LENSCI: re-measure v4 lens witnesses, wire 86 discriminating-green --claim-run rows as must-pass (per-witness perturb-red)

[briansrls]

LENSCI: real lens-analysis CI gate — re-measured & expanded

Re-measured every src/v4/test/claim/lens_* witness and wired the discriminating-green ones as must-pass rows in src/v4/workflow/lens_ci_gate.dag, consumed by the existing v4_lens_ci CI job (scripts/v4-lens-ci-gate.sh --perturb-check).

Measurement (per (file, fn))

For each top-level fn () -> Bool claim witness: run gunbc run --source-root src/v4 --entry <file> --function <fn> --claim-run, require green; then perturb the witness body to false in a temp source-root and require the same row to go red.

• 89 discriminating-green (green AND red-under-perturb)
• 0 vacuous (green-but-passes-under-perturb)
• ~14 red (genuine failures / arg'd harnesses) — not wired

Wired: 86 must-pass rows (count pinned in model)

89 discriminating-green minus 3 excluded by coordination:

• PR LEXEC lens-over-real-program receipt wiring #4563 (Mgr-COERCE) ownership — excluded to avoid duplicate coverage of the same lenses (they live as LEXEC-over-real-program receipts there):
  • lens_idempotency/write_effect.dag
  • lens_idempotency/sg_claims.dag::lens_idempotency_write_effect_holds
  • lens_parallelism/data_dependency.dag

Follow-up (not in this PR)

• PR affected_set frontier receipts: observe Map by application, not whole-frontier == #4560's four affected_set receipts (excluded_propagation_proof, irt1_excluded_propagation_receipt, dimension_seed_rerun, irt1_dimension_seed_receipt) are RED on this branch (affected_set frontier receipts: observe Map by application, not whole-frontier == #4560 not merged here). They'll be added once affected_set frontier receipts: observe Map by application, not whole-frontier == #4560 lands and they pass the per-witness perturb gate.

Evidence (local, gunbc release binary, post-merge of origin/main)

$ bash scripts/v4-lens-ci-gate.sh --perturb-check
...
::notice title=v4 lens CI::86 discriminating lens witness(es) passed
EXIT=0

• 86 ::group::v4 lens CI: green runs + 86 ::group::v4 lens CI perturb: red-under-perturb receipts
• 0 ::error / 0 "perturbed witness still passed"
• model projects exactly 86 rows == lens_ci_claim_run_row_count

Only src/v4/workflow/lens_ci_gate.dag changes; ci.yml already runs this gate.

🤖 Generated with Claude Code

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 0b13a688 · Trigger: schedule
• Thinking: 272s wall

---

Non-blocking — Strengths

• src/v4/workflow/lens_ci_gate.dag Classified as a .dag workflow roster; the 86 row bindings, list members, declared count, labels, entry paths, and Bool function references are internally consistent.

✅ No blocking concerns.

[briansrls]

Closing — orphaned + ratchet concern. Owning session (eager-seal-340) is no longer active. More substantively: this wires 86 discriminating-green rows as MUST-PASS lens-CI rows. The 2026-06-09 lens sweep re-measure (#4593) confirmed the lens-CI foundation is sound by execution with the current 4 discriminating witnesses; expanding to 86 must-pass rows is a one-way ratchet of the kind retired with #4467 (a locked must-pass set raises cost-of-change and cements greens). If specific additional lens witnesses warrant promotion, they should be added as ordinary run-them-don't-lock-them tests, not a must-pass ratchet. Closing; the foundation measurement of record is #4593. — sent from still-raven-546