Aranna title animator: owning CPU history, 2026-11-07

The fresh unchanged original reached the title doors but not the ordinary menu or world. No emulator repair is justified by this run. The narrower missing observation is the resolved OnUpdate/OnStartChore reflective dispatch for the selected clip resource, and its call into UpdateBlender/AnimationCallback, with admission on every CPU instance that can own those calls.

Evidence: scratch/runs/20261107-aranna-animator-owning-history/ (ignored). This follows aranna-animation-completion-20261107.md; its native/current/retained 126 controls were not repeated. Animator +0x7e remains consumed pose dirtiness, not an enable defect. No guest instructions, guest state, assets or configuration were patched or substituted. No completion was synthesized.

Runtime and observer provenance

The passive WAT hook uses emulator-private globals and retained nonfaulting reads. It records entry/return with matching ESP, actual float argument bits, decision boundaries, animator/manager/current-node words, binding installation and reserved completion lifecycle events. It does not evaluate guest flags or write guest memory. Detail limits are 128 entry/return pairs, 1,024 decision boundaries and 64 separately reserved lifecycle entries. Synthetic RET-stub checks passed before box creation at 05:48:12Z; the strengthened lifecycle/state-receiver check passed before the sole original launch at 05:51:45Z. Baseline and observed stub executions had identical guest bytes, stack and PC. These checks validate the observer, not original-program correctness. An intermediate local WAT syntax failure is retained in compile-observer-final.log; final compilation and checks passed.

Owning CPU results

analysis.json, animator-final-events.json and raw original/observations.jsonl retain the measurements. Addresses below identify read-only guest test evidence. The selected animator was 166835476, manager 147347508, frontend 150089148; these were resolved from the live title entry, not copied from an older run.

Observation Measured result
Title routine 0x4bf1bb First admitted at 05:54:55.961Z on slot 0
0x69b615 initial entry Actual dt = 0, caller return 0x4be0a5
Pending processing return 0x69b62e EAX = 1; pending +e4 changes 1 to 0
Readiness boundary 0x69b659 All 171 admitted decisions have EAX = 1
Subsequent animator updates 127 positive float arguments, 0.0380010009–0.0670000017 seconds; caller return 0x5b9adb
Entry/return window 05:54:55.961Z–05:55:02.346Z; 128 matching returns
Pose update return 0x69b6ac 170 samples have AL = 0 (full EAX = 147347456)
Detail window Ends at 05:55:04.554Z when the explicit 1,024-boundary cap is reached
Selected clip/request +c4 = +d4 = 13; request +dc = 1, +e0 = 0; +d0 = 0 throughout snapshots
Animator status +78 = 0, +7c dword = 257, +f0 = 0 throughout snapshots
Manager phase/status +34 = 0, +38 = 0, +10 = 0, +48 = 0 throughout snapshots
Manager current/end +3c = +44 = 152079244; the pointed node is self-linked

The initial zero-time update returns before binding installation. At 05:54:55.964Z, 0x69b012 receives, then stores at animator +f4, exactly [0x4c0236, 0, 150089148, 0x4f3cab]. Incoming and stored binding words match. Thus this run distinguishes successful pending/readiness handling and positive animator dt from phase-manager advancement. It does not support blaming zero dt, an unready clip, or the consumed pose-dirty byte.

No reserved 0x69b048 dispatcher, 0x4f3cab adapter, 0x4c0236 member, or 0x521f84 state-request event was admitted. No selected-manager 0x6a6a11, 0x6a6a2e, 0x6a4bd7 or 0x69dde1 event was admitted in the detail window. These are scoped absences, not proof of global non-execution: selected-object admission lives in each WASM instance's private globals, and only slot 0 saw the title entry that populated them. Calls on another instance would not qualify. The animator detail cap also prevents claims about later phase state. A later title-entry packet at 05:58:37.354Z reports 4,546 selected updates, establishing continued entry activity, without extending the early argument/state window. The terminal broker reports zero drops and no observation error; caps and cross-instance admission, rather than broker overflow, are the stated limits.

Static discriminator and concrete next observation

static-boundaries.json identifies 0x69b048 as exported Aspect::AnimationCallback(unsigned long) and 0x69dde1 as Aspect::UpdateBlender(float). A linear rel32/reference scan and DLL import scan found no static natural caller of AnimationCallback. That does not exclude indirect or script/reflection calls; the natural completion caller remains unresolved.

Bounded disassembly adds a specific next boundary:

The next useful observation is therefore the selected resource's incoming OnStartChore/OnUpdate arguments, cached descriptor, resolved dispatch target and return at 0x45f650, plus manager phase construction/advancement and natural completion entry on all CPU owners. Seed identity from shared resolved objects rather than requiring every instance to execute 0x4bf1bb; reserve separate small budgets for this reflective path and completion. This run did not capture those resolved descriptors/targets, so neither a skipped script dispatch nor a CPU/Win32 semantic mismatch is established. Independent native/CPU/API semantics and an old-implementation-failing regression are still required before an emulator repair can be claimed.

Frames, inputs, validation and lifecycle

The cheap rendered full-shell smoke was personally inspected before creation: one balanced Start-button click opened the Start menu (smoke-before.png, smoke-after.png, smoke-review.json). It ran no original guest. The single fresh original attempt then recorded these individually reviewed frames, paired with .state.json, .review.json, input tickets and receipts:

Frame UTC / visible state Following action
startup 05:52:26, EULA One balanced click at fractions .2481,.7093
response-00 05:53:06, black loading Passive observation
response-01 05:53:27, black loading Passive observation
response-02 05:54:06, black loading Passive observation
response-03 05:54:52, logo on black Passive observation
response-04 05:55:39, logo over closed title doors One balanced Escape
response-05 05:57:41, title doors unchanged Passive observation
response-06 05:58:21, title doors unchanged Normal quit

There was no ordinary menu or world to act on. No EULA/clock control was rerun; no clock sampling supports a timing claim. Observer compilation, byte-identical retained provenance, unchanged import/export contract, synthetic passivity and lifecycle checks, shell smoke, input receipts, fetched evidence hashes and normal cleanup checks passed. No production code changed, so no emulator regression or locked production build is claimed for this documentation-only result.

One box was created with a UUID Idempotency-Key, bx_d7r57pgd, at 05:49:18Z; original expiry was 06:19:18Z. No resume, risk/access response, nested worker or second heavy runtime occurred. Evidence was retrieved before the explicitly delegated normal box stop. Guest/browser/supervisor exited code 0 at 05:58:42Z; all 11 recorded process IDs, runtime/CDP ports 34273/43317, and browser profile /tmp/wine-assembly-input-Lx2gJi were absent, with all four runtime locks released (cleanup-verified.json). Normal stop requested 06:00:34Z; API inspection confirmed archived, stop completed, non-forced at 06:00:35Z. Evidence manifests and lifecycle responses remain in the run directory; no source/media archives were made.