Actual-input review — v0.4.0

Review date: 6 October 2026. This record separates deterministic verification from playing the rendered build. No simulation state was injected to complete the browser station checks. A safe practice watch was added to compare the offices without repeated interruptions; its completions are not campaign victories.

Browser observations

• In a focused Chrome tab, Q entered first person, drew the pistol and obtained actual native capture. Moving the mouse without holding a button changed yaw from approximately -1.621 to -2.399 and pitch from -0.613 to -0.470. Q released capture and holstered. A background-tab refusal correctly showed the capture prompt; focusing the tab resolved it.
• In an actual campaign, a nearby intruder's seventh strike switched the overhead camera into first person, drew the pistol, faced the attacker and captured the mouse. Two genuine Space shots killed the small manifestation. Killing left its wound open. Holstering returned free interaction.
• The shrine censer charged all five wax witnesses through control movement, then settled at its central hook. Spirit rose from approximately 81 to 98. An attack immediately afterwards drew the pistol; the completed office was not lost.
• Engine play exposed an excessively fast oil feed. Its feed and return response were slowed, then the physical oil-film lesson was completed through its real controls. Abruptly returning the weight produced bearing wear, as intended.
• The loader's shell and locking collar completed in practice: discipline reached 90, one seal was spent, and a forced shell movement left a small penalty. The gunner's rate wheel completed by following the visible moving reference; discipline reached 100 and one incense was spent.
• The cannon counterweight completed after its spring settled, improving the mechanism from 94 to 100. The command governor completed at the advance mark, restoring discipline to 100 while retaining the reckless trait. Moving the commander beside the bore required moving the priest's approach to the opposite side; the revised route was played.
• The new direct relief wheel was over-opened during review. The gauge reached zero, the interface reported 15 seal-wear marks, and closing the wheel allowed the real pressure to recover. This was a failure-and-recovery check, not a clean first attempt. A second measured bleed closed at 46.3, seated its return and completed. The low-pressure filter then followed the moving visible band through genuine mouse turns; heat fell from 37 to 21, pressure from 48 to 40, and the mechanism reached 100.

• The final practice return restored the campaign at 73.928 percent progress, mission time about 113 seconds and oil/incense/parts/seals 5/5/6/5. Q was also checked from the focused censer plate: it cancelled paid work, entered first person and requested capture. Resume visibly obtained native capture; Escape released it and left the watch paused.

Current screenshots: censer and longer overhead cones, gunner sight exercise, filter wheel in first person, captured pistol. Machine-readable observations: native capture and Escape, restored campaign, completed filter practice state.

• A fresh isolated origin confirmed immediate moving tracks and battlefield, two covering cannon shots, a running sound context, and Q/click native capture during the protected introduction. Q holstered and restored free controls. The final 240-body/18-rifleman scene ran at 59 FPS in this focused check. Introduction observation and moving overhead view retain the evidence.

Fixes arising from review

The first campaign attempt was left unattended while inspecting feedback and ended in defeat. Repeated Space firing could then activate the focused new-watch button; that was fixed. Lost focus now pauses the watch. Escape after native capture has a short event guard to prevent a release event from immediately undoing the pause. Capture is requested again from Resume.

Q/F/X were initially swallowed while an SVG tool had keyboard focus; they now remain available alongside its adjustment keys. Q and finite-ammunition fire also work during the moving protected introduction. Station selection clears stale movement input. The high/low cooling choice no longer rebuilds an active handwheel while pressure crosses its threshold. Shrine settling time now has a visible disabled reason. The desktop container's former deny-all permission handler was narrowed to permit the game's own pointer capture. Practice reloads now finish while its campaign clock remains suspended.

Objective checks

151 automated checks cover simulation, material costs and wear, all six new offices, impossible input and save states, fixed-step progress, interruption, cannon interlocks, actual close-strike records and wall occlusion, bounded battlefield events, audio cadence and ducking, native-capture lifecycle, navigation, rendering contracts and packaging policy. Static reference and syntax checks are run on the frozen release source. Native platform and publication receipts are recorded separately after their runs complete.

Focused Chrome rendered at about 60 FPS during the station checks, at a 720-pixel-wide internal framebuffer. Before the final exterior crowd increase, approximately 206 draw calls / 55k triangles were observed overhead and 98 / 38k in a first-person view; counts vary with visible compartments and effects. Background-tab throttling produced much lower samples and was not treated as an active-play performance result.

Acceptance still open

No complete three-watch victory is claimed for this revision. Long-run balance, first-time human comprehension, physical-tool preference, final character/environment fidelity, sustained listening and the cinematic remain in review. The purpose of this build is to give the user distinct, usable mechanics to compare in a running game.