SuperTux: personally reviewed airborne neutral settle
The review client now offers neutral-settle-200 after personal review of the
exact latest native Pause image. It requires nativePaused=true,
grounded=false, airborne=true, and a nonempty frame-specific review. The
page enables this choice for the airborne attestation and retains grounded
review for all five existing directional/jump choices. Every new capture clears
the attestations. No image analysis, automatic route, or landing claim is added.
The choice sends kind=neutral-settle, keys=[], holdMs=200 through the
existing numbered transport and actual production composition. The persistent
controller accepts that exact combination only: native Escape resume, a fixed
200 ms neutral wait, native Escape Pause, then the composed driver's existing
fresh screenshot. No directional/jump events are dispatched during the settle.
The resulting Pause and grounding states remain UNKNOWN until personal review
of the new capture; a transport completion does not prove landing or survival.
Ordinary empty-key actions remain rejected. Ordinary actions also enforce the
grounded attestation at controller admission.
Identity pins, latest capture matching, original timestamps, single-use capture tokens and command IDs, the 12-second freshness bound, runtime deadline/caps, ACK bounds, independent UP, real outstanding transport promises, and sticky first UNKNOWN remain in force. Cancellation, staleness or identity replacement after resume prevents another DOWN. A settle does not refresh its supporting image or authorize a following ordinary action.
Evidence: scratch/runs/20261109-supertux-reviewed-airborne-settle-0116/.
All checks were offline, using Node fixtures only:
controller-02-final.log.gz: 30 actual Node-owner cases passed, including the fixed neutral wait with only two Escape pairs, ordinary empty-key refusal, wrong neutral duration/keys/observation, stale/future capture, worker/module replacement, cancellation during neutral wait, insufficient deadline, independent UP/late ACK failures, and capture replay.client-03-final.log.gz: 15 control groups passed with 14 real Node children. The actual composed productionexecutefunction ran through the numbered consumer. All five ordinary choices retained their exact key sequences; neutral settle produced only Escape resume/pause and a fresh capture/token. The shipped page's rendering/gating ran in a DOM fixture. HTTP authentication, CLI lifecycle, stale/identity/image/latest/token rejection, and immutable first client UNKNOWN were exercised. Every child was awaited with stream EOF.composition-01.log.gz: actual Node transport/lifecycle test passed, including pending DOWN surviving the ordinary CDP timeout, independent UP, sticky stale refusal without success ACK, provider identity negatives, and terminal drain. Composition forwards the explicit neutral kind.build-final.log.gz:flock /tmp/wine-build.lock bash tools/build.shpassed. CPU SHA-256 stayeda320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295.
The initial client harness omitted the production function's sha dependency;
that offline fixture failed before input and its log is retained. The corrected
second run passed; the final run added all ordinary choices and DOM gating.
Unchanged accepted historical controls were verified and reused by PATH+HASH
in reused-pathhash.json.gz. Source recovery uses Git SHA and small own patches,
without source/media/module archives. Free disk remained above 1.5 GB.
No box, original, browser runtime, worker agent, peer file edit, or deployment was used. Box ID/state/expiry and stop receipt are inapplicable (none created). The next separately authorized gameplay run can personally review a fresh airborne Pause frame, choose one settle, and inspect its fresh result. Actual landing, onward movement, exit, and model deliberation within 12 seconds remain unverified; historical status-143 and other UNKNOWN limits are unchanged.