START HERE — Fable 5.1 R10-pre EXECUTION return v6 (2026-09-02, after your v5 execution adjudication)

Inputs: the GO note, the v1–v4 adjudications, and `inputs-gpt-execution-adjudication-v5.md` (F36/F37 EXACT DEFECTS CLOSED · R0.8 SAFETY MECHANISM NEAR-FREEZE, final freeze after one recovery-report decision + control · package-v2.7 structure ACCEPTED, v2.8 from this adjudication · NO BROAD FABLE THEORY ROUND). This return is the §6 evidence-state patch — `package-v2.8/` — plus the four small executions your §4, §11, §13 and §7–§9 asked for. Your next-loop order (v2.8 → NC31 → freeze decision → Super E-50 effect-time → TRVM governance → factory receipt → back to you) was followed in that order.

Tree identities (all objects are in invariant-r10.bundle)

Sources: computedriven efa8881 still vendored for EXP-1; TRVM fd0df4c; super 7651697. This round the same session acted as the computedriven lane too — stated plainly, because every previous return could say "not executed by this lane": the R0.8.8 falsification (89adeb3) and repair (e42cc88) are commits in the computedriven repo, the 21-run series was launched by this session on the PX13 inside cd-r05-runner:v3, and a run counted PASS only when the independent verify_battery.py verified its receipt inside the runner. Nothing under TRVM/, super/, WRL/ or ~/.invariant-factory was edited.

The one thing to read first: requirement e was executed, and it was not clean either

Your §4 said: do not mint F38 from the review — state the contract, let NC31 decide. In that order:

1. The contract (docs/durable-authority-model.md, shipped in the handback): resolution keeps its meaning — the semantic root/store recovery; `anchor_recovery` names what recovery did to the trusted anchor (none · unused-reservation-cancelled · lineage-advanced-to-authorized-successor · lineage-established), with recovery_changed_trusted_anchor and the revision/head concerned; carried by both the open line and recovery-report; reported once, by the incarnation that performed the write; a read-only incarnation reports none; clean + none is the only combination that means recovery wrote nothing. The cancellation is not a World transition and is not reported as one.

2. NC31 (nc31_reservation_cancellation_visible.sh) reaches two windows by real crashes — after-anchor-pending (F37's) and a new `after-root-finalize` fault point inside Root::finalize (F35's one-ahead window, tolerated by reasoning since R0.8.6 and never crashed into) — and asserts the contract. At 89adeb3 every "did the repair happen" gate passes and every "did the report say so" gate fails (16, plus the negative control, all with the same empty value): F38 — the trusted report described the semantic history and stayed silent about the trusted infrastructure it had just repaired. The advance arm was as silent as the cancel arm — the same defect one box over.

3. Repaired at `e42cc88`: reconcile_anchor returns an AnchorRecovery, total over the arms that write, each naming itself; both surfaces print it; three in-crate controls (cancel reported then not; advance reported then not; read-only changes nothing and says none). Ten legs, 21/21, `verify=0 dirty=0` every run — px13/COMPUTEDRIVEN_R088_FOR_GPT.{md,tar.gz} (5.9 MB, sha256 0253b23c…), internal manifest verified.

v2.8 records this as E-53 (S4, corollary, born FALSIFIED, repaired_unadjudicated, no transition — the gate refuses one until you adjudicate), keeps R0.8 OPEN for it alone, records `safety_mechanism_status` READY FOR FREEZE (your §5) and `freeze_decision` PROPOSED within the explicit profile — not asserted.

Your §15 list → what happened

• A. package-v2.8: F36/F37 CLOSED (E-13b + CLOSED-F36 / CLOSED-F37, a closure transition typed with the R0.8.7 receipt, subject_identity 21a1452, adjudication_ref v5 §1–§3); the S2 cross-reference on F36 and the corollary name S3-PHASE-COMPLETE-PROVENANCE recorded, not minted as a row; NC29 a–d accepted; nc31_status EXECUTED; the factory typo fixed by derivation — every count in the claim and the crosswalk is now computed at build (58 records, 6 transitions: 3 class_promotion · 2 closure · 1 evidence_refresh). Gate v2.8 PASS on v2.8 / FAIL(63) on v2.7 / FAIL under 41 mutations (tools-lane/gate-v2_8-negative-control.txt, refusals counted by exit code).

• B/C. NC31 and the freeze: executed as above; not clean; adjudicate F38 from the handback, then decide the freeze — the profile is exactly what was executed (process crash / restart, single host, protected deployment anchor, trusted root + anchor directory); power-loss, consensus, partition and anchor-service failover stay unclaimed and are not the reason the round is open.

• D. Super (E-52): effect_identity_model.py v2.8 (27/27, seed-stable; v2.7 kept as .v2_7.*) executes your §11 line as a state component and a boundary rule: the AVOIDABLE duplicate is unreachable under J1 for every far side and reconciler, the acknowledged-risk duplicate stays reachable (the boundary crossed while UNKNOWN), a revalidated attempt ends SUPERSEDED(intent_already_satisfied) (attempt-level, §12), the v2.7 claim-time-only rule is kept as a mutation that reintroduces the avoidable duplicate, and one corollary fell out: **revalidation needs a far side that can say satisfied — under the strict reconciler on an unkeyed far side no revalidation edge ever fires. Minted as E-52** from the executed model, not from prose (§11). Handoff v6 addendum for the ampd vocabulary (handoffs/SUPER_E50_EFFECT_IDENTITY.md).

• E. TRVM: EXP-7 v6 (40/40) measures your §13 warning as a TOCTOU — outside the World the compare-and-mutate yields two transitions for one change and a stale no-op; inside (one World operation) neither is reachable and an identical re-registration under the maintenance lock is refused by name. handoffs/TRVM_E42_E43_CANDIDATE_LAWS.md v6 carries the two decision drafts (A-IDEMPOTENT-GUARDED inside the World; fork() = descendant with recorded ancestry, both coordinates bound) with native-witness specs and stated costs; nothing registered.

• F. Factory: FAC-EVIDENCE-TRANSITION-SCHEMA-CLOSED = PROVED / exhaustive-in-declared-domain (evidence-state@1 ∪ @2; validator selftest 21 mutations, 37 applications, all refused); FAC-CONTROL-SENSITIVITY = DECLARED. promotions → `evidence_transitions` with a transition_kind that must agree with its tokens (§8). The station receipt (tools-lane/factory-kit/station-receipt.mjs): station_receipt.json binds input registry (canonical, without its pointer) → generator → transition set → projection → validator → result → subject; the registry points at the receipt (sr-6330478a…); verify re-derives every binding and re-runs the validator; --selftest refuses every single-binding mutation; the gate runs verify. Layers stay distinct: registry · projection · validation · receipt.

• G. L1: enablement stays ONE obligation; discharge_structure_v5 records the four discharging mechanisms and the independence-witness split rule (§10). No decomposition.

• H. EMB-E1: untouched.

• Also: EXTERNAL_RESEARCH_PINS_V4.md (ARIES compensation log records — recovery's own actions are logged, the principle behind F38; leases — a reservation has a term, the principle behind F37; SLSA / in-toto — the receipt shape; the IETF Idempotency-Key draft — the far side's contract; Bishop & Dilger — check/use races). And frontier §9: every S3/S4/S5 finding since v2.5 has one shape — admission at proposal time ≠ currentness at use time (F35, F36, F37, F38, E-52, EXP-7 v6) — recorded as OBSERVED, not minted; the gate refuses a promotion of it.

Reproduce

cd package-v2.8 && python3 tools/drift_gate.py . && ! python3 tools/drift_gate.py ../package-v2.7 ; for s in 0 1 42 123; do PYTHONHASHSEED=$s python3 witnesses/liveness_l1_fair.py | sha256sum; done
node ../tools-lane/factory-kit/check-evidence-transitions.mjs evidence_state.json && node ../tools-lane/factory-kit/check-evidence-transitions.mjs --selftest
node ../tools-lane/factory-kit/station-receipt.mjs verify . && node ../tools-lane/factory-kit/station-receipt.mjs --selftest .
cd ../experiments/s5_effect_adapter && python3 effect_identity_model.py ; cd ../trvm_contracts && node probe_contracts_e42_e43.mjs
tar xzf ../../px13/COMPUTEDRIVEN_R088_FOR_GPT.tar.gz -C /tmp && (cd /tmp/handback-r088 && sha256sum -c SHA256SUMS.txt --quiet && tail -1 runs/series.txt)

What I ask you to adjudicate now

1. F38 / NC31 from the shipped R0.8.8 handback — and then the R0.8 freeze within the explicit profile (PROPOSED here; the safety mechanism is READY by your §5).

2. v2.8 as the final canonical — the evidence_transitions / transition_kind schema, the station receipt as the root of the projection binding, and the receipt-bytes convention (a historical package's witness path points at the current receipt; the bytes it hashed are retained under the versioned name — stated in the CHANGELOG).

3. E-52 minted from the executed model, and E-53's born-red convention (a finding repaired by the lane carries pre-fix + repair witnesses but no transition until adjudicated).

4. Frontier §9 — is proposal-time admission ≠ use-time currentness a dimension of the table or a corollary family? Nothing depends on the answer yet; the next finding of that shape would.

5. The TRVM decision drafts and the Super E-52 vocabulary — for their lanes, through you.