Stronghold current-command review publication barrier
The operator now waits for the current command's immutable review request within that command's original absolute 40-second deadline. A retained request with an older integer command index means the current request is still unpublished. It grants no permission and does not cause a binding failure. Missing publication still consumes the same deadline and terminally fails when it expires. This repairs the operator fetch race exposed by 70b17e57b, described in split gesture reviews.
The replacement run-owned helpers are review_publication.py, review-request.py and ack-review.py in scratch/runs/20261104-stronghold-current-review-publication/. They supersede the predecessor's one-shot fetch and acknowledgement scripts. For a separately authorized replay, stage these three helpers alongside that replay's explicitly authorized runtime_io.py; retain the predecessor's operator-state.py, operate.py, wrapper and route closure unchanged. The entry points import the predecessor state machine by its explicit sibling run path. No transport allocation or launch authorization is supplied by this offline run. Fetch and ACK remain separate invocations, with personal VIEW between them; the fetch never creates a receipt. Do not use the predecessor's one-shot review-request.py for the next replay.
The remote barrier ignores only requests whose integer command index precedes the pending command. Unknown, malformed or future bindings fail. Once the current index appears, the helper validates request fields and selects capture metadata bound to its full PNG SHA-256 and exact tick steps, using the newest capture timestamp no later than the request. Repeated frozen captures may have identical pixels; an older capture at a different tick cannot supply metadata. The downloaded bytes and metadata then pass the unchanged review_bound check. The helper rereads the request and requires exact equality, checks that the operator cursor has not changed, and stores that exact request for separate review. It never silently switches to a replacement request.
ACK again verifies capture hash, command/tick binding and exact request equality after personal VIEW. Local inspection after transport, capture downloads, request reinspection and ACK upload all remain subject to the original deadline. Transport timeouts do not replace it. Any failure persists the existing first terminal reason locally and publishes the terminal latch for the wrapper. Only quit at the next command index remains admissible. Neither helper submits a command or acquires/releases the command owner's lock; the existing nonblocking exclusive lock remains held by operate.py through reply inspection.
The actual wrapper, route driver, input queue, physical driver and freshness checks are unchanged. Fresh personal VIEW, full output-rectangle RGBA, frozen tick, HWND, DOM/session identity and geometry comparisons still precede physical input. There is still one cumulative route driver and one immutable evidence queue: arrival 3072 steps, stability 768, aggregate route 4096, 64 per ACK, navigation 16384 and browser duration 360000ms. The split ChooseType review, hover review and press commands, launch window/TTL/reserve checks, reply-inspection deadline and quit NEXT behavior are retained. No deadline was extended and no counter was reset.
Ten offline controls execute the actual wrapper's personalReview, commandLive and installTransportEvidence functions against a filesystem/subprocess transport. The delayed publisher leaves command8's file in place until command9 publishes; the helper waits and fetches command9, with zero receipt uploads or input before the separate fixture acknowledgement. Another control begins with no request file. Negative controls cover future/unknown binding, malformed current request, wrong capture steps, replacement during transfer, local inspection at the deadline, and ACK reinspection at the deadline. The never-published control times out after 40.063 seconds while the old request remains; the actual wrapper then rejects without publishing a current request or producing input. These are offline guard fixtures, not a live personal review or gameplay claim.
Retained tests pass 18 operator controls, six actual-wrapper review controls, four transport cap/deadline/latch controls, six split-phase controls, nine cumulative route controls, and three positive/48 negative layout cases. A separate process verifies the actual nonblocking exclusive lock; a successful reply inspected at the original deadline remains a terminal failure and allows only quit10 NEXT. Predecessor evidence remains byte-for-byte intact: all 323 sealed files /14995403 bytes verify. Test outputs are redirected into the new run, leaving predecessor artifacts unchanged.
The new run contains test outputs, small inputs, transport logs, exact starting Git SHA, relevant binary diff, new-file patches and an artifact seal. Git remains the source archive; there are no source archives, copied WASM modules or game media. Only this note is committed. Production source is unchanged. No full build, heavy CLI, Chrome launch or box API/lifecycle operation occurred. The original box bx_dkj38gmx was normally stopped at 2026-11-04T12:00:23.614Z, per coordinator evidence; its historical expiry was 12:13:15.406Z. It was not resumed or reassigned here. Coordinator ownership is unchanged. Native replay remains a separate allocation after this offline fix.