MDK2 optimized observer: one original menu/Medium assessment

This assignment assesses published 2c8c7cc1fd353af2df2f7c1b3d3240bb40c69c1b in one ordinary original menu/Medium attempt. Its numeric comparison retains all35 pins/502 bytes and534 translations per fixture boundary. The passive fixture speed is not a gameplay measurement.

Evidence: scratch/runs/20261105-mdk2-optimized-original/ (RUN). Git is the source archive; evidence records the exact Git SHA, relevant binary diff and new-file patches, with original media/modules referenced by existing path and SHA only. No source snapshots, archives, copied modules or media are used.

Preparation began offline while stronghold-review-ack-original-0036 owned the heavy runtime. Source/module mutation and the full build were gated on its final runtime AND file release and independent verification of exact PID/startticks, ports and profile cleanup. Return Fire waits this assignment's final release. All builds must finish before that release.

The original50M/360s observer, owner/record/byte/input/ACK checks, active0 full-rest requirements, frozen review,16ms and512/544 limits remain. Queued Enter and its asynchronous-publication stamp are distinct from a native command press, balanced keyup/full rest, decoder return, maintenance return, loading and gameplay. Uncaptured memory remains unknown. No timer, render or guest call is forced. Every fresh checkpoint used for control requires personal pixel review. The first post-Chrome fault ends the attempt with normal browser close; any later remedy is offline and cannot authorize a second original.

Chrome cutoff is 2026-11-05T01:40:00Z, cleanup cutoff 01:43:00Z. Existing720-second browser,60-second bootstrap,40-second command and180-second cleanup caps stay unchanged. Box ID/state/expiry: NA; coordinator owns shutdown.

Stronghold's final runtime release at00:53:36Z and file release at00:59:49Z preceded every source edit and build. The independent check confirms original Chrome1570178/start188692357, Node1570155/start188692178 and flock1570154/start188692178 absent, HTTP34441/DevTools39957 refused, exact profile absent and external flock free. predecessor-cleanup-independent.json records it.

The old historical432-row manifest and accepted439-row manifest are retained unchanged. A separate current binding uses source 17e09475ecaeaa47de76684df0a0518124a6e387 and adds exactly these amendments, each attributed to that final corrected raster publication. All other served JavaScript/source rows keep their accepted hashes; no blanket rehash occurs.

Dependency Accepted SHA-256 Current SHA-256
src/09ab-handlers-d3dim-core.wat 5c892770004355099d016e301ff73366a560237d681faf0cd3c6bd3aa8e4fc07 dbcf489a787b4f399eadc926e13336976d6eb0cc30ffed5c8c360af5b496a75a
build/wine-assembly.wasm 018c94ffbde8c857587c25c860db988d603d0662df9a18901f00c183af7493cd e620f3adfc2850580004e204bc1fea495844e942806866f8ada24c215212e412
build/wine-assembly.compat.wasm 4e684af48e4ec06ae8aada0d5cd2b522345a7a51b8ed89988d4d8fa0d75ea7c2 66b37bd76da5a7b49c063b4eb0d44c898aaf00ff41e000a6fbd87c01edc136c8

The actual integration/preload/inspection explicitly selects current; the accepted and historical resolver options remain available for their original regressions. The original five controller sources receive only their two exact source/canonical identity substitutions. Assignment/deadlines and the named predecessor release gate are renewed without extending any relative cap. The optimized observer itself remains byte-identical to published2c8c7cc1f. RUN/caller-binding-ledger.json records old/new helper/controller hashes and provenance; run-specific caller inputs reference existing original paths.

Each continuous-trace checkpoint now uses the existing passive host memory hook to read the next48 stack words after the original48. A uniquely ordered 47ca7e → 47a928 → 47b393 saved chain and four exact original instruction windows must match before interpreting the saved stream and counter. The stream's128 bytes must bind callback47a880 and its actual stack argument. The counter lies52 bytes beyond the saved47b393 slot. This captures frozen state, not a witnessed callback/maintenance return. Missing mappings, changed instructions, absent frames or incompatible callbacks return explicit unknown with outer:null. All-Worker ownership/frozen fences remain before and after the read. There is no new per-boundary observer work or changed observer byte, record, deadline or instruction-pin allowance.

Validation before publication:45 local tests and24 explicitly selected integration tests pass, including the unchanged pin comparison mutation and failure-precedence cases. Thirteen current-binding stale/hash/module/source/ selected-live-dependency negatives pass. The passive reader tests scattered byte mappings,129 missing-byte locations, four changed instruction windows, missing saved frames, callback mismatch, frozen ownership and advancement; full fixture memory hashes stay equal. The full command flock /tmp/wine-build.lock bash tools/build.sh passes and reproduces both current module hashes without copying or restoring modules.

Initial fixture mistakes are retained: the passive-reader fixture allocated14 bytes for the15-byte outer loop pin, and one private install test used its old clock. Both corrected fixtures pass. The first negative harness altered every Git revision and therefore rejected a historical source before reaching the intended current check; its corrected scope changes only accepted/current Git reads. These are offline fixture failures, not original attempts.

Implementation 5e2542a569861d7458f336784b41012538d72926 was published with immediate strict pinned SSH pull/push before execution. The full actual composed FS+HTTP route then passed439 FS,439 proof HTTP,439 actual HTTP and2253 installed-caller checks with plain Puppeteer mocked, public/native SPAWN_FORBIDDEN, zero Chrome launches and both temporary servers closed. preflight/ and preflight.log preserve that boundary. The actual original page compiled, instantiated and sent the canonical current e620 module; the owning worker trace independently binds those bytes, slot0 and endpoint1.

ONE original launched at01:06:10.512Z under exclusive /tmp/wine-local-runtime.lock. Eleven fresh PNGs were personally inspected: startup0/512/1024/1536/2048/2560/3072/3584 and native selectors64/67/77. personal-views.jsonl binds each review to its PNG SHA. Startup progressed through gray/black surfaces, BioWare splash and MDK2 title/legal overlay to New Game. Selector64 showed New Game; selector67 showed the difficulty dialog with Medium highlighted. Selector77 showed the MDK2 background without the difficulty dialog or icons, with no comic, loading text, world or gameplay.

The existing ordinary controls used observe16, New Game64 and Medium304 steps:384/512 aggregate menu steps and53/113 commands. No544 release reserve was needed; no control limit, ACK,16ms, observer50M/360s or active0 full-rest guard changed. The first Medium checkpoints showed queued WM_KEYDOWN13 and WM_CHAR13 with asyncPressPublished:true, while native command62 remained raw/previous/active0 and the difficulty menu remained open. At global3840, the queue emptied and native command62 reached phase4, raw1/previous1/active0. This satisfies the original native press predicate; it must not be described as active1. Keyup was subsequently queued. At3952 it was consumed with active1 and showmenu0; only at3968 did phase/raw/previous/active all return0, with empty input queue, null active input and mouse mask0. The ordinary controller then accepted full rest. No pending/loading/cancel/comic/gameplay control was sent after this successful Medium assessment.

The continuous trace spans546000 native boundaries with19 samples, 2 decoder-output records,2 decoder returns,1 GC entry/return,3 deferred-load checks,1 command62 dispatch entry/return, and1 timer-set entry/armed/return. Decoder returns to47ca7e occur at blocks164483 and499381. GC enters443070 at 164511 and returns40218c at301232. Command62 dispatch enters42ae70 at500479 and returns46d0c4 at504653. Timer-set enters46cb20 at503366 and returns41e9b2 at503367; the captured timer object138660096 contains countdown 1.2000000476837158, with no inferred wall deadline. There is no captured timer handler, demand-load/type10/new-level entry, loading or gameplay progression. The three deferred checks each retain requestedLevel/checkpoint/loadFlag0; these are captured values, not defaults for absent memory.

All19 correlation snapshots preserve frozen16ms ownership and thread1, with no mismatching duplicate checkpoint. Ten have the supported saved decoder / stream frame chain. Each captures stream138913920, callback47a880, argument404232216, buffer14348388, DirectSound buffer136451064, flags298, blockBytes32768 and refillIterations1 at address122682724. They cover the first decoder at3680–3744 and the second at3872–3936. The other9 snapshots retain outer:null and the explicit absent-frame reason; they are not counter0. The stream's captured words25/28/29 change from3025920/3/142336 to 3065856/4/182272 between those groups. This is frozen stream/refill state, not a direct callback-return event. No direct outer callback or maintenance return hook was captured. The original static call chain places maintenance before GC entry, so its prior return is an inference from native continuation, not a separately witnessed maintenance return. original-analysis.json retains all19 snapshots and the native event sequence.

This establishes that the optimized observer can complete this original menu/Medium boundary under the existing guards. It does not establish a controlled whole-history speedup or gameplay speed. The prior accepted attempt left Medium queued at4096; this attempt reaches native press, key release and full rest at3968. However, New Game needed64 steps here versus48 before, so the Medium trace begins3664 rather than3648, and the explicitly amended current module differs from the earlier accepted module. These confounds prevent sole causal attribution to the numeric comparison. The35-pin/502-byte/ 534-translation equivalence and passive fixture timing remain separate facts. No second original was attempted and no gameplay result is claimed.

The original had no surfaced post-Chrome fault. After the final pixel review, the normal quit path called Browser.close at01:12:31.165Z, Chrome exited0/null at01:12:31.312Z, and Browser.close resolved01:12:31.313Z. No kill fallback ran. Independent cleanup at01:13:00.062Z verified Chrome1594185/start188797975, Node1593763/start188797530 and flock1593762/start188797530 absent; proof HTTP46457, actual HTTP39117 and DevTools46143 refused; exact profile absent; and external flock inode141907 free. cleanup-verification.json records this. The local heavy runtime was released01:13:11Z; file ownership is released only after this final note is published. All builds are finished. Return Fire may then independently verify the recorded exact cleanup before its own attempt. Box ID/state/expiry remain NA; coordinator owns shutdown.

Final offline audit rechecked all11 PNG hashes,19 frozen ownership snapshots, 384-step/53-command ledger, exact module binding and final active0 full rest. result.json is reviewed and scoped to menu/Medium, with loading/gameplay false. Relevant source evidence remains Git plus binary diff/new-file patch and existing input SHA references; no snapshots or media/module copies were added.