Caesar III: housing placement, parent loss, and cancellation repair
Run: scratch/runs/20261117-caesar3-reviewed-hover-city-construction/result.json.
The one authorized original reached Assignment 1, Aventine. A personally reviewed housing placement at (420,470) returned an actual owner USER message within 254.326 ms, independent physical UP within 430.188 ms, and a complete observation within 649.636 ms. The later city image shows Dn 1,990, down from Dn 2,000, and a housing placement marker. Population remained 0. This proves a charged housing plot placement; it does not prove an occupied dwelling. No new road, population increase, or native Quit was completed.
The original used published c993e385cf5be638ffec7594da6ebb392aa1ea22, module
9099bb3f694d5ec6dac2886b70c1920fa33f39ad9258b1e662d1ebe05dc2da0a, unchanged
defaults, threads and real ticks. All 114 recursive media paths, sizes and SHA-256
values matched before the native clock and after closure. The first capture
followed a passive current-owner unfiltered GetMessageA return, about 54.45 s
after startup. Fourteen finite actions returned native confirmation, including
separate pointer arrivals and freshly reviewed presses. Every original image
used for an action was personally viewed. No forced pump, synthetic receipt,
private game-state read, media modification, or second original was used.
The earliest terminal boundary was launch-parent loss. The outer supervisor's pinned parent process 740412 disappeared at 1794908003.8794484, about 268.56 s after the native clock began. The outer correctly cancelled its process scope. It retained actual child and both watchdog waits, both EOFs, and an empty final scope. The origin of the parent loss is unknown. A later road-tool attempt used a 34.009 s old city image and was correctly refused before intent/DOWN, latching permanent UNKNOWN. There were zero later HTTP requests, retries or rebindings.
The original adapter imported the shared VB runtime without installing its shutdown handlers. Consequently outer TERM killed the adapter before its native wait, drain joins, watchdog wait, and descriptor-release receipts were retained. The original has a native cancellation receipt, but no retained actual native wait or native EOF proof. Outer waits do not substitute for those missing records, and host cancellation does not constitute native Quit.
The repair installs the imported shutdown handlers, factors native execution
into the same cancellation-safe path used by authored tests, reads partial pipe
output promptly with read1, and records actual waits and both EOFs in finally.
No input deadline, image age, game clock, guest code or shared helper changed.
Real authored TERM and INT controls respectively exit 143 and 130 while retaining
native exit 0, actual wait, both EOFs, independent watchdog wait and explicit
host-cleanup disposition. This repairs missing cancellation evidence; it does
not establish why the launch shell disappeared or authorize another original.
The first full preflight failed before any original because its independent UP ACK took 1,987 ms against the unchanged 1,750 ms limit. It remains recorded as a failure. After the locked build finished, full qualification passed before the single original launch. The exact published slow-frame combined-expiry negative and separate-arrival/press positive are reused only after checking identical source maps and module bytes. Repair qualifications use the pinned c993 graph plus the recorded owned patch, with a freshly compiled retained module and main/CPUWorker/OS-child seals. No whole-graph equality with unrelated later main commits is asserted.
The repair's focused TERM/INT tests, full authored native controls (including
pointer mapping, printable input, all ten consumer modes, freshness and deadline
refusals), and actual outer printable-input run passed. Native, outer and both
outer watchdogs were actually waited with both EOFs and an empty remaining scope.
The required locked build passed before publication. repair-verification.json
records that only the two owned adapter/test files differ from the original
source map and that module bytes remain equal.
Next work requires separate original authorization, a launch owner that survives
transient tool sessions, and a fresh city review before road placement. Both
prior original endpoints remain permanently closed. No box or worker was
created; box ID and expiry are null, state NOT_CREATED, stop not applicable.