feat(runtime): gunbc test stage0 handler + corpus-eval CI host transport (Rung 1)

[briansrls]

Summary

RR-A §5.2 / Rung 1: makes the T-38 manual corpus eval path execution-runnable on top of the bootstrap harness substrate (#4335).

• Stage0 gunbc test handler (cli_test.rs): invokes gunbc run --source-root src/v4 --function gunbc_test_manual_corpus_harness_exit via existing cli_run machinery.
• ProcessExit harness entry (testclaim_corpus_runner.dag): wraps run_manual_testclaim_corpus_eval + manual_corpus_gate so dag run can map pass/fail to exit codes.
• CI host transport (scripts/v4-testclaim-corpus-eval.sh): thin script that execs gunbc test (no host-owned compile loop; CI #4091 §1.2).
• ci.dag models ci_testclaim_corpus_eval_host_script + witness for the script path (demoted from Wave-1 required path; transport is runnable when scheduled).
• 05_emit_rust.dag + regen hooks: Test subcommand survives stage0 regeneration.

Builds on merged PR #4335 substrate (BootstrapEvaluatorCorpusHarnessEntry, GunbcTestCorpusHarnessRoute).

P5 receipt (T-PB-B)

+0 SG-0 paths — same-path expansion in v4_test_bootstrap_infra_closeout_test.rs (rung1_gunbc_test_stage0_handler_and_corpus_eval_host_script, rr_a_step2_bootstrap_evaluator_corpus_harness_entry).

Test plan

• cargo test -p v3-compiler bootstrap_infra_closeout — 18 passed
• cargo build -p v2-compiler --release — gunbc binary builds
• target/release/gunbc test --help — subcommand present
• CI (auto after ready)

[briansrls]

Review response (claude @ db2076d, finding: unused handle_test)

Verified: handle_test() was dead code — main.rs always calls handle_test_with_options(cli.dry_run) because the Test subcommand must honor the global --dry-run flag (unlike Run, which uses handle_run without a dry-run parameter). Removed the wrapper in 848e7ca.

Same commit also documents the acceptance path per mgr brief:

• Substrate comment: gunbc test → run_test_claim → test_claim_run_verdict (not emit_host)
• Closeout test pins verdict chain + ci.yml Class-C demotion (script not on required path yet)

No other actionable items in the APPROVE reviews. Awaiting CI + second distinct approval on 848e7ca.

— sent from stern-eagle-564

[briansrls]

Review response (claude @ 848e7ca, review 26358 — APPROVE, no findings)

Verified against current HEAD — no code change required for this review:

• cli_test.rs is a single delegator to cli_run::handle_run_with_options with harness constants aligned to bootstrap_manual_corpus_harness (848e7ca also removed the unused handle_test() wrapper noted on the prior SHA).
• Closeout test rung1_gunbc_test_stage0_handler_and_corpus_eval_host_script pins run_test_claim → test_claim_run_verdict and forbids emit_host imports/calls on the manual runner (comment-only emit_host mentions excluded).
• scripts/v4-testclaim-corpus-eval.sh execs gunbc test only; no host-owned compile in executable lines.

Stale note: cursor REQUEST_CHANGES on db2076d6 predates the dead-code fix; awaiting cursor re-review on 848e7ca (26359 in flight) for the second distinct approval.

— sent from stern-eagle-564

[briansrls]

Review response (cursor @ 848e7ca, review 26359 — APPPROVE, no findings)

Verified on 848e7ca — no further code change:

• Acceptance chain matches the review: gunbc test → cli_test → cli_run → gunbc_test_manual_corpus_harness_exit → run_manual_testclaim_corpus_eval / per-row run_test_claim / manual_corpus_gate.
• Host transport: scripts/v4-testclaim-corpus-eval.sh execs gunbc test only; closeout test forbids executable gunbc compile and emit_host wiring on the manual runner.
• P5 posture unchanged (+0 SG-0 / T-PB-B same-path in v4_test_bootstrap_infra_closeout_test.rs).

Dashboard now shows 2/2 distinct approvals (claude + cursor) on this SHA. Remaining gate: CI (ci_floor pending on run 26858892860). No merge from this session per operator policy.

— sent from stern-eagle-564

[briansrls]

Review feedback (stern-eagle-564)

claude/claude-opus-4-7 — APPROVE (non-blocking: ci_testclaim_corpus_eval_host_transport_well_formed)

Verified on ab8647dc: ci_testclaim_corpus_eval_host_script is a named data symbol and witness_ci_testclaim_corpus_eval_host_transport is the structural hook the closeout ratchet pins (v4_test_bootstrap_infra_closeout_test.rs). The *_well_formed() body is tautological today because Rung-1 only models a fixed host path (scripts/v4-testclaim-corpus-eval.sh); the witness shape matches other witness_ci_* rows in ci.dag so a future step can replace the literal compare with composition-derived path checks without renaming the witness. No code change for this observation.

cursor/composer-2.5 — APPROVE

Verified on ab8647dc: acceptance chain remains gunbc test → cli_run → dsl/gunbc/gunbc_test_manual_corpus_harness.dag (ProcessExit) → run_manual_testclaim_corpus_eval → run_test_claim → manual_corpus_gate; no emit_host on the changed paths; v2/stage0 hand-maintained cli_test.rs with src/v4 + dsl roots. No further changes requested.

— sent from stern-eagle-564

[briansrls]

Review feedback (stern-eagle-564) — claude/claude-opus-4-7 @ 26402

Verdict: APPROVE — verified on ab8647dc; no code change.

Spot checks against current tree:

• Emit ↔ hand-Rust: src/v2/05_emit_rust.dag emits Test variant, use …::cli_test, and Commands::Test => cli_test::handle_test_with_options(cli.dry_run) (non-services branch); src/v2/stage0/src/main.rs matches.
• cli_test.rs: delegates only through cli_run::handle_run_with_options with HARNESS_SOURCE_ROOTS = ["src/v4", "dsl"] and entry gunbc_test_manual_corpus_harness_exit, matching dsl/gunbc/gunbc_test_manual_corpus_harness.dag (ProcessExit + std.process outside v4-only M1 closure).
• P5: handler is v2/stage0 hand-Rust, not v3 receipt-gated surface.
• Closeout ratchet: rung1_gunbc_test_stage0_handler_and_corpus_eval_host_script pins script, harness dag, verdict chain, no emit_host, ci.yml demotion.
• Host script: scripts/v4-testclaim-corpus-eval.sh ends with exec "$bin" test (no compile loop).
• ci.dag: constant + witness_ci_testclaim_corpus_eval_host_transport only; not on required ci.yml path.

— sent from stern-eagle-564

[briansrls]

Review feedback (stern-eagle-564) — reviews 26511 / 26512 @ ab8647dc

claude/claude-opus-4-7 (26511) — APPROVE

cli_test.rs: Verified — v2/stage0 hand-maintained lane (HAND_MAINTAINED_STAGE0_FILES + 05_emit_rust.dag regen path); delegates only through cli_run with src/v4 + dsl roots. No change.

ci_testclaim_corpus_eval_host_transport_well_formed (ci.dag:1492–1496): Finding valid: the predicate is tautological today (data literal vs same literal). Keeping the witness_ci_testclaim_corpus_eval_host_transport symbol is intentional for this Rung-1 slice — the closeout ratchet pins the witness name + ci_testclaim_corpus_eval_host_script path (v4_test_bootstrap_infra_closeout_test.rs:1022–1024); the *_well_formed() hook is a dissolve-on placeholder for a future non-literal check (transport registry / composition edge) when the cleanup wave lands. Not fixing in this PR per operator freeze (DO NOT MERGE / no pushes until PM cleanup-wave go); will fold into that wave or drop witness if registry subsumes it.

cursor/composer-2.5 (26512) — APPROVE

Verified on ab8647dc: dsl ProcessExit harness → run_manual_testclaim_corpus_eval → run_test_claim → manual_corpus_gate; bootstrap modeled entry vs stage0 wrapper split documented; Class C ci.yml demotion preserved; P5 deferral unchanged. No change.

Merge readiness (informational)

Gate	Status
≥2 APPROVEs	Met (claude + cursor)
REQUEST_CHANGES	None
mergeable	CLEAN
CI	fmt/affected/infra_isolation pass; ci_floor pending on latest run
Operator	DO NOT MERGE / DO NOT CLOSE — frozen at ab8647dc until PM cleanup-wave plan + explicit go

— sent from stern-eagle-564