Find the work — interface review, v0.3.2
6 October 2026. This is an implemented usability revision with actual control evidence. It does not approve final art, listening, difficulty or the completed demo.
Findings reproduced in v0.3.1
- Entering first person could show a wall without a useful target. E selected the nearest station regardless of which way the player looked, and did not turn to it.
- The opening objective listed three unfamiliar tool acts before providing a clear route or preparation step. It did not change with ritual progress.
- Overhead station labels were tiny, abbreviated and restricted to the exact room. The player was hard to locate. Clicking a distant station had little visible route feedback.
- A station showed description, empty diagnosis space, repair, purge and ritual options at once. The one necessary act sat below substantial prose. Ritual labels were as small as 10 px.
- In first person, clicking the scene could fire a weapon when a player intended to interact. A keyboard Space on an ordinary focused button could also reach the fire handler.
- Opening the manual from Pause and leaving it resumed the simulation unexpectedly. Menus did not isolate background keyboard focus.
Implemented
The duty card gives one reported problem, its reason, and a route or interaction button. Engine, Cooling, Cannon and crew names are consistent between the tabs, scene, prompt and panel; religious names remain as subtitles. Station tabs show explicit text states, including crew lapse, loose witness, reported sign and work in progress. Guidance uses announced symptoms and visible gauges; it does not read a hidden covenant route, taint value or correct post-lesson tool choice.
The overhead view marks YOU, names local/adjacent stations and traces the current walking route through the exposed passages. The rest of the tank remains dim. In first person, the nearby interaction gets a named E prompt and an attached label. E favors an accessible station ahead, can explicitly turn to one behind, and frames the station when opened. It cannot operate a nearby station through a solid rib. Clicking a station tab still walks to its actual approach, with controls opening on arrival.
First person distinguishes hands free from pistol drawn. Q draws/holsters; the reticle, weapon model and controls reflect that state. A hands-free click opens the nearby station. Drawing clears service tools and stops automated walking; a nearby bodily threat offers an explicit defend action. The opening weapon is secured. After a kill, the wound route and three-second hold instructions remain visible until binding succeeds. Threat guidance requires an actual line of fire, not just distance through a wall.
Station panels lead with purpose, current symptom and the main act. Extra orders, emergency repair and advanced rites use a disclosure. Unavailable procedures show a written reason and resource requirements. Ritual instructions and witnesses have larger, clearer typography. The protected first rite numbers its tools, marks the current and completed acts, and gives an exact held-input instruction for each stage. Later rites retain diagnosis and wrong-choice consequences. Closing work explicitly warns that spent materials remain spent.
Esc always pauses. X closes the station; it interrupts unfinished work. The short field manual starts with navigation, interaction, held work and combat; detailed procedures are expandable. The game behind a modal is inert, focus stays inside the modal, and help entered from Pause returns to Pause.
Actual input QA
Local play used ordinary browser mouse/keyboard input. Read-only diagnostics supplied the final receipt; no simulation commands or state were injected.
- Started fresh, used the duty route to reach Engine, switched views, closed its panel with X and reopened the named first-person target with E.
- A short tool click did not advance the protected rite. A sustained 2.3-second grip advanced isolation. Paused, reloaded and continued: the paid second step remained prepared. Help from Pause returned to Pause. Completed the unction and answering contacts.
- Followed the Loader warning and route. His panel presented correction as the primary act; correction cleared the lapse and returned inspection to the primary position.
- Drew the pistol from the combat cue. Two ordinary aimed shots killed a small body. The source remained open. After a deliberate inspection delay, its returning body required another two shots. Holding the visible cautery control bound the source and spent one seal. Kill count 2; bound count 1.
- Walked to Cooling, opened the emergency disclosure, saw why an undiagnosed purge was unavailable, and performed a parts repair. It spent one part and six spirit. Started the conditional cooling rite and completed its first two held acts; arrival at the repair halt interrupted the final act. Do not claim that rite completed.
- Reached the first halt at mission time 168.2 seconds; total game time 318.9 includes protected instruction. Patched hull 66.4, spirit 67, parts 5, seals 5. Maintenance was deliberately delayed during interface inspection; this is not an optimal play or full campaign balance run. Halt receipt.
- A separate protected replay checked 1280×720, 1024×768 and 390×844. No horizontal overflow. All three held tools and the close control fit within the operating panel at the measured sizes. The narrow check establishes layout only, not complete touch gameplay.
- No captured runtime error or warning in the inspected local run. Sound remained off; no listening approval is inferred.
Overhead route · 1024-pixel controls · First-person rite.
Automated guidance tests cover the staged lesson, actual station approaches in both views, blocked interactions through machinery, public symptom privacy, urgent defence while working, living body versus wound binding, and resolution of crew/pressure duties. All 134 automated tests pass. The source-reference check accompanies release verification.
Remaining usability work
This is an agent-driven expert playtest, not a first-time human usability study. The complete covenant, measured-relief and loaded-bearing journeys need broader novice testing. Final physical instrument art and priest/tool contact still need work. No full three-watch victory, full touch journey, sustained listening approval or demo acceptance is claimed. Cinematics/narration still require creator approval of the demo.