Stronghold fresh setup: readiness prerequisite remains unproved, 2026-11-05

The retained command-14 image is populated setup artwork, but it does not prove a completed setup frame or a ready control. The actual native history stops inside an unreturned fourth draw. No recognizer/readiness acceptance change is warranted by this evidence. The three production helpers and their focused tests remain unchanged; no screenshot/hash whitelist, tolerance expansion, unknown-scene bypass or control permission is introduced.

This is the concrete blocker permitted by the assignment's instruction to document missing native prerequisites when control readiness cannot be proved. It does not resolve the runtime prerequisite or authorize another original. Source/retained analysis only was performed, starting from published 666b5ee4c211d20baa5c7927b07bc04084b1683f and the predecessor report. Evidence is in scratch/runs/20261105-stronghold-fresh-setup-readiness/.

Actual images and independent pixel comparison

I personally viewed the original full-size captures, rather than relying on their filenames, prior annotations or hashes. Paths below are under scratch/runs/20261105-stronghold-repaired-bypass-original/browser/.

Capture Steps Visible observation Existing composed recognition
route-native-boundary-7.png 10176 ChooseType / Custom Game hover choose-type-custom-hover
route-native-boundary-8.png 10432 Sword/shield background and unavailable text; no populated panels or numbered buttons setup-transition-observation diagnostic only
route-native-boundary-9.png 10688 Human/Lord Crusader, map, Deathmatch Game, 10000/10000, Crusader Demo row, numbered buttons and unavailable text unknown

The final PNG SHA-256 is 89bb4fe560a2f8cb79b64091597f98c3be9694ea128ec05ee423715a89f37d1f. The preceding PNG is b9ccb3f4a3078cd008964db6794a9a9eb9f6ebd250b4d801b1ea25a4ceee928d. The predecessor note's description of the preceding image as having colored buttons is inaccurate for that retained file. This observation corrects the description without altering its original evidence.

I also viewed historical setup scratch/runs/20261104-stronghold-current-request-native/review/route-native-boundary-12.png. The existing classifier accepts that historical image as actual-scene20-observation; this remains an observation label, not native input permission. Direct decoded-RGBA comparisons cover all 574,480 pixels of the opaque game rectangle (10,194,860,668), plus separately measured Human/Lord, map, Deathmatch/gold, player-list, numbered-button and unavailable-label regions. There is no alignment search, masking, threshold fitting or derived screenshot.

Compared with that historical setup, step10432 differs at 397,531 pixels and step10688 differs at 3,130 pixels. The latter still differs in every named region: 1,841 Human/Lord pixels, 184 map pixels, 410 Deathmatch/gold pixels, 60 player-list pixels and 77 numbered-button pixels. The unavailable-label subregion differs at 1,594 pixels and overlaps the Human/Lord region. These measurements distinguish the actual populated layout from the preceding transition and historical setup without treating a digest as readiness proof. Deleting the unavailable text from recognition would not account for all other differences or establish their acceptability.

Native evidence defeats screenshot-only readiness

browser/preset-attachment-107.json binds each capture SHA, step and epoch to the same original trace. Its actual terminal status is command-failed, fault null, 43,338,271 continuous boundaries. Its last native PC is 0x44f810; the final capture annotation reports 0x44f7d8. Both addresses are inside the original RGB565 lookup pixel loop, verified against original executable bytes. Neither is a draw return.

The trace establishes more than the older partial-frame investigation:

Thus even fade0/direction0/suppression0 is an actual retained negative for the proposed inference “setup is complete and controls are ready.” A prior completed fade-zero draw and a later zero suppression sample cannot be joined into a fabricated completed unsuppressed draw or pixel publication.

Exact missing native prerequisite

The original executable remains at its existing path test/binaries/candidates/stronghold-crusader-demo/installed/App_Executables/Stronghold Crusader multiplayer demo.exe, SHA-256 9024c533095f2227b0105b7e8a9e4cf28ec4fb206b165629787b2c97cf99f13f. Fourteen byte anchors are checked directly; bounded fresh static disassembly is retained in this run. Original setup, draw and widget source contracts agree with the earlier static investigation.

For setup-frame completion, missing evidence is a real same-owner/epoch return of the unsuppressed fade-zero draw, followed by an identified ordinary presentation of its output and fresh independently reviewed pixels. The actual pending fourth draw cannot supply that return. Neither its remaining boundary headroom nor a host capture supplies the missing presentation identity.

For the numbered preset, original hover 0x42d850 checks mode99 and feedback [0x995674]; active input suppression clears feedback at 0x4e9ee0–0x4e9ef3. Original activation 0x42d880 independently requires mode99 and [0x69165c]==0 (no pending preset). The retained observer reads mode and fade, but does not read pending preset or widget feedback/activation state. Generic widget visibility/disabled flags and the selected widget's ownership also need evidence before interpreting artwork as an enabled control.

Application is another event: updater 0x42db40 tests the pending preset and native press edge [0x1bd26a4], then calls 0x42e490 and clears pending state. The native edge and application are not observed here. A press edge would have to come from an authorized ordinary gesture, not be required or fabricated at host rest. Hover, activation and application must retain their distinct evidence; none is inferred from a visible numbered circle or description. Strict Zephyr/Start/world conditions remain separate and unproved.

The existing helper intentionally forbids route-stability, route-review, route-next, preset hover/press, handoff and native gestures in this diagnostic lane. Even the historical complete-layout label stops with “no readiness grant.” Extending the fresh image list would at most permit more diagnostic arrival observations; it would not remedy the native/control prerequisite. No such extension is installed. Unknown scenes still stop at first failure.

Scope, validation and handoff

analyze-retained.cjs performs only retained-image comparisons, real capture/ epoch/ACK/report binding checks, current recognizer calls and original-EXE byte inspection. analysis.json and analysis.log retain the results and all existing input paths/SHA-256 values. Known ChooseType and historical setup are positive recognition controls; the background-only transition and current unknown frame distinguish the actual cases. There is no retained positive for fresh control readiness, and no synthetic native history is substituted.

No shared code edit, executable heavy suite or full build was performed. ReturnFire0305 then MDK0320 retain their queue priority and neither FINAL release had arrived at the source-analysis decision. Their release/cleanup gate was not bypassed. No code implementation is being committed, so a code build pass is not claimed or reused as fresh validation. Documentation is checked with git diff --check.

The original first failure/reply14, terminal latch, complete ACK/full-rest contract, original clock, unknown-scene guards and 50M/360s/2048-record/1MB caps remain intact. No command15, browser/game/original attempt, native fixture, guest write, forced render/load/time, second attempt or new runtime occurred. The next technical prerequisite is the missing native draw/publication and control-state evidence above, within unchanged caps and first-failure rules; this report grants no new runtime authority.

Recovery uses exact Git SHA plus relevant binary diff and owned new-file patches. Existing captures/media/modules stay at their original paths; no source archive, snapshot, tar or repeated media/module copy is created. Box ID/state/expiry: N/A; no assigned box was used. Shutdown ownership remains with the coordinator. This worker releases its documentation/evidence and stops.