Iron Litany — actual browser playthrough, October 5, 2026

Method and acceptance boundary

Played the local candidate at http://127.0.0.1:8786/game/ through visible game controls in the Codex in-app browser. The shared Chrome session became unavailable, so this run resumed the in-app browser's older, protected first-watch tutorial save. It did not resume the parent's separate Chrome second-watch save.

Movement used the actual station walking buttons and floor clicks. Camera changes, religious acts, inspection, correction, repair, purging, crew replacements, requisitioning, saving, pausing and restoration used the game's controls. No simulation state was injected, no time was accelerated, and no console command changed gameplay. document.body.dataset.gameState was read only to record instruments and the result. Pauses were used while inspecting evidence and planning the next action.

The report proves a complete playable campaign and the specific interactions below. It does not establish human audio acceptance, completed art direction, or a verified outdoor glimpse.

Completed campaign

The final screen displayed SANCTUARY REACHED — The machine endures. The final visible result was Hull 100 · Spirit 94 · 54 shells fired · 3 crew replaced. The read-only state recorded phase: "victory", leg 3, spirit 93.5273, three replacements, 54 cannon responses, six oil, six incense, five parts and six seals. The HUD rounded spirit to 94. The repair halt patched the hull; immediately before the final halt it was approximately 95, so the final 100 does not mean there was no damage during the watch.

Evidence: actual sanctuary screen and read-only final state. The result tab was marked to remain open. No defeat occurred in this run. Victory appeared at the last gate; a separate final-report button was not exercised.

First crossing: training, correction and the first repair halt

Resumed the saved protected tutorial, walked to the Motive Altar and performed all three tutorial responses. Completing the final response started the moving watch, with actual loading and cannon fire.

Walked to Loader Holt, inspected his bearing and issued formal correction after the initial profanation. Discipline rose from approximately 53 to 88 and the immediate lapse was arrested. This did not remove the underlying Mutinous trait.

An unattended interval during context recovery advanced the watch from approximately 18% to 93% of the crossing. This was a testing-process interruption, not a demonstration of efficient real-time play. Six faults accumulated; cooling fell to approximately 23%, heat reached 92 degrees, pressure reached 94%, and hull remained approximately 93. This interval is explicitly included in the result rather than hidden or reset.

At the cooling reliquary, the observation described a strong pump and a pressure head trapped behind the return. Performed close return → bleed trapped pressure → unseal return. The rite cleared the cooling fault and its old recoil disturbance, lowered heat and pressure, and allowed arrival at the first repair halt.

At the halt, replaced Commander Rusk with Vale, Gunner Venn with Korda and Loader Holt with Ilya. The counter visibly reached three replacements. Spent the remaining five allotment on three oil and two incense, saved at the halt, reloaded the page and used Continue saved watch. The replacement names, resources and repair-halt phase survived the reload. This also loaded the revised first-person loader approach.

Second crossing: mechanical need versus intrusion

The carried engine recoil witness was settled with the free 1.4-second response. The dry-bearing observation then required isolate motive feed → anoint responsive bearing → double relay response. This cleared the physical bearing fault and restored motive function.

Walked to the vox, used Inspect the mechanism, then performed the targeted purge. The recorded chaos vector disappeared. Walked to the shrine and used the personal examination to clear the custodian's compromised internal response. A separate breech recoil witness was reseated without spending a reagent.

Cooling attention at approximately 40% progress used the strong-pump/pressurized branch again; the gauge was around 72%. Completing it reduced heat from approximately 81 to 61 degrees and pressure from approximately 72 to 54%.

Deliberately selected Bank the motive demand against a dry-bearing observation. The game rejected the act, retained step zero, reduced spirit by three and trust by two, increased heat, and displayed a consequence tied to the physical witness. Correct isolation, unction and the double response then completed the rite. A wrong choice is therefore not silently accepted as an equivalent ritual button.

The breech observation described a hot chamber with already-oiled teeth. Performed lock firing circuit → vent hot chamber → witnessed dry cycle. The firing sequence paused during the observance and resumed after completion. Later personal examination and a free coolant recoil response completed successfully. Arrived at the second halt with three replacement crew still in service and no defeat, then replenished ritual supplies, parts and seals for the final crossing.

Final crossing: expedience, hot motive rite and false answer

Used an expedient part repair on the carried breech obstruction. It restored physical function while costing six spirit and producing the pipe-bellow/uneven-relay caption, documenting the actual tradeoff.

The port-visibility check below consumed substantial live time. Mechanical and religious threats continued to accumulate while the player was looking outside.

Returned to the engine when the casing was approximately 81 degrees and motive function had fallen sharply. The observation said full demand was heating trapped air and fresh oil would cook. This time the correct sequence was bank motive demand → vent overheated casing → slow piston cadence. The same Bank act that had failed against the earlier dry-bearing condition was accepted against the hot condition. Followed with an expedient repair to recover function before travelling to the shrine.

At the shrine, an answer arrived before the mechanical relay moved, and the custodian's own cant was compromised. Performed break neural tether → compare mechanical witness → seal foreign response. This removed the shrine intrusion, the personal intrusion and the shrine's physical recoil fault through the appropriate observance. Repaired the breech obstruction and reached sanctuary. A new coolant recoil disturbance, only approximately 6.5 seconds old and still inside its grace period, remained in the final read-only state.

Visual and movement checks

• Revised first-person loader: the new approach framed the adult crew member at the left and the loading cradle/breech to the right. It substantially improved the prior composition, where the central breech face hid the worker. Before, revised actual frame.
• Overhead: actual floor movement and station routing worked; the player moved to the right-hand aisle through floor clicks. The interior, lamps, crew and cannon remained legible while the surrounding scene was predominantly black. Actual overhead frame.
• Port aperture: floor clicks brought the player to approximately (3.2, -1.6), beside the starboard slit centred at (4.1, -1.5). Switched to first person and used actual pointer drags to centre the physical aperture. It is a small, deep slit in the armour rather than a panoramic window. However, the exterior remained black across approximately 35 live seconds of observation/rotation. Actual dark port frame. No enemy, civilian, rubble or terrain glimpse was positively observed. This remains a visual QA gap even though exterior masking has separate geometry tests.
• Read-only runtime sampling was generally approximately 59–60 FPS; the final sample reported a 720 × 291 internal render, 111 draw calls and 0.8 ms reported render submission. These are sampled debug counters, not a hardware performance benchmark.

Issues and follow-up

1. No proved outdoor glimpse. The desired predominant darkness is visible, and the physical slit is present, but recognizable passing fragments were not witnessed in this playthrough. Diagnose alignment, aperture transmission, actor-height/placement and brightness before claiming this part of the user experience is verified.

2. Sound could not be verified in this browser. Repeated trusted clicks on the visible Sound off button left it off; no toast, warning or error was captured. The read-only browser scope returned undefined for window.AudioContext and window.webkitAudioContext, but that scope may restrict globals, so this is not proof of the browser's actual AudioContext support. No sound was listened to or creatively approved here. Root was informed to add visible feedback for a suspended or unavailable sound system.

3. Stale recoil button. Immediately after expedient breech repair had resolved its shock, its panel still displayed Brace and reseat the recoil witness while the live omen list contained only a coolant shock. Include unresolved local shock state in the panel's render signature so the old button disappears immediately.

4. Mechanical diagnosis labelled as intrusion. A discovered ordinary dry-bearing fault produced the readout Intrusion evidence recorded. Mechanical and shock signs should be labelled as fault/witness evidence; reserve intrusion wording for confirmed chaos. This matters to the causal distinction the religious system is meant to teach.

5. Balance remains forgiving. The long unattended first-crossing interval produced extreme cooling distress and six faults without killing the crew or relic. Efficient intervention works, and late low motive health visibly slowed progress, but the severe exponential pressure envisioned in the brief is not yet demonstrated by this candidate. Tune after more uninterrupted playtests rather than treating this successful recovery as final difficulty approval.

6. Crew and rite interaction remain a foundation. Adult crew silhouettes and working poses are clearer, but the close-up art still reads as blocky geometry. Rites currently challenge condition selection and ordered physical response. Pressure-band manipulation, contact-based unction and active causal witness challenges remain proposed next interactions in RITUAL_AND_SPIRIT_DESIGN.md; they are not claimed as implemented here.

The campaign is playable end to end through the actual interface. Creative art acceptance, audible feedback acceptance, and positive outdoor-fragment proof remain separate from that result.

Follow-up regression: sound controls

Opened a separate local browser tab, preserving the victory tab, and began a fresh protected training watch. A header sound click while the briefing overlay covered the game had no visible effect. After Begin the first watch dismissed that overlay, the same actual Sound off control immediately became Sound on, with its pressed state true. The new source also provides an activation timeout and visible failure feedback for genuinely blocked Web Audio.

Opened Spirit's language, exercised all six visible Listen to the cadence buttons, returned to the watch, and muted with Sound on. The control visibly returned to Sound off. No error or warning was captured while exercising the previews. This resolves the previously unverified control activation in this browser; it does not substitute for a human listening review of the audio itself. The initial blocked header click is explained by the modal overlay, rather than treated as proof of unavailable AudioContext support.

Follow-up campaign: balance, ports and save orientation

Completed a second, separate three-crossing campaign through the actual interface. It began from the new watch opened for the sound regression above; the previous victory tab was preserved. Loaded the updated balance during the protected tutorial, then continued that same campaign through the later port and camera fixes using normal Save and Continue saved watch controls. This was an iterative local candidate check, not a claim that one untouched installed binary was played from start to finish.

The second final screen displayed Hull 100 · Spirit 99 · 53 shells fired · 3 crew replaced. The read-only state recorded phase: "victory", leg 3, spirit 98.9816, no remaining omens, four oil, three incense, eight parts and four seals. Immediately before the last halt, hull was 88.7; the final halt patched it to 100. Explicitly clicked Submit final report, which produced SANCTUARY REACHED and Campaign complete. No defeat occurred in this follow-up. Actual follow-up victory, final read-only state.

Real port and camera evidence

Walked via an actual floor click to (3.2, -1.605), next to the small starboard slit at (4.1, -1.5). Entered first person and turned by dragging the mouse until the player's actual forward direction was approximately (1, 0), directly toward the slit. The initial near-black view was investigated in a separate diagnostic tab; no diagnostic controls or altered masks were used for the evidence frames below.

After the port readability fix, the normal first-person view showed a rough outside surface through the narrow physical aperture, with black above and around it. Reloaded the final irregular texture and captured the view again. Final normal first-person port.

Switched back to overhead without turning away from the port. During ordinary forward travel, a portion of a passing green armoured surface became visible to the right of the tank at road-cycle approximately 42.6. The surrounding exterior remained black. This is positive rendered gameplay evidence of an outside fragment through the eye-aligned port mask; it does not prove that every proposed actor category, such as civilians or gore, has been visually reviewed. Normal overhead passing fragment.

The first-person save bug found during this check initially restored the direction toward negative Z rather than toward the port. After the fix, used Continue → first person → a small real vertical mouse drag → Pause → Save → reload → Continue. It restored first-person mode, the same port framing and forward direction (0.999999, 0.001204) at the same player position. Actual restored first-person view. The earlier negative-Z restore is resolved by this later actual save-and-load check.

The final loader approach also remained usable in the updated watch: the adult crew member, loading posture and rear cradle were visible beside the breech. Inspected the certified Loader Ilya and received No active intrusion detected at loader. Final actual loader frame. This verifies framing and interaction, while close-up character art still needs the user's visual acceptance.

Maintenance and religious decisions in the follow-up

• Completed protected training, corrected Holt's initial lapse, identified and purged vox corruption, and corrected the original gunner. At the first halt, replaced all three original crew with Vale, Korda and Ilya, then provisioned and restored the save. The crew replacement counter and supplies survived.
• A local recoil response disappeared after its 1.4-second attention finished; the remaining mechanical fault was labelled Fault witness recorded. A confirmed vox or shrine corruption was labelled Intrusion evidence recorded. These are the resolved stale-control and incorrect-diagnosis-wording checks from the original issue list.
• Port diagnosis consumed live crossing time. By late watch two, motive power fell into the fifties while the existing self, shrine recoil, coolant recoil and engine bearing faults aged. Targeted vox purging and the dry-engine rite recovered enough motive function to reach the next halt at approximately 94 hull. This was deliberately recorded rather than erased by a reset.
• Entered the final crossing with those carried faults. A cool-breech observance required lock → anoint teeth → witnessed dry cycle. The compromised shrine required break tether → compare mechanical witness → seal foreign response, clearing both the personal intrusion and the shrine's old recoil disturbance.
• One real mistake was made: walking away immediately after choosing the last shrine act interrupted the still-running response and consumed its prepared reagents. Returned and repeated the rite, then waited for its completion before moving. The interface displayed the interruption and preserved the consequence.
• The engine later described trapped hot air, a long resting gap and oil that would cook, even with a 68-degree instrument reading. Followed its actual witness rather than assuming a fixed thermometer-only rule: bank demand → vent casing → slow cadence. It restored the engine and its new local recoil witness.
• Reseated the old coolant recoil witness, then performed a preventive coolant observance. Its observation described lost prime and said bleeding would expel its remaining reserve. Correctly chose close → prime silent return pump → unseal, a distinct branch from the earlier trapped-pressure bleed.
• At 56% of the last crossing, the live omen list was empty, hull approximately 91.9 and spirit approximately 100. Later cleared a fresh hot-cartridge hang by lock → vent hot chamber → dry cycle, reseated a new shrine recoil witness without reagents, then identified and purged a separate inward-peeling shrine seal after it appeared. Sound was enabled again successfully during this final advance.

The updated pressure curve produced meaningful attrition and falling motive power during prolonged neglect, while competent intervention remained able to recover. This follow-up does not independently reproduce the automated 116-second worst-case balance comparison in BALANCE_V4.md, and a successful operator run does not settle the user's final difficulty preference. There are now two complete actual-interface victories, positive normal port evidence in both views, and a passed real camera save/restore check. Human listening review, final art approval and further depth of the physical rite interactions remain open creative review items.