Caesar III fresh intent: local negatives passed; new capture closed before original
Evidence: scratch/runs/20261118-caesar3-fresh-intent-city-growth/result.json.
No Caesar III original was admitted. Occupied road-connected housing, positive population, natural immigration, money changes, Save and ordinary native Quit remain unproved in this task. No previous city or closed endpoint was reused.
The run-local operator was prepared before controlled capture. It loads the
published cf132e191 helper closure from Git in memory, serves its immutable
production graph, and compiles module
a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8.
All 114 recursive registered Caesar media files were hashed before the
authored scope and verified unchanged afterward. HTTP module bytes match the
compiled module. Independent child debugger inspection was omitted; actual
executing module bytes and child roles are explicitly unknown, rather than
inferred from served bytes. Actual page/document/process/window identity,
current-display readback, independent release and shared disk admission remain
required.
Heavy preparation and release prearming precede each whole capture. Each intent has a new UUID. A strictly local negative invalidates its scene, joins or cancels the old release lease, records zero DOWN dispatches and permits new preparation only while the scope is healthy. The whole-review limit stays 15 seconds; the ordinary release lease is predeclared at 20 seconds. Any uncertain request or other validation failure permanently closes the scope. No in-flight request is relabelled NOT_SENT.
One fresh authored native GDI scope produced two personally reviewed whole browser images of its blue window and instructions:
- An explicit local preflight negative returned
NOT_SENT, cancelled and joined the independent release child, with zero input requests in either transport. - A deliberately short one-second release lease expired before review completed.
The intent returned
NOT_SENTwith zero DOWN requests. Its independently prearmed C UP completed, physical C state was zero, and its child joined. This control is zero DOWN, not zero transport: it contains the real C UP.
Both negatives left the scope healthy. A third, distinct operation successfully prepared a new release lease. Its new capture failed before any review or DOWN: the static window's retained publication was from 10:34:01.911, whereas readback completed at 10:34:32.598. The published unchanged validator rejected the over-30-second display age at 10:34:32.617. This is a harness preparation failure, not evidence of a Caesar III or runtime defect. A new whole frame, accepted new intent and visible input consequence were therefore not proved.
The shared scope closed permanently, all three late ordinary attempts refused locally, and wire audit found zero ordinary requests after closure. The third release lease cancelled and joined with zero input. No old scope was reopened, no cap extended and no second authored scope launched: preparation had declared one authored scope, at most one original, 300 seconds for the control, 180 seconds startup, 80 captures and 160 commands. The unused original policy was 1,200 total, 180 Quit reserve and 20 cleanup seconds. Published prior guard/UI controls remain historical evidence; they do not qualify this failed new-intent control.
Browser, actual driver, native supervisor, watchdog, enclosing launcher, release children, Xvfb, window manager and finite board watcher retain actual close waits with both EOFs. The first board watch's receipt was overwritten by the final finite watch and is missing; the final watch has its own real close/EOF receipt. The outer tool invocation returned exit 1; separate EOF flags for that tool master are unavailable. Native supervision reports no leak; recorded owned PIDs and listeners were independently checked absent, and the ephemeral browser profile and Xauthority were removed. Sampled control disk minimum was 1,777,836,032 bytes, above the unchanged 1,500,000,000-byte floor.
No production code repair or build/test pass is claimed. Only this report is committed; run-local operator inputs, relevant initial/final binary diffs, module, selected images, recovery recipes and actual waits remain in the ignored run. Foreign work, UT partials/ISO and Blood II's hold were preserved. No config, registry, save, media or guest-state edit, forced focus/poll/clock/speed, synthetic ACK, agent, worker, box, Python, Wine or deployment occurred. Box ID null, state NOT_CREATED, expiry null, stop receipt null.