docs(workbench): consolidate review + IA reorg into Wave M Phase A (re-sequenced before C)

Boils the design review and the information-architecture decision into one
integration. Phase B's heart (calibration/serving discipline) is done;
Phase A (Structure & Polish) now runs BEFORE resuming Phase C because two of
its items are structural prerequisites the Cognition cluster lands into.

Key synthesis: the command-palette drift bug and the grouped-nav idea are
the SAME fix — navigation derives from one registry that also encodes
structure. Standing IA constraint: one workbench, one address space, one
Chain Rail; grouping is a wayfinding skin, never an architectural fork; no
separate workbenches (the single evidence model is the thesis); 'levels' are
depth-within-a-surface, not top-level categories.

Phase A brief pack (wave-M-phaseA-briefs):
- A1: navigation registry (one source of truth) + grouped nav by the
  organism's loop (Converse/Cognition=core-logos/Evidence/Determinism/
  Discipline/Substrate/Settings) + palette fix (Demos+Calibration are
  currently unreachable via ⌘K; the ⌘1-0 map is stale)
- A2: Calibration earned-state — the centerpiece undersells (reads the
  sub-N_MIN report.json; the earned license lives in the disagreeing
  ratification_queue). Primary fix: regenerate coherent practice artifacts;
  fallback: reader surfaces both with provenance
- A3: Doctrine station ('how this UI can't lie') — elevate contracts/checks
  into a first-class surface; trails, larger

Parked: B4 (needs engine-side license stamping; no re-derivation),
Calibration EvidenceSubject kind.

docs/handoff/wave-M-phaseA-briefs-2026-06-13.md

@@ -0,0 +1,128 @@
+# Wave M · Phase A — Structure & Polish Brief Pack
+
+Date: 2026-06-13
+Plan: `docs/workbench/wave-M-worthiness.md` § Consolidated re-sequencing.
+Runs **before** resuming Phase C — A1/A2 are structural prerequisites the
+Cognition cluster lands into.
+
+## Standing constraints
+
+- Worktree off fresh `origin/main`; green-local (`pnpm build && pnpm test`,
+  Python lane where touched) before push; **STOP after checks green; Shay
+  merges.** Token-only (hexScan); schema/enum/conformance gates; no new
+  mutation endpoints; no invented data.
+- **One workbench, one address space, one Chain Rail.** Grouping is a
+  wayfinding skin, never an architectural fork. Do not split into separate
+  apps; do not break the Evidence Address or cross-route co-addressability.
+
+## Order
+
+A1 ∥ A2 (disjoint files) → A3 trails (larger). Each is its own PR.
+
+---
+
+## Brief A1 — Navigation registry (one source of truth) + grouped nav + palette fix
+
+Kills a live bug **and** delivers the grouped IA in one move: today
+`LeftNav.NAV_ITEMS` and `CommandPalette.NAV_COMMANDS` + `NAV_PATHS` are two
+hardcoded lists that have drifted — **Demos and Calibration are reachable
+from the nav but not via `⌘K`**, and the `⌘1–0` map is stale. The fix is the
+registry-driven move (same principle that made KeyboardHelp honest).
+
+### Read first
+- `src/app/LeftNav.tsx` (`NAV_ITEMS`), `src/design/components/primitives/CommandPalette.tsx`
+  (`NAV_COMMANDS`, `NAV_PATHS`), `src/app/Shell.test.tsx` (the nav assertion),
+  `src/app/shortcutRegistry.ts` (the registry pattern).
+
+### Deliverables
+1. `src/app/routes.ts` (new): the single source of truth —
+   `export const WORKBENCH_ROUTES: RouteSpec[]` where
+   `RouteSpec = { id: string; label: string; path: string; section: Section;
+   shortcut?: string }` and
+   `Section = "Converse" | "Cognition" | "Evidence" | "Determinism" |
+   "Discipline" | "Substrate" | "Settings"`. Group the current routes:
+   Chat→Converse; Trace/Runs/Audit/Vault→Evidence; Replay/Demos→Determinism;
+   Evals/Calibration/Proposals→Discipline; Packs→Substrate; Settings→Settings.
+   The **Cognition** section is declared now (empty) so Phase C slots in.
+   Keep `shortcut` only on the routes that warrant a `⌘` (don't force 12 into
+   `⌘1–0`) — primary picks get `⌘1…⌘9`; the rest are palette-only.
+2. `LeftNav` renders `WORKBENCH_ROUTES` **grouped by section** (section
+   label + its items, in `Section` order); empty sections render nothing.
+3. `CommandPalette` derives its navigate commands and the path map from
+   `WORKBENCH_ROUTES` — delete `NAV_COMMANDS`/`NAV_PATHS`. Demos + Calibration
+   become reachable; shortcuts come from the registry.
+4. `Shell.test` nav assertion updated to the grouped structure; add a test
+   that **every** `WORKBENCH_ROUTES` path is reachable from the palette (so
+   drift is caught by a test, not a human).
+5. Constraint: routing, the Evidence Address, and the Chain Rail are
+   untouched — this is nav presentation only.
+
+### Verify: `pnpm build && pnpm test`
+
+---
+
+## Brief A2 — Calibration earned-state (the centerpiece must show its thesis)
+
+On committed data the Calibration route shows three classes all "not yet
+licensed", empty bars — it never shows a class crossing θ. The reader reads
+`evals/gsm8k_math/practice/v1/report.json` `per_class`, whose committed copy
+is a sub-`N_MIN` baseline (`additive` committed=0), while the *earned* state
+(`additive` committed=100, measured 0.861, PROPOSE-licensed) lives in the
+separately-committed `ratification_queue.json` — and the two disagree
+(correct 3 vs 95, committed in different commits).
+
+### Primary fix (preferred — coherent data, reader unchanged)
+Regenerate the committed practice artifacts from **one coherent run** via the
+sealed practice runner (`evals/gsm8k_math/practice/v1/runner.py` —
+`build_report` / `write_report`, and the ratification-queue generation) so
+`report.json` `per_class` and `ratification_queue.json` agree, and the
+unchanged reader surfaces the earned class. **Trust boundary:** sealed
+practice, deterministic regen, reviewed; practice regime may carry wrong>0
+(attempt-and-eliminate) — that is NOT the serving wrong=0; do not conflate or
+weaken either. Verify post-regen: the two artifacts agree, and
+`read_calibration_classes()` shows ≥1 licensed class.
+
+### Fallback (workbench-only, if regen is out of scope)
+Extend the B1 reader to surface **both** artifacts with provenance: the
+`per_class` ledger AND an "earned licenses" set parsed from
+`ratification_queue.json` `proposals` (class_name, committed, measured,
+required, action), each labeled by its source so the two are never silently
+merged. The route gains an "Earned licenses" panel. Honest, but messier than
+coherent data — prefer the primary.
+
+### Verify
+`PYTHONPATH=$PWD .venv/bin/python -m pytest tests/test_workbench_calibration.py -q`
+plus, for the primary, a check that the two committed artifacts agree.
+
+---
+
+## Brief A3 — Doctrine station ("how this UI can't lie") — trails; larger
+
+A read-only surface that elevates the "contracts/checks" instinct into a
+first-class station: list the doctrine gates and the load-bearing invariants,
+each with *what it proves* and a pointer to its executable check. Every other
+AI UI asks for trust; this shows the proofs.
+
+### Deliverables (sketch — full brief authored when A1/A2 land)
+- New route under the **Cognition/meta** section: a static-but-real catalog of
+  the gates — hexScan (token-only), schemaDrift (TS mirrors the engine schema,
+  both snapshots), enumCoverage, route conformance (ADR-0162 §6), golden-file
+  layout (DAG) — and the invariants — `wrong=0`, `versor_condition < 1e-6` —
+  each row: name · what it proves · the file/command that executes it · doc
+  (ADR) reference.
+- Data is read from a small committed manifest of gates (so the station can't
+  drift from reality silently — the manifest itself is conformance-checked
+  against the test files it names). No invented "all green" badges: a gate is
+  listed because its check exists, not because we assert it passes.
+- Read-only; no execution from the UI.
+
+### Verify: `pnpm build && pnpm test`
+
+---
+
+## After this pack
+
+Resume **Phase C** (Cognition / core-logos cluster: pipeline visualizer,
+field substrate, identity, contemplation) into the grouped structure A1
+created. B4 stays parked (engine-side license stamping). Then D (tour),
+E (continuous).

docs/workbench/wave-M-worthiness.md

@@ -172,8 +172,88 @@ Detailed brief pack: `docs/handoff/wave-M-phaseB-calibration-briefs-2026-06-13.m
 - No timelines — phases/priorities/scope-sizes; sequencing is the dependency
   DAG, not a clock.
 
-## Execution order
+## Consolidated re-sequencing (amended 2026-06-13, after Phase B heart landed)
 
-**B → C → D**, with **A in parallel**. The worthiness gap is widest at B; the
-tour (D) lands hardest once B and C exist to show off. Phase B brief pack is
-authored first (this commit); subsequent phase packs follow as each lands.
+Phase B's heart is **done** — B1 (#724 readers), B2 (#725 Calibration route),
+B3 (#726 wrong=0 frame) make the serving-discipline loop visible. A design
+review of that work + an information-architecture decision now fold into a
+single **Phase A (Structure & Polish) that runs BEFORE we resume Phase C** —
+because two of them are structural prerequisites the Cognition cluster (C)
+should land *into*, not extend a flat list past.
+
+The governing insight: **the command-palette drift bug and the
+grouped-navigation idea are the same fix.** Navigation must derive from one
+registry that also encodes structure; build that once.
+
+### Standing IA constraint (binds everything below)
+
+**One workbench, one address space, one Chain Rail.** Grouping is a wayfinding
+skin, never an architectural fork. A calibration class and the trace it
+relates to stay co-addressable. We do **not** split into separate workbenches
+— the single evidence model is the thesis. "Levels" are expressed as *depth
+within a surface* (calm default / infinite depth), not as top-level
+categories.
+
+### Phase A — Structure & Polish (NEXT; runs before resuming C)
+
+- **A1 — Navigation registry (one source of truth) + grouped nav + palette
+  fix.** Today `LeftNav.NAV_ITEMS` and `CommandPalette.NAV_COMMANDS`/`NAV_PATHS`
+  are two hardcoded lists that have **already drifted** — Demos (#723) and
+  Calibration (#725) are in the nav but unreachable via `⌘K`, and the `⌘1–0`
+  map is stale. Replace both with one `routes` registry `{ id, label, path,
+  section, shortcut? }`; LeftNav renders it **grouped by section**;
+  CommandPalette derives its nav commands and the `⌘` map from it (drift
+  becomes structurally impossible — the same registry-driven move that made
+  KeyboardHelp honest). Sections follow the organism's loop:
+  **Converse · Cognition (core-logos) · Evidence · Determinism · Discipline ·
+  Substrate · Settings.** Current routes group as: Chat→Converse;
+  Trace/Runs/Audit/Vault→Evidence; Replay/Demos→Determinism;
+  Evals/Calibration/Proposals→Discipline; Packs→Substrate; Settings→Settings.
+  The **Cognition** group is created now (empty or near-empty) so Phase C's
+  surfaces slot into a home rather than a flat tail.
+- **A2 — Calibration earned-state (the centerpiece must show its thesis).**
+  On committed data the Calibration route shows three classes all "not yet
+  licensed" with empty bars — it never shows a class crossing θ, because the
+  reader reads `practice/v1/report.json` `per_class`, whose committed copy is
+  a sub-`N_MIN` baseline, while the *earned* state (`additive` committed=100,
+  measured 0.86, **PROPOSE-licensed**) lives in the separately-committed
+  `ratification_queue.json` — and the two artifacts disagree (correct:3 vs 95,
+  different commits). **Primary fix:** regenerate the committed practice
+  artifacts from one coherent run via the sealed practice runner so
+  `report.json` per_class and the queue agree and the reader (unchanged) shows
+  the earned class — deterministic regen, reviewed, no metric weakening.
+  **Fallback** if regen is out of scope: the reader honestly surfaces *both*
+  artifacts with provenance (the per_class ledger AND an "earned licenses"
+  panel from the queue), each labeled by source. Either way the disagreement
+  is resolved and the route shows its moment.
+- **A3 — Doctrine station ("how this UI can't lie").** Elevate the
+  "contracts/checks" instinct into a *surface*, not a folder: a read-only
+  station that lists the doctrine gates (hexScan, schemaDrift across both
+  snapshots, enumCoverage, route conformance, golden-file layout) and the
+  load-bearing invariants (`wrong=0`, `versor_condition < 1e-6`), each with
+  *what it proves* and a pointer to its executable check. Every other AI UI
+  asks for trust; this one shows the proofs. Larger scope than A1/A2 — may
+  trail into/after C, but it is the single most novel "worthy of the model"
+  surface and belongs in the Cognition/meta neighborhood.
+- Remaining Phase-A polish (density pref wiring, DAG's other two consumers,
+  accessibility pass, tabular-nums/balance sweep) continues as before,
+  parallel-safe.
+
+Brief pack: `docs/handoff/wave-M-phaseA-briefs-2026-06-13.md`.
+
+### Parked / deferred
+
+- **B4 (per-turn leeway attribution)** is **blocked**: a served turn carries
+  no calibration-class/license/θ. It needs an **engine-side** change to stamp
+  ReachLevel/class/θ onto served results (its own ADR) — then B4 is a trivial
+  display. Do not re-derive in the workbench (theater).
+- **Calibration EvidenceSubject kind** (the route uses local selection today)
+  — a conscious deferral, not an accident; revisit if a calibration class
+  needs to thread the Chain Rail.
+
+### Order
+
+**A (A1 → A2 in parallel; A3 trails) → resume C (Cognition / core-logos
+cluster) → D (tour) → E (continuous).** A1+A2 land before C so the Cognition
+surfaces arrive into a grouped structure that already shows discipline
+honestly.