Iron Litany — rites, recoil and mechanical speech

5 October 2026. Original game mechanisms informed by the religious maintenance, canticles, stubborn machine spirits and scrapcode sources catalogued in LORE_RESEARCH.md. These rite names, mechanical thresholds, sound grammar and gameplay rules are inventions for Iron Litany, not quotations or claims of a universal canonical procedure.

Seven mechanism directions

1. Condition-led observance, implemented. The priest reads an analog witness and chooses physical acts. A dry bearing and an overheated casing need different service; a pressurized return and a pump that has lost prime need different middle acts. The diagnosis text describes material conditions, the consequence of a bad intervention and an audible cadence. It does not print an ordered answer key. The manual must teach the relationship between isolation, service and witnessed restoration. Three choice stages share verbs across the vessel. The number of a choice is only an input shortcut; successful acts vary with the observed condition.

2. Recoil witness, implemented. Each sixth actual cannon response loosens one witness seal or valve detent. The blast and gauge rattle announce it. The priest must reach that station, brace the witness, re-seat it and return two taps for 1.4 seconds. The response is free of reagent cost. Eighteen seconds of grace allow travel and prioritization; after that, neglected recoil faults gradually cause wear. A full rite or rapid repair can also settle one, but costs more. A spiritual purge cannot tighten a physical seal.

3. Remembered covenant, implemented at a first level. The spirit records answered observances, affronts, trusted station responses and trust. Returning to an already familiar altar yields a small additional accord gain. Material-only repairs buy time but leave a debt, lower trust and elicit a broken-vigil cadence. Temperament shifts among watchful, strained, affronted, besieged and accordant. It communicates through machinery rather than human speech. This is a simple personality model; richer persistent preferences remain future work.

4. Witness challenge, implemented in choices; deepen next. A corrupted receiver or inner tether answers before its physical relay moves. The player can compare two independent channels, isolate the connected branch and restore only the witnessed response. Genuine mechanical distress has an analog counterpart. A mimic is not defeated by guessing which scary noise means Chaos. Next, let the player tap a chosen historical cadence and require the machine to answer from the correct physical part; counterfeit replies may arrive from the wrong location or before the causal act.

5. Measured relief handwheel, implemented in v0.2.2. The Office of the Measured Return begins only with a trapped head of 62 or above and spends one oil. Isolation interrupts cooling; the player sets a persistent relief opening against actual pressure, seals in the 38–48 band, holds a closed witness for 1.2 seconds and explicitly restores the return through three pipe knocks. Flow follows head and opening, while the live heat/motion model continues. Overbleeding admits air and wears the real sealing faces, visibly and to a bounded maximum; completion cannot erase that wear. This is not a universal upgrade: ordinary coolant service gives stronger repair and a fixed 22-point pressure drop in 4.4 seconds; measured relief can shed a severe head down to the band but requires longer regulation. It does not purge Chaos. Captions, keyboard range control and explicit shut controls preserve learning with sound off. Reload retains the paid action and opening, reopens its panel, and pauses until explicit Resume. Hand contact, tactile resistance and metered coolant inventory remain future work.

6. Unction stroke and contact, future physical input. The priest chooses the bearing that answers the grind, holds a brush/tool in contact through a partial turn and meters oil. Too little leaves the knock; too much cooks on a hot casing or contaminates a neighboring surface. The tool path, material response and mechanical sound give the result. Cloth, metal, oil and wax must retain the Diablo I material direction; a glowing rhythm ring would weaken the theme.

7. Shared isolation and crew witness, future planning input. An isolated motive branch could permit two nearby services during one interruption. The gunner can park the turret for a safe access window; the loader can act as a second witness after reliable correction. Each delegation has a certification prerequisite. Skipped safeguards must state the risk. Religious authority becomes a way of organizing labor, route and exposure, rather than a universal bonus for indiscriminate scolding. The existing swift correction, quarantine and common first-depot replacements remain intact.

The first playable observance

The existing complete by-book rites remain available for old saves and protected engine training. They cost the same prescribed reagents and take 7–8 seconds of explicit responses. The new condition-led observances take 4.4 seconds of performed acts, plus the player's reading/decision time. They require the priest within two metres of the station for preparation and each response. Walking cancels an unfinished observance through the existing movement integration. Its prepared reagents stay consumed, as with ordinary rites.

Full motive power is temporarily reduced after isolation is performed. A breech observance locks actual firing until completed or interrupted. A wrong available act does not advance the rite: it causes a small spirit/trust loss and, for a physical station, a modest heat/pressure consequence. It repeats the witness evidence so that recovery is possible. The snapshot of the witness at preparation determines the sequence; ordinary thermal drift cannot secretly change the required choice halfway through. A Chaos variant clears the witnessed station vector; the shrine variant can clear the priest's own unwitnessed response. It does not clear unrelated corruption.

Nonverbal vocabulary

These IDs and cadences are stable sound contracts. Layer real machinery; avoid recognizable human screams, words, musical success jingles or arbitrary random groans. Captions describe the sound and permit learning when sound is muted. A learner glossary may associate an observed pattern with its material need, but should not present a reliable oracle for every form of future corruption.

Distress repeats on a measured cadence of 14–19 simulation seconds; a settled vessel speaks less often, about every 27 seconds. Successful attention creates one answer and postpones the next unsolicited phrase. Pausing freezes the cadence with the rest of simulation. The highest current need chooses the phrase, which makes simultaneous faults a triage problem rather than an overwhelming pile of loops. This first version does not spatially synthesize sounds itself; the sound system must place the components at the reported target. A self-target echo should feel internal but must remain captioned.

Integration contract

simulation.js exports normalizeState, observanceChoices, SPIRIT_PATTERNS in addition to the established APIs. spirit.js holds the pure pattern and adaptive-plan logic.

• Normalize an old parsed campaign with normalizeState(state). It adds absent spiritEntity only, never replaces malformed present fields. Save validation accepts the truly absent extension for backward compatibility and checks a present extension. Existing in-progress by-book rites retain their labels and timers.
• command(state, 'observance-start', {station, position: {x,z}}) prepares the new local observance. Use this as the normal station interaction after training. Training still calls the old engine rite-start and rite-step.
• The action has kind: 'observance', station, rite, variant, observation, step, remaining, total and canonical steps. Display observation as witness evidence. Display choices returned by observanceChoices(state); each has id and label. Send the chosen ID with command(state, 'observance-choice', {choice, position: {x,z}}). Disable choices while remaining > 0. Do not use the internal required step ID as an answer hint. Number keys may choose the displayed alternative by its current position.
• A recoil incident uses the existing record shape with kind: 'shock', a system target, age and resolved flag. The usual station urgency can show it. command(state, 'shock-attend', {station, position: {x,z}}) starts kind: 'attention', one 1.4-second response already running. No number/choice confirmation is needed. Walking/cancel leaves the fault unresolved. Generic rite-step rejects both new action kinds, preventing a bypass through legacy numbered responses.
• state.spiritEntity holds {version, temperament, trust, answered, affronts, familiar, nextUtterance, serial, utterance, lastNeed}. familiar lists altars the spirit remembers as devotionally attended. The current utterance is {id, serial, time, target, caption}. The sound system should cue only when its serial changes, use SPIRIT_PATTERNS[id].components to map layered assets and show the descriptor caption. Never replay a stale phrase just because a muted page reloads.
• Actual cannon count/lastShotTime continue to be the only source for blast, shake and handling. Recoil faults are generated on every sixth actual shot, alternating coolant, breech, engine and shrine. They do not create cosmetic shots, consume extra ammo or invent enemy attacks.

Verification and remaining review

Focused simulation checks cover proximity before spending, hot/dry and pressure/slack choice differences, rejection of legacy bypass, frozen witness snapshots across drift, actual-shot-linked recoil attention, cancellation, physical/spiritual fault distinction, authentic versus premature patterns, and save migration/tamper rejection. The competent campaign test now handles physical recoil faults with their attention command instead of incorrectly trying to purge them. This is a necessary update to its triage strategy, not a relaxation of the win conditions. Four seeded full campaigns still require victory, three replacements, hull over 40 and corruption below 35.

Those automated campaigns approximate three seconds between independent interactions. They are not actual walking playthroughs or a sound listening/art approval. Root gameplay QA must inspect whether spatial travel, reading under pressure, competing faults, captions, volume and recoil shake are fair, legible and enjoyable. Thresholds, repeat cadence and sixth-shot frequency are initial tuning values. No cinematic or narrated gameplay approval is implied.

Measured relief feedback contract

The material vent loop follows the actual retained valve opening and pressure: gain follows opening, playback rate follows head, and direction follows the coolant outlet relative to the priest. One source serves a continuous vent; no source is created per gauge repaint. Shut, cancellation, pause and mute fade and stop it. Three return knocks are scheduled against simulation time, preserve pending beats across pause and skip elapsed beats on reload or late sound activation. The final accord remains a single authoritative completion response. The handwheel and small world gauge reflect those same simulation values. These are authored mechanical candidates, not human listening approval.