Caesar III: ordinary owner readiness and the first city input

The one authorized original reached Assignment 1, Aventine, with Dn 2000, Pop 0, Jan 500 BC. It did not build housing or a road or change population. The first housing-selection press expired at the unchanged 500 ms native confirmation deadline. UNKNOWN permanently closed the endpoint; there were zero subsequent HTTP requests, retries, rebinds or replays. Shutdown was host cleanup with actual process waits and both pipe EOFs, not guest/native Quit.

Evidence: scratch/runs/20261117-caesar3-owner-ready-city-building-progress/result.json. The original allowance is spent. No boxes or other workers were created.

Published readiness repair and original proof

Commit 583db55a972328829ddddb91a30c7fd1bc1b3f7a fences passive readiness to a fresh ordinary public GetMessage/PeekMessage return on the current process, owner thread and live window generation. Entry and return must retain the same exports object and memory buffer, focus/main window context, unfiltered target, valid stack and no yield. Correct owner PM_NOREMOVE with an empty return remains valid readiness. A readiness observation is not an input ACK. No forced pump, idle or callback is introduced.

The four-second authored first-pump delay produced zero pre-pump captures, binds or DOWNs, followed by a strict native positive. Stale, foreign, filtered, changed-memory and changed-generation observations are rejected. Commit 9d6f127f65d1a3f0209c9a4bf2a8a2a34d51b74b bounds the extra readiness trace allowance and joins the fixture watchdog before assertions. The full native qualification passed on that exact published graph, including its parent supervisor, main/CPUWorker/OS-child inheritance and actual waits.

The original freshly compiled that same immutable graph, module SHA-256 9099bb3f694d5ec6dac2886b70c1920fa33f39ad9258b1e662d1ebe05dc2da0a. All 114 recursive media entries were checked before the clock and unchanged after shutdown. Defaults and real ticks were retained. The 2400-second native deadline, independent 2420-second watchdog and 240-second preparation reserve were prearmed. Passive readiness arrived about 29 seconds after launch. Only then was the complete title image captured and personally reviewed.

Five separately reviewed finite actions received native confirmation: dismiss title, Start New, enter C, Continue and enter the city. Their capture-to-intent ages were 7.085–8.979 seconds. The city image showed the genuine map and construction controls. Housing selection used an 8.333-second-old personally reviewed image, within the published 15-second gameplay policy. Menu policy remains 30 seconds; native ACK/independent UP/immutable observation limits remain 500/1750/3000 ms.

Measured next cause and tested repair

Retained public native observations locate the housing failure:

Event Time after housing intent
Ordinary WM_MOUSEMOVE returned 251.211 ms
Ordinary empty owner poll completed the hover barrier 255.847 ms
Expired DOWN purged 499.651 ms
Next ordinary owner poll 507.699 ms
Independent physical UP 1457.266 ms

The combined arrival/press queued DOWN behind the existing hover barrier. Ordinary city work delayed the next poll beyond its fixed deadline. DOWN never returned from the native owner. Handler application alone was correctly rejected. The public trace supports this timing cause; it does not establish any built structure or successful housing selection.

Commit c993e385cf5be638ffec7594da6ebb392aa1ea22 adds explicit finite hover-intent and review-hover-intent modes. A person may review and confirm actual WM_MOUSEMOVE first, then obtain and review a fresh image before a separate press. Each action keeps the existing actor, focus, modal, generation, context, nonce, target, release and deadline checks. Nothing schedules a later press or drives the game automatically.

An authored fixture performs 350 ms of real ordinary frame work. Combined arrival/DOWN expires at 500 ms, becomes permanent UNKNOWN and makes zero later HTTP requests. Separately reviewed movement and press both receive real native owner returns, independent UP and immutable captures; the final ordinary snapshot verifies the callback's UP after DOWN title. On exact published c993, movement/press native confirmations were 262.587/254.626 ms, independent UP 383.896/411.836 ms, aggregate observation 424.795/454.871 ms. These are authored controls, not another original or a claimed city consequence.

The repair's exact published graph was freshly compiled and sealed and its focused positive/refusal tests passed. Its module and all engine, host and runtime sources match the original proof. The complete graphs differ in the Caesar helper/test and unrelated helpers; c993 is not claimed to have the 9d6 whole-graph qualification. Both required locked builds passed. The early trace-bound failure and two working hover-test failures are retained with their actual waits and recovery metadata; they are not counted as passes.

Next: a separately authorized original can use an explicitly reviewed native pointer arrival, then a fresh review and press, to attempt housing placement and observe a real structure or population change. The closed original here must never be queried or resumed.