RO-LIVE-05

PHASE4-N-M-C (2026-05-26) CLOSED RO-LIVE-05 on the bounded statement: (1) live wire integration end-to-end against docker preprod is proven (committed wire-only transcript at slot 76910188 with handshake_ok v15 + peer_tip_read); (2) the closed LiveConsensusInputs import authority + canonical Blake2b-256 fingerprint shipped (CN-CONS-IN-01, DC-CONS-IN-02), exercised against a live preprod bundle (epoch 179, 384 pools); (3) the GREEN LiveLedgerView + epoch-window guard (DC-VIEW-01, DC-ADMIT-11) + closed AdmissionPeerEvent emission contract from the RED wire pump (CN-PUMP-01, DC-PUMP-01, DC-PUMP-02) + Undecodable-tightening (DC-ADMIT-12) shipped; (4) adversarial false-accept corpus across 4 mandatory mutation classes (DC-EVIDENCE-02) proven fail-closed.

PHASE4-N-M-A1.1 (2026-05-26) CLOSED the original A1.1 reference-script gate AND the freshly-surfaced A1.2 Byron-address + Plutus-integer-field gates. The seed importer now handles the FULL preprod UTxO dump: 1,909,985 entries import cleanly at the bundle's tip (committed transcript phase4-n-m-a1.1-admission-bootstrap-transcript.jsonl records admission_started + bootstrap_complete with the canonical consensus_inputs_fingerprint_hex and initial_ledger_fp_hex). Two-imports byte-identity preserved across the full dump.

2026-05-27 follow-up: with the peer at 100% sync, the operator pass was re-run against the live tip (epoch 291, source_tip_slot 124134280). Bootstrap completes cleanly (committed bootstrap transcript) but the admission wire pump exits with BlockFetchDecode immediately after chain-sync intersect succeeds and block-fetch RequestRange is sent. Root cause (diagnosed 2026-05-27): the session reducer (ade_network::session::core::drain_connected_frames) emits ONE DeliverPeerFrame per mux frame, but cardano-node fragments large block-fetch responses across multiple mux frames (mux payload max = 65535 bytes; Conway preprod blocks exceed this). Per-mini-protocol payload reassembly across consecutive mux frames is MISSING in the session reducer. The capture binary crates/ade_network/src/bin/capture_block_fetch.rs::read_for_protocol already implements the correct reassembly pattern (per-protocol byte buffer + try_decode predicate); the session reducer needs the same logic ported into its closed-sum drain path.

This is a session-reducer slice (PHASE4-N-L or PHASE4-N-M-FRAG scope), NOT an A1.1 / C5 strengthening. A1.1 stands closed; the open obligation re-tags to reflect the newly-surfaced gate.

See docs/evidence/phase4-n-m-c-operator-pass-README.md §8 for the diagnosis runbook + the captured BlockFetchDecode evidence; [[feedback-shell-must-not-overstate-semantic-truth]] doctrine load.

PHASE4-N-M-FRAG (2026-05-27) CLOSED at HEAD 4d3dc98 shipped the session-reducer reassembly + tag-24 unwrap + chain_dep epoch_nonce wiring. Live pass against fully-synced peer then surfaced Header(VrfCert(VerificationFailed)) at the pool_active_stake_missing stage with epoch=0 — pointing at the era-schedule's epoch-number plumbing, NOT VRF / nonce / pool data.

PHASE4-N-M-SCHED (2026-05-27) CLOSED RO-LIVE-05 outright. Root cause: ade_node::admission::bootstrap::make_schedule_for_imported_window hardcoded start_epoch: EpochNo(0) instead of canonical.epoch_no. With the schedule treating all imported-window slots as epoch 0, LiveLedgerView (which gates lookups by epoch == self.inputs.epoch_no) refused every per-pool lookup → header_validate short-circuited at the pool_active_stake.ok_or(VerificationFailed) line. Fix: thread canonical.epoch_no into make_schedule_for_imported_window.

Committed live transcript at docs/evidence/phase4-n-m-c-operator-pass-transcript.jsonl records the full DC-EVIDENCE-01 statement:

• 1 admission_started with consensus_inputs_fingerprint_hex=8166ba41…
• 1 bootstrap_complete with initial_ledger_fp_hex=65461b64… and chain_tip_slot=124136968
• 1 block_received at slot 124140368, peer hash d111e613…
• 1 block_admitted at slot 124140368, our hash d111e613… matches peer hash, post_fp 9528023e…
• 1 agreement_verdict { kind: "agreed" } at slot 124140368, our_hash == peer_hash == d111e613…
• 0 agreement_verdict { kind: "diverged" }
• 0 block_admitted with mismatched hash
• clean admission_shutdown { reason: "signal_received" }

PHASE4-N-M-FOLLOW (2026-05-27) strengthens by demonstrating sustained admission across the natural chain-sync stream. Committed transcript at docs/evidence/phase4-n-m-follow-sustained-transcript.jsonl records 34 consecutive block_admitted events from real chain-sync follow + block-fetch (slots 124137045..124137868), 0 diverged, 0 mismatched-hash, 0 input_not_found. The pump now walks the peer's chain in order from intersect: each RollForward extracts a Point from the header envelope and triggers a block-fetch; block-fetch BatchDone gates the next chain-sync RequestNext. The op-cert counter rule was fixed inline — equal counter is now no-op (same op-cert re-used across blocks within a KES period, normal pool operation per the Cardano protocol). Previous strict-<= semantics over-rejected legitimate repeated op-certs and was masked by the corpus's one-block-per-pool design.