Playthrough V6: physical work and independent witness

6 October 2026 · local v0.2.3 candidate. This records actual browser input and read-only observations. It is not a release, creative approval or complete player campaign victory.

Evidence method

Work was performed through the visible game: walking, station controls, held pointer/keyboard contacts, shutter adjustment, continuous tracing, Pause, save/reload and depot controls. No hidden-state injection was used. Read-only DOM diagnostics supplied the numerical readings below; those readings confirm state, while the input sequences establish that the interaction was actually performed.

Run A: ordinary service, interruption and a real loss

The protected engine instruction completed through held contacts. At coolant, a fully applied wrong tool incurred its penalty and reset the service sequence; the wrong press did not immediately reveal its correctness. A covenant retained earned witness proof and a partial trace through pause/save. At the repair halt, all three crew duties were replaced through the depot controls.

This run subsequently lost Watch 2 at simulation time 280.75 seconds, with corruption reaching 100. It demonstrates that successful individual tasks and the first halt do not establish campaign survival. No human full-campaign victory is claimed.

Run B: challenge, miswire, counterfeit and return

The shrine covenant was worked with continuous pointer tracing and keyboard stylus guidance. Drawing the unwitnessed 0 → 1 branch produced a visible miswire, cleared proof and required a fresh challenge. Rechallenging restored the independently witnessed route. This is negative as well as positive evidence: submitting a plausible neighboring connection did not complete the rite.

Cooling service then completed through actual Close 1.2 seconds → Bleed 1.8 seconds → Unseal 1.4 seconds contact holds.

At vox, an early counterfeit response was observed about 550 milliseconds into the challenge. The diagnostic charge was 0.1375, the echo was at node 6, and there were no impressed or raised pins. The later physical witness produced the five impressions [0, 1, 4, 3, 6] after an actual 4.25-second hold. The early answer therefore had no matching material proof.

After roughly 0.17 seconds of rightward trace movement, the visible header's Pause control was used. The resting point was approximately (.34, .18); the tool was down=false and earned proof=true. Saving and reloading retained that work without continuing its held input. An explicit regrip and real-time keyboard traversal completed the binding. The vox incident resolved; independent gunner, breech and engine problems remained. At the latest observation, simulation time was 136.2 seconds, with four of each reagent remaining.

Recorded frames:

• Counterfeit answer without physical pins
• Witnessed trace and sealing work
• Miswired branch requiring a new challenge

Remaining gates

The initial 0.66 m and 1.65 m display surfaces remained effectively invisible beside the tracks, even with positive capture statistics. Rotating the eye-image mapping alone did not close that failure. The final bounded 3.5 m surface unfolds the same tiny eye-relative capture past the track edge, with its width still capped at 0.74 m. It changes display size rather than the admitted view.

At normal 1280 × 720 gameplay size, the moving overhead frame visibly shows a small broken terrain strip extending from the left armour slit into black. Turning toward the interior and allowing 200 milliseconds for the next frame removes that left strip; an independently admitted far-side slit can then become visible. The same eye in first person sees the shared green road-cut fragment through the actual tunnel, with all overhead surfaces disabled. This establishes initial visible port presentation, not final scenery composition: the unfolded strip remains small and can look rod-like, and authored exterior art is still required. Screenshot filenames containing “readable” or “final” are evidence labels, not creative acceptance.

The mechanism worker reports 107/107 automated tests, including the later-sender ledger and pause-focus regressions. Those checks are separate from this playthrough and do not establish human difficulty, hand contact, crew art or sound listening approval.

Native and live delivery are verified separately in the release receipts; this playthrough does not certify either channel. Art and listening approval remain open. Cinematic and narrated gameplay still require the creator's demo approval.