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

tests/test_learning_publish.py

@@ -107,13 +107,22 @@ def test_the_corrected_lever_number_replaces_the_earlier_one(tmp_path: Path) ->
         )
     )
     roadmap = tmp_path / "roadmap.md"
-    roadmap.write_text(ROADMAP + "\nThe lever recovered 7 of 17 holdout pairs at Youden +0.059.\n")
+    stale = "human holdout profile pair_correct 0 -> 6 of 17, Youden 0 -> +0.353, round 2 `no_proposals`."
+    roadmap.write_text(f"{ROADMAP}\nMeasured 2026-09-05 with the vibe-py candidates: {stale}\n")
     report = publish(learning_dir=out, measurements_dir=measurements, roadmap=roadmap)
     text = roadmap.read_text()
     assert "-0.0588" in report.table or "-0.059" in report.table
     assert "5 of 17" in report.table or "detected 5" in report.table
     assert "holdout pairs stopped teaching" in report.table.lower() or "train split only" in report.table.lower()
     assert text.count("learning-harness:begin") == 1
+    # Req 5.3 says *in place of* the earlier one. Leaving the old claim above the new section is how a document
+    # ends up asserting two different numbers for the same measurement.
+    assert stale not in text
+    assert "superseded" in text and "learning-harness" in text
+    assert "Measured 2026-09-05 with the vibe-py candidates:" in text, "the sentence is corrected, not deleted"
+    before = roadmap.read_text()
+    publish(learning_dir=out, measurements_dir=measurements, roadmap=roadmap)
+    assert roadmap.read_text() == before, "a second publish must be a no-op once the claim is corrected"
 
 
 def test_the_roadmap_section_is_replaced_not_appended_again(tmp_path: Path) -> None: