SuperTux early second-row platform approach
Evidence: scratch/runs/20261114-supertux-early-platform/result.json.
The sole original died during the second-row approach, before any second-row
jump was submitted. Native Abort Level, map Quit Game and title Quit produced
[Exit] code=0 at 544.486 seconds. The early second-row launch, crossing,
first pit, level exit, map unlock and next-level entry remain unproved. There
was no movement on the new life, retry or second game.
Read-only original level/tile inspection places the second row at columns 34–40, x1088–1312, top y320. Tiles 77, 83 and 102 are solid; solid floor begins at y448 beneath the decorative snow edge. The manual plan was to stage roughly 128–160 pixels before this 128-pixel rise, substantially earlier than the prior failed 34-pixel launch. The first pit remains x3072–3168, with wood platforms x3040–3200 at y352 and x3232–3392 at y256. These data describe geometry; they do not select runtime inputs.
The original README documents P pause/unpause, Escape menus, arrow navigation and Enter level entry. Previously observed native Space Jump/Left Control Run bindings were reused after all 789 original media/settings hashes matched. No configuration, lives, physics, level or save edits, downloads, source-tree archives, media/WASM copies, boxes or deployments were used.
The new tools/supertux-platform-controller.py imports the released Forsaken
transport, including its control-before-fsync fix. Seven shared helper/test
files matched release 95afbb284e71cf1d1b416ff022e0b0b9288a15c4; none was
edited. The adapter enforces an immutable 180-second Quit reserve before
every DOWN and an irreversible manually entered Quit phase. Jump choices bind
personal floor, obstacle and launch-margin reviews. Gameplay/native P choices
require a full scene no older than five seconds. There is no player tracking,
automatic jump, route delivery or automatic pause/resume.
After Q2 guarded 1200 and Airstrike 900 actual terminal/outer waits and heavy releases, the first locked full build passed. The production control suite then reproduced the shared helper's parent-death acknowledgement-bound failure: renderer UP was 1998 ms, but release ACK was 2027.571 ms, exceeding its two-second check. Six other production checks passed. This was a controller fallback/acknowledgement issue, not a guest physics diagnosis.
A focused actual-CLI adapter test reproduced 2013.070 ms ACK before repair.
The owned adapter now durably publishes a 1750 ms fallback before DOWN,
constrains each guard deadline to it, admits pulses up to 1500 ms, and
checks that the complete pulse fits before every edge. The independent child
reads that cap even after parent death before consumed-time schedule delivery.
The same parent-death test passed at 1764.447 ms ACK. Shared files remain
unchanged. Initial failures, the first fixture's insufficient reserve, and
adapter-red-to-green.patch remain in evidence.
Final checks passed: seven policy tests, eight actual-CLI production tests,
29 HTTP transport controls, six terminal controls, three persistence controls,
production guard/input-owner/progress tests and the locked full build.
The eight production checks cover the adapter's parent-death and normal
two-key pulses plus unchanged pipe-EOF, slow-fsync, normal pulse, late-ACK,
busy-server and dropped-ACK controls. The final build used
flock /tmp/wine-build.lock bash tools/build.sh while holding the heavy lock.
The original launched at 19:12:29 UTC, fixed at 1200 seconds, after its
independent 1220-second watchdog was observably prearmed. It used registered
supertux_013, default 1000-block batches, real ticks and production guard
port 8465. A full native title was personally reviewed before the sole bind.
Later fresh scenes, rather than ACKs, established actual UI transitions.
The tested source base is
2bd7e6bfe2580a933159c2338cb93ec50a21e7df; module SHA-256 is
c94ccf1b179f18bd4b47867445afed8cef2f1393b19aabfd259c7f3b94a7be09.
Exact recovery inputs are in source-pins.json, media-pins.json,
runtime.patch, new-runtime.patch and operational-inputs.patch.
Postflight matched 1630 pinned source files, all 789 media/settings files
and the module. Other games' actively owned helpers were excluded from this
control/build freeze; they are not imported by this SuperTux route. Git and
patches are the source archive.
The native baseline was 50 score / two coins / four lives. An early opening
jump from grounded Tux near screen x201, with the first row at x384 and about
152 pixels of leading-edge margin, produced first-jump-pose.png and
first-jump.png. first-row-settled.png then shows Tux standing on that row's
top. This repeats opening progress and is not second-row completion.
paused-floor.png visibly shows the native “PAUSE - press 'P' to play”
overlay; host execution and real ticks continued normally.
Subsequent manually chosen approaches were 350 and 300 ms, separated by fresh
full scenes and released/settled observations. The last reviewed supported
approach, early-approach-fresh.png, shows Tux near screen x201, the second row
near x523 at y320 and an enemy near x518: approximately 290 pixels of
leading-edge margin. Right300 was followed by second-row-staging.png, where
the row is near x289, Tux is absent and an enemy is near x284. The following
second-row-outcome.png proves native respawn: timer 295, three lives,
versus the preceding first-life timer 155.
The short approach did not produce a personally reviewed safe 128–160-pixel staging point. The exact movement/contact interval was not captured. Enemy contact is consistent with these scenes but not individually proved. No second-row jump occurred, so the earlier-launch hypothesis was never tested and the prior late jump was not repeated. No failed jump or approach is called evidence of an emulator physics defect. The pit plan was not reached.
The one-death cap triggered irreversible Quit at 471.863 seconds, before the 1020-second gameplay cutoff. Reviewed Escape/Up/Abort Level, map Escape/Up/Quit Game and title Up/Quit completed native shutdown. Final counters were 50 score / two coins / three lives. No after-cutoff gameplay, deadline extension or native relaunch followed.
All 22 consumed pulses / 23 DOWN receipts retain one permanent owner, unique nonces, independently prearmed release and actual release-child wait 0 with separate stdout/stderr EOFs. Maximum precision-scene age was 4.910 seconds; maximum consumed-to-release ACK was 820.807 ms. All releases received ACK. Five choices were positively NOT_SENT before arm because scenes expired; each used a distinct fresh scene/intent afterward. One optional observer found no armed pulse after such a refusal; its actual child wait 0 and observer failure are retained without an invented airborne frame. The original has no UNKNOWN, replay, rebind or sticky input failure.
Native, watchdog, supervisor-child and additional outer waits are 0, with actual waits and both EOFs recorded. The watchdog exited on the actual native wait without firing; the heavy lock was released. Invoking tools expose combined output, so their separate EOF fields remain unknown. Deliberate parent-SIGKILL regressions explicitly retain missing release-child waits.
Sixteen result-named PNGs passed dimensions, CRC, inflate and SHA validation; unselected PNGs were removed with hashes retained. Runtime PIDs are absent, port 8465 is closed, and about 12.6 GB remained free on /home. No box was created: ID none, state NOT_CREATED, expiry null, stop receipt not applicable. No full27 or parity claim is made.
This bounded task stops here. A new assignment would need to establish an actually reviewed early second-row launch, then first-pit and exit/unlock proof. This run is not a completed crossing and its new life must not be retried.