Iron Litany — verification record

Current v0.4.0 fitted-office review

Physical offices records the experiments, constructed tank changes, native aim capture, moving introduction, longer slit views and shared battlefield sound/geometry. Actual-input review separates completed station exercises, campaign combat, failure recovery and saved-watch restoration. 151 automated checks pass. The runtime receives separate native-platform and publication checks before delivery; their receipts follow in RELEASE_V040.md without changing the frozen runtime.

All six new offices and the motive and measured-pressure controls were completed through browser input. Q interrupted a focused fitted instrument and entered captured first person; Resume obtained native capture and Escape released it with the watch paused. The practice return restored the campaign at approximately 73.93 percent progress with its original supplies. Art, sustained listening, first-time human usability and a full new-build three-watch victory remain open. Cinematic work is still gated by demo approval.

The v0.3.x sections below are historical evidence, not current-build acceptance.

Current v0.3.2 interface check

UX review records reproduced ambiguity, the implemented guidance/interaction revision, an ordinary-input watch reaching the first halt, two kills and one bound wound, saved partial-work restoration and responsive layout checks. Published release evidence verifies both desktop bundles, public readback and ordinary controls on the final live build. All 134 game tests pass. Historical results below are not new-build acceptance.

Current v0.3.1 crew and cannon check

Published release evidence is in RELEASE_V031.md. Both native bundles match all 148 frozen game files. Actual immutable-origin play passed the held-tool lesson, new approach and loading-stage contact checks; canonical scripts/runtime and public download bytes were verified. The creator's canonical saved watch was preserved at the updated welcome screen.

CREW_AND_CONTACT_V031.md records actual input play through the first halt, all three crew replacements and restored save, followed by the final breech approach in Watch 2. Five shell stages were captured during real automatic shots; recorded crew overreach was zero. Three small bodies were killed, one wound bound and a six-round reload spent six reserve rounds. Cooling and breech recoil were reseated; two emergency coolant repairs cost two parts and twelve spirit after deliberate maintenance neglect. The full suite has 126 passing tests. No full three-watch victory, final art, desktop gameplay or human listening acceptance is claimed.

Snapshot: 5 October 2026 · Foundation 0.1 · Updated after the final browser walkthrough and automated rerun.

This record distinguishes automated simulation checks, live control checks, art acceptance and listening approval. It is not a completed-demo certificate.

Current v0.3.0 verification

124 tests pass for the current narrow-hull, weighted-bearing and physical-incursion revision. The suite includes every ordered station route through actual doorways, local elevation/save migration, both distinct slit visibility contracts, paid interrupted binding, finite ammunition, projectile obstruction, growing creatures and a minimum recovery window after a kill. The final local browser's bounded console check captured no error or warning.

PLAYTHROUGH_V7.md records two ordinary-input runs. An intentionally extended halted run ended in ward loss and exposed an immediate-rebirth defect, which was fixed. A recovery-focused watch reached its first repair halt at 2:58, sealed three physical sources, killed two intruders, completed the loaded-bearing rite with zero wear, replaced all three crew and restored that halt from a save. It continued into Watch 2 and inspected a physical first-person slit. A separate bearing save/reload retained its exact oil film and preparation. No full v0.3.0 human-control campaign victory is claimed.

The eight synthetic experiments in watch-experiments-v030.json all reached victory, with three replacements. Prompt breach attention used 19–21 personal shots against maintenance-first's 47–54. Their ideal aim, omitted lesson and legacy maintenance oracle are explicitly limited in INTERIOR_AND_INCURSIONS_V030.md; they do not establish perceptual difficulty or creative acceptance.

Desktop packages receive separate native startup/GPU/offline-page/save checks before publication. Their release receipts identify the exact frozen source and are separate from these actual player-control runs. Authored crew/creature art, complete new-campaign pacing, touch journey and sustained human listening remain open. Cinematic and narrated gameplay production awaits creator approval.

Release verification is now complete in RELEASE_V030.md: both native packages and six public assets are verified, the full Kadabra package and owner-private mirror are published, and a real post-publication watch exercised combat, reloading and continuous source binding. This later evidence commit does not change the frozen runtime/dossier bytes shipped in v0.3.0.

Everything below is the historical verification record. Its earlier versions and test counts are not current-build claims.

Automated baseline

All 25 simulation/navigation tests pass against the final UI and exterior edits. A second reviewer also repeated the tests and syntax checks with no new blocker found. The final commit contains this report and the exact checked source.

Reproduction entry: npm test from the repository root. The recorded suite is tests/simulation.test.mjs plus tests/navigation.test.mjs.

The competent policy wins the three crossings on seeds 1, 41, 1984 and 4096 with three replacements and three depot visits; assertions require hull above 40, corruption below 35, more than 35 shots and bounded supplies.

The checked neglect regression reaches defeat on default seed 41, even while continuing through depots. No four-seed neglect claim is established by that test alone.

The competent policy uses simulated commands with a three-second attention allowance. It does not walk the actual browser character, read partial clues or prove comfortable human pacing.

Live Chrome walkthrough — demonstrated outcomes

The following observations came from ordinary visible browser controls, not direct simulation-state mutation.

The active game was then left unattended during tooling. It reached a Chaos defeat at displayed time 4:27, with hull 89 and spirit 64. The unresolved intrusion remained deadly despite substantial hull and spirit values. This records an observed neglect outcome, not a completed three-crossing victory through human-paced browser play.

Protected instruction intentionally shields the first wrong-order attempt. The separate simulation wrong-liturgy consequence test concerns unprotected play; the observations must not be conflated.

Camera/control functionality was demonstrated, but final visual quality in either camera remains unapproved.

Runtime audio

Three local AudioX candidates pass audio_info.py: stereo 48 kHz PCM-24 WAV, peak approximately −3 dBFS, zero clipped samples and no malformed/silent output.

Engine/ventilation: 10 s, RMS −14.81 dBFS. Breech/recoil: 5 s, RMS −14.60 dBFS. Ritual iron: 8 s, RMS −22.06 dBFS.

Exact prompts/seeds are in generation-manifest.json; measures/hashes in audio-qc.json; generation limits in provenance.json.

WAV total is 6,624,132 bytes. Generation used existing local AudioX offline; no models/environments were downloaded or restored. Available disk after generation was 104,528,859,136 bytes, a historical measurement rather than a current guarantee.

Human listening and creative approval remain pending. Seamless ambience loop boundaries were not authored; runtime fades/crossfades and sustained listening require verification.

Known production gaps and release gates

• Runtime art and gesture animation are procedural foundations, not the approved Diablo I target. Authored priest/loader/engine assets and physical loading contact remain outstanding.
• Three simulated crossings and depot interfaces do not constitute the complete intended campaign. Walkable refuges, religious gatherings, long space journeys, leaders/teachers/peers and broader crew selection remain unimplemented.
• Exterior presentation remains a foundation; authored pieces, aperture grime/shutter treatment, bounded pool evidence and coherent cross-aperture observation need integrated acceptance after the latest edits.
• Current simplified meters are game variables, not a validated physical tank simulation. Deeper traction, fluid, thermal and load behavior remains planned.
• An available page, cassette or concept image is not evidence of demo completion, live publication, final runtime art or a finished film.
• The creator explicitly requires the demo to be done and approved before cinematic and narrated gameplay production. That approval has not been recorded here; both film tasks remain gated.

Append-only live evidence area

Append dated observations here: build/commit, browser/session, starting state/seed, actual actions, observed result, evidence path and remaining failure. Record the pending final rerun, successful purge and runtime sound results separately. Amend an earlier claim explicitly if a later check disproves it.

Do not mark a milestone complete from test counts, screenshots or code inspection alone; apply the independent acceptance gates in ART_DIRECTION.md and DESIGN_FOUNDATION.md.

Final browser pass, 5 October 2026

A fresh seed-41 watch completed the opening rite, corrected the loader, and reached the vox through ordinary controls. Inspection at displayed 2:07 confirmed the intrusion. Targeted purge at 2:11 reported the vector sealed: unresolved omens fell from two to one; incense and seals each fell from six to five, while parts remained six. The remaining omen was the independent coolant fault.

Sound on and Sound off both changed the visible control state. Browser resource records confirmed delivery of all three WAV files (2,880,344 / 1,440,344 / 2,304,344 transfer bytes including HTTP overhead). This proves loading and controls, not human listening quality.

Actual screenshots are saved in docs/screenshots/overhead-watch.jpg, first-person-vox.jpg and repair-halt.jpg. The last watch was left paused with sound off. Browser console history contained the earlier, repaired shutter-initialization exception; no later exception was observed after the corrected reload.

The bounded renderer reported 230 meshes after batching, four point lights, 71 draw calls, 9,202 triangles, 11 textures and a 720×285 internal framebuffer at the inspected ultrawide view. A sampled render submission took 0.8 ms while automation-visible frame cadence varied; these counters are diagnostic evidence, not a smooth-frame-rate certificate. System and destination resolve to the same filesystem, with 104,480,210,944 available bytes at the final storage check.

Production exemplar pass — 5 October 2026

The user continued the same project under Sol High. This pass added a generated target plate from the actual overhead screenshot, then separately built code-authored priest, loader, engine and breech details in the shared scene. The target plate is a reference image, not a gameplay capture; its prompt, provenance and hash are in art-exemplar-v2-manifest.json.

The camera now frames the tank more closely (orthographic half-height 7.4 → 6.2), with a small exposure and ambient lift. Actual runtime evidence is saved at screenshots/overhead-exemplar-v2.jpg and screenshots/loader-first-person-v2.jpg. The latter shows a readable loader and breech with clear first-person access, but the face, hands and surface work still have simple polygonal form. The art gate remains open.

Live first-person inspection found that the old crew approach positions overlapped the loader mesh. The new loader, gunner and commander standing points remain within action reach and pass the same collision/path tests. Reloading an existing save made at an old overlapping crew position relocated the priest to the new loader standing point; the inspected state reported player (-1.65, -1.1), first-person view, Watch 2 and all three replacements intact. No save schema change was needed.

The visual loading cycle now derives from actual cannon shot time: recoil, acquire, lift, align, seat, clear and ready. It cannot grant or alter shots. One new test verifies its timing, gate closure and quarantine suppression. All 26 automated tests pass, and the local source/reference check passes. A short full-speed motion capture, shell-to-hand contact inspection and approval across both views remain pending. The sampled browser reported 242 meshes, 102 draw calls, 11,306 triangles and a 720×285 internal frame before the final material tweak; active frame cadence remains unapproved.

The browser playtest reached the first repair halt again, replaced all three crew and displayed Loader Ilya in Watch 2. That watch was allowed to run unattended while first-person evidence was gathered; unresolved faults ended it during the second crossing. A new protected first watch was then started for the saved exemplar screenshots. This is a test save, not a completed human victory.

Loader reach and framing pass — 5 October 2026

Live Chrome play at the loader station reproduced a visible flaw in the prior candidate: the stationary loader's arm stretched across the breech as the shot-linked shell moved to its seat. The worker now shifts his feet through rack, grip, lift, align and return positions. A focused deterministic test samples both hand targets through the transfer and bounds shoulder-to-target distance below 1.25 metres; the shot, ammunition and firing interval remain controlled by the simulation, not by this visual sequence. The ready breech still closes before the next simulated shot at the current minimum firing interval.

The station approach moved to the side aisle at (-2.45, -1.15). Selecting the loader in first person now frames the worker and gun together. The actual 2560×1217 Chrome frame at the 720×285 internal resolution is loader-transfer-v3.png: Watch 1, first-person align stage, shot 17, game time 177.15. It proves the revised position and camera presentation were exercised in live play. The shell remains partly occluded and the block-built face, clothing and hands do not pass the art or contact-motion gates. Browser cadence during this run was variable; this capture is not a performance certificate.

One unattended observation reached repair halt 1 with the original crew, replaced Holt and Venn, and entered Watch 2; accumulated intrusions then caused defeat while gathering screenshots. This is a genuine loss with neglected maintenance, not evidence of final difficulty balance or a live campaign victory. A fresh first watch produced the saved frame above. No cinematic or narrated gameplay approval has been given.

Rites, sightlines and crew pass — v0.2, 5 October 2026

At the initial v0.2 source snapshot, all 47 automated checks passed: seeded campaign outcomes, proximity and migration, adaptive witnesses, recoil faults, eight slit tunnels, real adult arm reach, geometry budgets and secure offline paths. Source/module/local-link check passes. These checks are supplemented by an actual Chrome playthrough through the protected engine rite, cannon fire, a coolant recoil reseat, a deliberate wrong physical response (penalty without advancing), three first-halt crew replacements, requisition, save/reload restoration and the second crossing's condition-led coolant rite. The complete live campaign outcome is recorded separately once the ongoing QA watch ends; no automated victory is represented as human play.

Exterior runtime was inspected in overhead view: surrounding scenery is black outside the player's actual slit sightlines, with no panoramic exterior backdrop. The shared physical mask also applies in first person, using player eye position and facing. Shader console errors were absent in the inspected Chrome run.

Crew geometry is replaced with original adult profiles, face planes, folded uniforms/cowl, hands and role equipment. The shell handling route is extended to 7.1 seconds and works around the cradle's rear. A test samples the actual 0.729m arm rig through the transfer, requiring less than 0.035m contact error instead of accepting the former 1.25m arm reach. Final motion, visual composition and subjective art approval remain separate from that geometric check.

Four new original AudioX material samples pass file QC (48kHz stereo PCM24, -3dBFS peaks, no clipping). Cannon concussion layers impact and loose steel; the six repeatable mechanical cadences, directional station cues, captions and listening page are connected to deterministic spirit memory. Human listening approval remains pending.

Offline Linux and Windows packages are configured with native-runner smoke checks, a stable local save origin, a sandboxed renderer and bundled cassette/manual. A successful native workflow is required before publishing download availability. Cinematic and narrated gameplay remain gated on demo approval.

Campaign completion and revised candidate — v0.2.1

The actual interface campaign in PLAYTHROUGH_V4.md reached sanctuary after all three crossings, 54 cannon responses and three replacements. The final halt patched hull to 100, with spirit rounded to 94. No state was injected or time accelerated; pauses were used during evidence collection. The first candidate was too forgiving during a documented unattended interval, which led to the calibrated revision in BALANCE_V4.md. That replay now retains structural loss after late recovery, and gives coolant-first triage a measurable advantage. It is numerical evidence, not final human difficulty approval.

The misleading mechanical-intrusion readout and stale recoil button found during play are corrected. First-person saves now preserve actual look direction and pitch instead of the last walking direction. Sound controls share bounded activation feedback. A fresh unobstructed game start enabled sound and its learning controls successfully; the earlier apparent failure was the briefing overlay blocking the header. Human listening approval remains pending.

The small physical apertures remain unchanged. Their steel shutters now match the masked opening. A real-GPU probe using the production shader recorded 640 colored pixels out of 4,096 through the aligned slit and zero for closed, facing-away and obstructed cases. Changing eye position removed the original centre view. This probe is required by both native smoke runs, supplementing the pure geometric checks. It does not approve visual art.

A normal first-person frame now positively records rough shared terrain through the slit: actual exterior fragment. Original GPU uniform values were correct; overly dark flat exterior materials made the visible region appear empty. Shared bounded terrain and more readable rough materials address that defect. A narrow passing armoured fragment is also proved in overhead: actual overhead glimpse. A real Save/reload/Continue sequence preserved first-person +X look direction and vertical look. The revised campaign also reached sanctuary through actual controls and an explicit final-report submission: 53 shots, three replacements, hull 88.6 on approach and 100 after the final patch, spirit 99. Updated sanctuary screen. The complete follow-up records the deliberate mistakes, interruptions, recovery and planning pauses. The background remains black outside the eye-relative rays.

All 52 automated cases and the source/link check pass at this candidate. Both revised v0.2.1 native runners passed rendered startup, view switching, offline resources, persistent origin and the real-GPU visibility probe at source commit 506d8bde17a78cded37549a7c82d82666f624f6b. The archive and anonymous download receipts establish publication availability separately. Windows checkout text uses CRLF; its manifest hashes match the shipped bytes, and all text matches canonical source after line-ending normalization. Binary/audio bytes are unchanged. A new checkout policy pins future text to LF. Cinematic and narrated gameplay still await the user's approval of the completed demo.

Measured return and crew refinement — v0.2.2, 5 October 2026

The Office of the Measured Return is now a physical input against actual coolant pressure. Measured return QA records both an overbleed/recovery and a fresh zero-wear replay through ordinary controls, including a paid 30% opening preserved exactly on reload while paused. Real vent gain/rate follow opening/head; three return knocks follow simulation time and do not replay elapsed beats. Separate Chaos remains after physical service. No human listening approval is inferred from runtime routing evidence.

Actual first-person inspection exposed a crowded loader approach during transfer. The revised side aisle stays at least 1.25 m from the moving worker in the regression samples; all station routes pass. Actual replacement-loader frame and actual overhead frame document candidate 5. The face, workwear, surface response and hand/seat motion are clearer but remain visibly procedural. Diablo I art acceptance, driver/gunner contact and final sound remain open. No cinematic approval exists.

The first test run passed 63 cases; the final suite includes a 64th case for inspection clearance across the whole loading cycle and turret motion. The live second crossing was deliberately allowed to deteriorate during inspection and ended in a ward-loss defeat; it must not be reported as a v0.2.2 full campaign win. Existing v0.2.1 sanctuary runs and seeded campaign regressions remain separate evidence. Exterior eye-relative apertures were preserved.

Work, witness and overhead slits — v0.2.3, 6 October 2026

The V6 playthrough records actual held mechanical service, a miswired covenant and re-challenge, an early counterfeit response without lifted pins, and a weighted witnessed trace. Header Pause/save/reload retained a partial stylus position; explicit regripping completed the vox connection and left unrelated duties unresolved. A separate run replaced three crew at the first halt and lost the second crossing to neglected corruption. No human full-campaign victory or final difficulty acceptance is claimed.

Earlier capture statistics failed to predict actual overhead legibility. The final bounded eye-image presentation extends past the moving tracks: normal-size moving slit frame. Turning away removes the original left-hand view. The same source scenery is visible through the physical tunnel in first person, with overhead images disabled. The image remains a small candidate fragment requiring further art composition; shared geometry, darkness and exact eye admission are preserved.

The integrated suite includes contact non-accumulation, bounded wrong-work recovery, no instant covenant purge or self-cleansing bypass, no shrine/vox inspection oracle, paid/save-restored work, positive incident identity, later intrusion preservation, valid physical trace ledgers, keyboard pause-focus handling and four-seed campaign regressions. Real play remains separate from these checks. Art, hand/tool contact, crew motion, human listening and demo approval remain open.