT, and report per-axis Spearman ρ under Jirak's rate (`I-NOISE-FLOOR-JIRAK`). Shared axes with ρ below the floor falsify the HYPOTHESIS. The `.bin1024` file beside the JSON is the proposal side for Haspelmath adjacency (Hamming between patches), not the CLUT.
+
 ## ISS-SUPERVISOR-RESPAWN-BACKOFF-NOT-APPLIED — a crashed consumer actor is respawned at once and the respawn is rescheduled forever (2026-10-13)
 
 **Status:** RESOLVED 2026-10-13 by #1540 (merged `f2bf199`). The delayed cast is now the only respawn path and is scheduled from `ActorTerminated`; the `RespawnG` arm spawns once and does nothing if the slot is already live; the crash count is carried and restarts at 1 when the previous crash is more than five minutes old (spec §5.3); more than 10 crashes in that window escalate and the child stays down. TEST-PINNED in `tests/supervisor_respawn_backoff.rs` (one respawn per crash after its backoff; escalation; window reset), each disable-verified red. The bullets below describe the defect as found. **Was:** OPEN. **Basis:** VERIFIED-IN-CODE (`crates/lance-graph-supervisor/src/supervisor.rs`, `RespawnG` arm); found by the codex review of #1536.

.claude/board/LATEST_STATE.md

@@ -1,3 +1,12 @@
+## 2026-10-16 — D-QPROF-0 qualia profile (branch `qualia-profile-icc`, in PR)
+
+- **Contract delta (branch):** `lance_graph_contract::qualia_profile` + generated `qualia_profile_table` — the first named-emotion calibration set in the repo (219 rows from Upstash `ada:qualia:calibrated:*`, previously cited in `thinking-engine/src/qualia.rs` as "220 calibrated pairs in Upstash" and absent). Data: `data/qualia/calibrated_219.json` (+ `.bin1024` sign-binarized Jina embeddings), generator `scripts/gen_qualia_profile.py` (deterministic, rustfmt-stable, FNV-pinned).
+- Open: the device→PCS CLUT for the 12 non-shared axes is a chart probe, not a constant (`ISS-QUALIA-A2B-CLUT-UNMEASURED`). Σ-codebook slot assignment (64 glyphs vs 219 named) is NOT decided here; the module only defines the palette numbering `0 | 1..=219 | 220..=255 reserved`.
+
+### Current Contract Inventory — net delta (`qualia_profile.rs`, `qualia_profile_table.rs`)
+
+- `lance_graph_contract::qualia_profile::{PCS_AXES, PCS_DIMS, Pcs, NamedQualia, UNNAMED, RESERVED_FROM, named, index_of, pcs_l1, nearest_named, nearest_named_in, Gamut, RenderingIntent, Rendered, render, Mix, SHARED_AXES, device_to_pcs_shared}` and re-exports `{NAMED, NAMED_COUNT, PROFILE_ID, SOURCE_FNV, TABLE_FNV}` — an ICC-style profile for felt states. Three spaces kept apart: **device** (`qualia::AXIS_LABELS`, convergence observables), **PCS** (`PCS_AXES`, 17 phenomenological axes, 219 calibrated patches in hundredths as `[i8; 17]`), **observer** (e.g. the six Ekman archetypes). `nearest_named` is B2A on the PCS side and its residual is the K channel — the `classification_distance` the i4 cutover stopped storing, measured against the inner vocabulary. `Gamut` is a 256-bit mask per rendering device; three intents adapted to a discrete palette (absolute / relative-in-family / perceptual); `Mix` = up to three inks (what `ada:qualia:active` carries). Only the five same-named axes (valence, arousal, tension, warmth, clarity) have a device→PCS mapping, and that is a CONJECTURE until the chart probe runs. 12 tests, each with a stated falsifier (`TABLE_FNV` and `SOURCE_FNV` mutations both verified red — the pair (source, table) is pinned from both sides, after codex P2 on the first push; intent inertness; anti-vacuity on gamut and nearest).
+
 ## 2026-10-15 — main after #1567 (merged `b1f1365d`); D-HOLO H6c/H14 and the writer's version clock on main
 
 - **On `main` (#1566, merged `c18aefcb`):** D-HOLO H6c and H14 probes and results; H14b open (awaits the operator).

crates/lance-graph-contract/src/lib.rs

@@ -147,6 +147,11 @@ pub mod plan;
 pub mod property;
 pub mod proprioception;
 pub mod qualia;
+/// ICC-style profile for felt states: the 219-row named-qualia table (PCS), palette index,
+/// K residual, gamut + rendering intents, mix. See the module doc for the three spaces.
+pub mod qualia_profile;
+#[doc(hidden)]
+pub mod qualia_profile_table;
 pub mod rail_geometry;
 pub mod rbac;
 /// RBAC hot-plug: the [`hotplug`] pattern for authorization — a consumer's