docs(review-suite): recheck s2/s3 strata under grader 1.1 for the same surface-in-prose defect

[shaug]

Summary

• Follow-up recheck to fix(review-suite): grader surface-in-prose defect + v2 s1 re-score comparison #84's grader fix (surface named only in a finding's
  prose was invisible to finding_surfaces()): confirms whether the same
  defect reaches s2-solution-simplicity-lens and s3-code-simplicity-lens,
  neither of which fix(review-suite): grader surface-in-prose defect + v2 s1 re-score comparison #84 touched.
• v1's raw attempt data does not survive anywhere on this machine for these
  two strata (matching s1's documented pattern) - a fresh 40-attempt run
  (5/case, unchanged runtime/model pin: claude 2.1.92, claude-opus-4-6[1m])
  was required.
• Direct counterfactual re-grade of this run's own fresh attempts under the
  pre-fix (85ccf13) grader shows: three of the four material cases were
  already correctly routed to referred under the old grader (no material
  change); setup-service-path-gateway is materially affected by the same
  surface-named-only-in-prose defect, at the referred-vs-false-positive
  boundary (2/5 referred + 3/5 false-positive under the old grader vs. 5/5
  referred + 0/5 false-positive under the fix) - closely mirroring v1's own
  reported 4/5 false-positive rate for this exact case.
• All four negative controls unchanged (zero false positives, perfect
  stability, both eras).

Why

• The grader defect's reach beyond the two already-fixed s1 cases had never
  been checked for s2/s3, even though their material cases' non-empty
  ever_referred lists made them look less likely to be affected. Checking
  directly (rather than assuming) surfaces a real, smaller-scope effect on one
  case's false-positive rate that a report-only before/after comparison would
  have missed - this PR's document verifies the claim with a same-attempts,
  two-grader-versions counterfactual instead of just diffing two reports.

Scope guardrails

• Adds new files only under review-suite/evals/v2/; does not touch
  review-suite/evals/baseline/v1/, gate-manifest.json,
  DECISION-RECORD.md, FAILURE-TAXONOMY.md, or audits/.
• Does not re-run or touch s1-correctness-orchestrator.
• Reports facts only; does not edit any tracker issue or record a policy
  decision.

Spend

$2.15 of 40 attempts, against this task's $5.00 hard ceiling.

Test plan

• just format - all checks passed
• just lint - skills-ref validate + plugin packaging validation passed
• just test - 252 review-suite tests + all skill test suites passed
• One full review-code-change cycle (solution-simplicity, correctness,
  code-simplicity) - all three lenses clean; correctness independently
  re-derived every load-bearing numeric claim in the new document
  (including the counterfactual re-grade) from source data rather than
  trusting the PR description.

🤖 Generated with Claude Code