docs(plan): record features 011 and 012 as Phases 10 and 9

[mbeacom]

What

Records the two newly-merged features in plan.md's spec-kit realization table. The table stopped at Phase 8 while #140 and #141 had already merged, so the ledger and the specs/ tree were out of sync — the exact condition that table exists to prevent.

Phase	Feature	Status
9 — Evaluator calibration	specs/012-evaluator-calibration/	scoped (#141)
10 — Probabilistic evaluator passes	specs/011-probabilistic-evaluator-passes/	scoped (#140)

The phase numbers are inverted, deliberately

Feature 012 is the lower phase, and its feature directory is numbered higher. That is not a mistake, and a note under the table says so.

012 gates 011: no probabilistic pass may ship without a holdout frozen before that pass produced its first score (ADR-0027 §3). plan.md already carries the rule that implementation of a phase MUST NOT begin until the phase beneath it has landed. Numbering calibration as the lower phase makes that existing rule enforce the dependency mechanically, instead of leaving it to prose inside each spec that a reader may never connect to the rule.

The feature directories run the other way because both were scoped concurrently.

Phase 7+ deferral corrected

"LLM rubric passes" was still listed as deferred, which the new rows contradicted. Removed — but their implementation is still blocked, now on named gates rather than on deferral:

• feature 012's frozen holdout, with its precondition gate observed failing (ADR-0016)
• an accepted record ratifying the harness-driven architecture (adrkit emits a prompt bundle and never opens a socket, which keeps Principle II intact)

Neither is satisfiable today, which is the correct state for a scoped feature.

Figures were counted, not carried

Every number in the two rows was derived from the merged specs rather than copied from a session report — this PR's predecessor shipped a fabricated duration ("eleven months" for a 25-day span) precisely because a figure was asserted and then repeated:

• 011 — 29 FRs, 17 SCs, 42 tasks, four open [NEEDS CLARIFICATION]
• 012 — 52 FRs, 53 SCs, 77 tasks, 36 [OBSERVE-FAIL], six distinct [NEEDS CLARIFICATION] labels

Counting distinct labels rather than token occurrences matters here: 012's spec contains more occurrences than questions, and 011's contains a prose sentence naming the token. Both rows state scoped per ADR-0014 and nothing above it.

Verification

• All four relative links in the new content resolve against the tree
• adr lint — 27 records, 0 errors, 0 warnings

Documentation only; no code, no corpus changes.

[Copilot]

Pull request overview

Updates the phase ledger to reflect recently scoped evaluator work.

Changes:

• Adds calibration as Phase 9 and probabilistic passes as Phase 10.
• Documents their inverted feature numbering and dependency.
• Removes LLM rubric passes from deferred work.

---

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.