fix(learning-harness): the round-3 review findings, remediated RED-first

Eight of nine group-3 tasks were rejected. Each defect was hiding a number,
and the numbers are why the loop was not learning:

- The fixed side was judged against the *vulnerable* side's function
  ranges. 191 of 227 pairs with a resolvable labeled span (84%) have
  different spans on the two sides, so a leak was recorded as silence,
  always in the direction that flatters Youden.
- Round zero measured every family with the generalist. `family:unknown`
  earns no credit against any real family, so every family but `unknown`
  baselined at a structural zero and every noise floor with it.
- The cross-family sweep counted absolute failures instead of negative
  flips. On a corpus where most pairs fail at round zero that rejects
  every round for damage it did not do. Unscorable rows voted too.
- K runs was declared and never threaded; `pairs --k-runs` was parsed and
  discarded, and its only test grepped `--help`.
- The meta-agent proposer was a stub, so `learning round` could never
  propose anything. It now composes MetaAgent.evolve against a scratch
  copy of the family directory, and `--proposals FILE` runs a recorded
  sequence offline.
- `refused_holdout` was a vocabulary word no code path could produce, and
  the covering test accepted either outcome.
- Cost was metered once, after everything had been paid for; a stage that
  raised left no record at all.
- The corrected lever number was published beside the earlier one rather
  than in place of it, and the test named for that property never checked.
- Refusals were `continue` statements: `Refusal.degradation()` existed and
  nothing called it. A train-split identical twin was reported as
  `split:train`, which is untrue.

Test repairs that each changed a number: two split tests measured an
`x = 1` fixture that teaches nothing whatever the rule says; two CLI
scoreboard tests asserted against all-zero output produced by a stub that
answers twice and puts its finding outside the labeled function.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01T7XD7x5KonMdmGRkp63EQG

src/openultrasast/learning/publish.py

@@ -12,6 +12,7 @@ rather than appended, so the file cannot accumulate stale numbers beside fresh o
 from __future__ import annotations
 
 import json
+import re
 from collections.abc import Mapping
 from dataclasses import dataclass, field
 from datetime import UTC, datetime
@@ -69,6 +70,7 @@ def publish(*, learning_dir: Path, measurements_dir: Path, roadmap: Path, date:
         )
         artifacts.append(artifact)
     _replace_section(roadmap, table)
+    _supersede(roadmap)
     return PublishReport(command=COMMAND, table=table, families=families, models=tuple(sorted(models)), artifacts=tuple(artifacts))
 
 
@@ -169,6 +171,32 @@ def _cost(cost: float, correct: int) -> str:
     return f"{cost / correct:.2f}" if cost and correct else "-"
 
 
+# Claims the split correction supersedes, as they were written before holdout pairs stopped teaching. Req 5.3 says
+# the corrected number is published *in place of* the earlier one; a document that carries both asserts two numbers
+# for one measurement and the reader has no way to tell which run is current.
+_SUPERSEDED = (
+    re.compile(r"human holdout profile pair_correct \d+ -> \d+ of \d+, Youden [-+]?\d+ -> [-+]?[\d.]+(?:, round \d+ `[^`]+`)?\."),
+    re.compile(r"The lever recovered \d+ of \d+ holdout pairs at Youden [-+][\d.]+\."),
+)
+_SUPERSEDED_NOTE = (
+    "superseded by the learning-harness section below: that number was measured with holdout pairs teaching "
+    "(see `benchmarks/measurements/2026-09-06-mechanism-lever-split.json`)."
+)
+
+
+def _supersede(roadmap: Path) -> None:
+    """Rewrite every superseded lever claim to point at the regenerated section. Idempotent: the pattern is gone
+    afterwards, so a second run changes nothing."""
+    if not roadmap.is_file():
+        return
+    text = roadmap.read_text(encoding="utf-8")
+    updated = text
+    for pattern in _SUPERSEDED:
+        updated = pattern.sub(_SUPERSEDED_NOTE, updated)
+    if updated != text:
+        roadmap.write_text(updated, encoding="utf-8")
+
+
 def _replace_section(roadmap: Path, table: str) -> None:
     body = f"{BEGIN}\n\n## Classified detector numbers\n\n{table.rstrip()}\n\n{END}\n"
     text = roadmap.read_text(encoding="utf-8") if roadmap.is_file() else ""