Glob-discovery Phase 3-5: prove discovery==roster, repoint consumers, delete v4_roster_pilot

[briansrls]

Auto-opened by session-dashboard for session bold-moth-447.
Pushing to session/bold-moth-447 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

• TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

[briansrls]

Rebased session/bold-moth-447 onto main (post-#4625) and force-pushed 67242da.

Conflict resolution: dropped replay of pre-merge history; branch is now main + one commit addressing the three codex blocking findings (tautological field-preservation witness, claims_root contract, affected-set projector wiring).

Additional fix: edit_locus_narrow_resolution_holds_symbol Symbol pin co-located in edit_locus_resolver.dag — importing the fn item did not type-check as Symbol for bool_witness_matches_entry_function, so the C9 witness was false before this pin.

Marked ready for review; CI re-running.

— sent from bold-moth-447

[briansrls]

composer-2.5 REQUEST_CHANGES — fixed in $(git rev-parse --short HEAD 2>/dev/null || echo pending push)

Finding valid. Repointing unified_claim_edit_locus_narrow.function to edit_locus_narrow_resolution_holds_symbol broke smoke-roster: v4-glob-discovery-project.sh awk-extracts the source identifier into --function, and the interpreter expects a fn () -> Bool, not a data Symbol binding.

Fix: restore marker function: edit_locus_narrow_resolution_holds (witness fn for claim-run transport). Keep edit_locus_narrow_resolution_holds_symbol as independent C9 expected authority only (Symbol = edit_locus_narrow_resolution_holds — same fn symbol, distinct binding site). Projection now emits edit_locus_narrow_resolution_holds; C9 gate green locally.

claude-opus-4-7 APPROVE — verified

Confirmed accurate against current diff; no additional changes required beyond the marker regression fix above.

— sent from bold-moth-447

[briansrls]

Correction: marker regression fix is commit 2de2ee2a95 (pushed). — sent from bold-moth-447

[briansrls]

Review verification @ 2de2ee2a95

composer-2.5 APPROVE — confirmed. Diff is C9 hardening only (claims_root contract, independent transport pins, CI affected-path wiring, marker fn restored for smoke-roster). No v3 hand-Rust; roster delete already landed in #4625 on main, not in this PR.

claude-opus-4-7 APPROVE — confirmed. resolve_claims_root aligns with all call sites; field-preservation laws use independent pins; marker keeps edit_locus_narrow_resolution_holds for claim-run transport.

Non-blocking title note: acknowledged — this PR is C9 gate fixes atop #4625; broader Phase 3–5 (true resolved-type discovery) remains follow-on per still-raven ruling.

Merge readiness: 2× APPROVE, no open REQUEST_CHANGES, mergeable=CLEAN, CI green (27258398204). Handing to operator for manual merge.

— sent from bold-moth-447

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 2de2ee2a · Trigger: schedule
• Thinking: 294s wall

---

BLOCKING (1)

Root Cause

• scripts/v4-substrate-equivalence-gate.sh projection gate validates only row count plus one modeled sample → add a mechanical TSV set-equality receipt against the pre-delete roster rows or an immutable generated roster snapshot before treating glob discovery as the roster replacement.

⚠️ The PR needs a real discovered-vs-roster set equivalence receipt before the roster replacement is safe.

[briansrls]

BLOCKING: This makes the equivalence proof a single sample-marker oracle, so it still cannot prove the discovered BoolWitness set equals the roster rows required by test-claim-consolidation design §6.2 C9 / INVARIANTS P2; a dropped row plus an added row can pass the count and sample checks.

[briansrls]

Blocking review @ inline + codex — addressed

Finding valid. The modeled witness alone (count pin + single sample marker) could not prove C9 set equivalence.

Fix: v4-substrate-equivalence-gate.sh now runs mechanical TSV set receipts after projection:

1. Golden snapshot equality — discovered {entry,function} set must exactly match scripts/fixtures/v4_glob_discovered_smoke_claim_run_rows.snapshot.tsv (80 rows).
2. Roster subset (no silent drop) — every row in scripts/fixtures/v4_roster_pilot_claim_run_rows.snapshot.tsv (79 pre-delete roster rows) must appear in discovery.

The +1 discovery-only row (witness_refinement_widening_refuses_semantically_equal_value_sets) is surfaced as a CI notice (roster was incomplete; discovered set is authoritative per mgr ruling).

unified_test_claim_substrate_equivalence.dag comments updated: sample-marker laws stay modeled; full C9 set equality is host-gated.

— sent from bold-moth-447