destiny-2-Sunrise/MISSION_EMBER_REACTOR.md

Follow-up: the latest live run reached the escape endpoint but reported missing beam effects/hazards, a stale mission banner, inaccessible Controllers, and a freeze before the ending movie. See the newest entry in MISSION_EMBER_STATUS.md for the fixes and their remaining gameplay checks. The earlier candidate below was not visually validated for those behaviors.

1AU reactor sequence — 2026-09-06

Reference: 1AU playthrough, 21:20–28:30. Reviewed the cached video across that interval at five-second spacing, then at one-second spacing around 23:28–24:27, 25:32–26:39 and 27:15–28:04. Subtitles informed dialogue placement; this was visual inspection, not an audible validation of the music mix. Contact sheets live in build/reactor-audit/.

Reference sequence

Time	Observed behavior	Implementation
21:25–22:55	Interceptor/security fight, then access to Light's End	Two authored single-Psion squads DISPENSER_SUPPORT_A/B form the Electron Controller gate; boarding and support deaths do not open the security door.
23:00–23:10	Ghost reacts to the weapon, then identifies thermal exchangers	Cue 40 at entry, cue 41 nine seconds later; duplicate triggers do not interrupt/replay either.
23:38–23:44	Beam/screen surge and transition, then opening clamshell shutters	Native placed laser/ring objects are active. Their device transitions and alarm sequences precede the exposure window.
23:44–23:54	Purple target exposed for roughly ten seconds	Six-second warning, ten-second exposure, fourteen-second recovery. This thirty-second clock is reconstructed pacing, not a recovered original script.
23:51	First target destroyed; “That's one!”	Valid health/death receipt sets one durable completion bit and activates its authored destruction scene.
~24:55–25:00	Second target destroyed; access across bridges to central island	Both deaths extend both bridges, unlock/power central doors, instantiate the central target and activate coffin squads.
26:00–26:04, 26:25–26:30	Central target visible, shutters close, reopen, then target destroyed	Central doors and shield inherit the same clock, including an exposure already in progress. They continue cycling until central target death.
26:30	Ghost says temperature increase is insufficient	Cue 48; leave central housing open, unlock the fusion-cell route.
27:28–27:38	Ghost identifies fusion cell; player picks it up	Cue 50 via approach or first ownership receipt. Native MOTHER_BRAIN_CARRY_OBJECT; recovery handles expiration.
27:45–28:00	Cell carried to reservoir and inserted	Carry navpoint switches to EMBER_DIRECTIVE_REACTOR_MOTHER_BRAIN_DELIVERY_NAV_POINT; accepted deposit consumes the cell and begins escape once.
28:01–28:30	Overload/explosions, escape objective, Ghost calls Zavala	Existing authored explosion scenes/escape devices, cues 51–54, escape checkpoint and two ending cinematics retained.

Mechanisms fixed

80809562 damage Sense fields health and shield have reflected flags 0x21, meaning mandatory raw float fields. Optional presence is bit 0x02. The previous decoder consumed nonexistent presence bits, misaligning both health and the Auth revision. Native A774B0 publishes the fixed twelve-byte health/shield/revision state. Independent complete packet fixtures now cover healthy, zero-health and absent-target (-1) states, plus truncation. Lua accepts only the current encounter generation and a previously observed living target. Unloading an object, initial zero, old generations and duplicate deaths cannot advance it. Redundant lethal-effect requests after an already acknowledged target death were removed.

Sequences previously searched only the opening seed state. Scene lease validation likewise rejected a later local state even with a current connection/region. The resolver now prefers the exact live-region occurrence; the existing seed occurrence remains a fallback for shared mission objects. It retains scenario, object, publication, connection generation and ambiguity checks. Scene validation also accepts the exact live state. This addresses the recorded target_unavailable alarms and the later-area scene wrong_state failures without changing the selected mission state or republishing the opening area.

Device position, power and lock are independent lanes. The central doors are initially closed/locked, then explicitly unlocked/powered at the second clamshell death. The reactor bridges use position 1 initially and position 0 to extend, consistent with the reported inverse pose and the landing bridge. This final-area polarity still needs visual confirmation. Beam VFX use the authored SPECOPS_APEX_RING_LASER_DEVICE/RING_DEVICE and their placed models (80B7117C, 80B711C6, 80B71218); no fabricated overlay or model replacement. The exact beam/screen response and animation latency require an in-game check.

Native music sections now advance through access, controller defeat, weapon reveal, thermal-exchanger combat, first destruction, central access/exposure, cell and escape (19–28). Section identities beyond the recovered tumbler_exit are inferred from bank order; finale section names and audible transitions have not been independently verified.

Validation and installation

• Release DLL builds; 21 portable native tests pass, including mandatory damage packets and live-region behavior selection (missing area, cinematic state, foreign scenario, ambiguity).
• Five Lua suites pass. Full route exercises both-controller gating, warning before exposure, both-clamshell bridge gate, central object creation and repeating doors, stale/missing health, wipe generation, cell expiration/recovery, consumption, duplicate events and ending cinematics.
• Peak full-route usage: 228/512 variables, 61/63 recorded intents per callback, 2/32 timers, 13,000/100,000 instructions per mocked callback. Reset publishes cleanup before repopulating.
• Core initialization is deferred while outside region zero and idempotent on duplicate timers.
• Installation manifest: build/reactor-finale-installation.json. Existing save/settings/SDK and prior replication cleanup remain intact. Game launch is manual.

This is an implemented and regression-tested candidate. Native sound, screen surge, actual bridge pose and shutter timing remain live-test observations, not established results.