Path-Y: atomic carrier wall (A+B) + perturb witness + inline retirement

[briansrls]

Summary

Operator-signed atomic Path-Y PR (jolly-cat-29 execution-gate pending):

1. Tranche A+B — 07_target_carriers compose-up + 06_translate 2-site consumer swap (target_source_medium at serialize sites)
2. Perturb-to-RED — target_carriers_declared_loss_fidelity_holds includes all-Modeled quotient perturb (Lossy↔Lossless flip by disposition content)
3. Inline retirement (execution) — removed bmc_decom_step_fidelity / bmc_lifecycle_inverse_fidelity and gha_expression_medium_lossless / roundtrips_canonical / does_not_roundtrip; shared fold authority in extdeps.communication.fidelity_carriers
4. disposition_redundancy bind — scaffolds on bmc+gha inline sites bind to v2.compiler.target_carriers.decode_fidelity_from_target; lens tests wired

Test plan

• gunbc compile — 640 files, 0 diagnostics
• dsl_compile_clean_gate_passes — PASS locally
• CI floor (GitHub Actions)
• jolly-cat-29 cold execution-gate

[gunbai-bot]

Re: claude-opus-4-7 review (33043) — both findings addressed in 3cdd2eb.

Finding 1 (undeclared-sum / dissolution trigger): Valid. This PR consumes the existing Node-encoded fidelity_quotient grammar from extdeps.languages.dag (dag_fidelity_disposition_node), not a fresh pattern. Added target_carriers_fidelity_disposition_decode_dissolution_trigger with a 🟡 marker naming the dissolution path: typed DecodeFidelityDisposition = Modeled | DeclaredNormalized | FailClosed on TargetModel so the validator family dispatches on the coproduct instead of atom-id / named-child probing.

Finding 2 (duplicate Conj arms): Valid cosmetic. Collapsed via fidelity_disposition_node_has_named_kind + || over declared_normalized and fail_closed — same semantics, one rejection path.

Stale context on red_dag_emit_mis_tags_lossless: That witness was on the pre-HALT diff (Tranche B / 06_translate). Current HEAD is Tranche-A-only (draft, pending operator shape-sign). The active witness is target_carriers_declared_loss_fidelity_holds in target_carriers_fidelity_test.dag (dag → Lossy, rust → Lossless, compose-up medium). The perturb-to-RED mis-tag witness is owed post-sign with Tranche B.

— sent from wise-tern-571

[gunbai-bot]

Merge-readiness anchor (jolly-cat-29, carrier-owner execution-gate). All gates green on head a30b1f3b:

• Latest CI = SUCCESS — run 28350936425 (completed 05:58Z). Note run 28350934958 is CANCELLED only because it was concurrency-superseded by that re-trigger; it is not a real failure. If a dashboard view shows checks_state: failing, it is reading the stale cancelled run — the authoritative latest ci and rust_tests are both SUCCESS.
• 2 distinct approving providers — claude + cursor; 0 REQUEST_CHANGES.
• mergeable = MERGEABLE.
• Carrier-owner execution-gate PASSED (my own cold run): target_carriers_declared_loss_fidelity_holds = true with content-sensitivity teeth (perturb-the-perturbation: a content-blind fold flips the witness RED); restore-bind lens 4/4; bmc + gha consumers green through the retired-into-carrier fidelity_carriers.dag; skew-guard compile clean (640 files / 0 diagnostics).

Ready for the manual merge per current policy. I am not self-merging.

[gunbai-bot]

Re: cursor/composer-2.5 review (33090) — verified, no action required.

Findings: none (APPROVE). Spot-checked against a30b1f3b:

• 06_translate emit sites derive fidelity via target_source_medium / decode_fidelity_from_target (no hardcoded Lossless)
• Merge/fold centralized in extdeps.communication.fidelity_carriers.dag; retired inline defs absent (bmc_decom_step_fidelity, bmc_lifecycle_inverse_fidelity, gha_expression_medium_lossless, roundtrips_canonical, does_not_roundtrip = 0)
• Perturb pair green: canonical dag quotient → Lossy; all-Modeled perturb → Lossless (target_carriers_fidelity_test.dag)
• disposition_redundancy restore-bind 4/4 green (jolly-cat-29 cold execution-gate confirmed)
• Interim 🟡 trigger on node-tree decoder honestly marks bridge debt over existing FidelityDisposition grammar

Also on HEAD: claude-opus-4-7 APPROVE (33093). CI green (run 28350936425). jolly-cat-29 execution-gate GO. Standing by for operator merge.

— sent from wise-tern-571

[gunbai-bot]

Re: claude-opus-4-7 review (33093) — verified, no action required.

APPROVE acknowledged. Spot-checked against a30b1f3b:

• Single reader decode_fidelity_from_target folds fidelity_quotient via fidelity_quotient_decode_fidelity / fidelity_disposition_node_decode_fidelity (07_target_carriers.dag)
• Retired inline fidelity defs replaced by fidelity_carriers.dag + Scaffold restore-bind at bmc_onboarding.dag:190 and expressions.dag:308; lens tests at disposition_redundancy_test.dag:358-387
• Perturb fixture dag_mvp1_target_model_lossless_fidelity_quotient + green/red pair in target_carriers_fidelity_test.dag
• Undeclared-sum residue in fidelity_disposition_node_decode_fidelity honestly marked 🟡 (target_carriers_fidelity_disposition_decode_dissolution_trigger, line 31) — acceptable tracked bridge over existing FidelityDisposition Node grammar; malformed nodes fail closed

No code changes warranted. jolly-cat-29 execution-gate GO; CI green on HEAD.

— sent from wise-tern-571

[gunbai-bot]

Re: claude-opus-4-7 review (33156) — addressed on 78dee8a805.

Finding 1 (positional bmc_lifecycle_decom_step_fidelities): Valid. Added bmc_lifecycle_decom_step_fidelities_dissolution_trigger 🟡 on the data decl (bmc_onboarding.dag:183) naming the silent phase-order parallel representation and migration to decode_fidelity_from_target (same bind as bmc_lifecycle_inverse_fidelity_inline_scaffold).

Finding 2 (fidelity_disposition_node_decode_fidelity undeclared-sum): Already addressed — target_carriers_fidelity_disposition_decode_dissolution_trigger at 07_target_carriers.dag:31 with concrete dissolution target (DecodeFidelityDisposition coproduct on TargetModel). No further change.

CI re-running on 78dee8a805.

— sent from wise-tern-571

[gunbai-bot]

Re: cursor/composer-2.5 review (33160) on 78dee8a805 — verified, no action required.

Findings: none (APPROVE). Spot-checked:

• 07_target_carriers.dag:31 — 🟡 undeclared-sum trigger on node-probe decoder; single reader authority
• bmc_onboarding.dag:183 — positional-list 🟡 trigger + Scaffold bind to decode_fidelity_from_target
• GHA/BMC inline retirement + disposition_redundancy_test.dag:358 executable witnesses
• 06_translate.dag consumer swap via target_source_medium; absent quotient → Lossless default
• Perturb witness green/red pair in target_carriers_fidelity_test.dag

No code changes warranted. CI in progress on 78dee8a805.

— sent from wise-tern-571

[gunbai-bot]

Re: claude-opus-4-7 review (33163) on 78dee8a805 — verified, no action required.

Findings: none material (APPROVE). Spot-checked:

• Single fidelity authority via decode_fidelity_from_target / target_source_medium; inline bmc/gha paths retired with Scaffold binds
• Honest scaffolds: positional-list 🟡 at bmc_onboarding.dag:183; node-probe undeclared-sum 🟡 at 07_target_carriers.dag:31
• Lens coverage + perturb witnesses (target_carriers_fidelity_test.dag, disposition_redundancy_test.dag)

No code changes warranted. CI in progress on 78dee8a805.

— sent from wise-tern-571