mass-sim-liturgy

Decisions and provenance

Recorded from the user’s requests in the development session on 14 October

1. These are product decisions, not ecclesiastical approvals.

Request / decision	Source	Implemented surface
Animate the celebration mainly from Roman Missal rubrics	Initial user request	mass-rite.js, original floor-plan animation
Paraphrase rubrics; show an illustrative Sunday in Ordinary Time	Initial implementation choice	41 baseline director steps; no complete translated Missal
3D scene, selectable views, no-deacon/no-server scenarios	User, approximately 21:25 +02:00	mass-3d.js, mass-scenarios.js
0-5 servers, combining duties for smaller teams	User selected the recommended option after the 22:27 request	Team allocations in buildScenario
Altar reverences should face towards the altar; longer chasuble	User correction during server-count work	Altar-relative orientation and priest geometry
Camera follows wherever activity is happening	User request during that work	activityView, smoothed action camera
Server-training mode and editable church layout	User, 22:37	mass-features.js, follow-server and next-duty cues
Optional speech, distinct character voices, different languages	User, 22:39	mass-speech.js
English and Latin first; additional languages importable later	User’s explicit answer	English script excerpts, partial Latin pack, JSON imports
Static, pre-generated voices; congregation should sound like a group	User, 22:56	Audio-source selector, MP3 playback, layered congregation
Use AI/free or cheap TTS; one-off generation is acceptable	User’s explicit answer	Local Kokoro ONNX generator; no paid API calls or text uploads
Common layout presets	User during audio generation	Central, mirrored, compact, broad, axial sanctuary arrangements
Private roman-mass-explorer repository	User selected private repository in the follow-up	GitHub progress snapshot

Implementation choices and limits

• Browser-first, no build system: HTML, classic JavaScript, local vendor scripts and file-relative assets let both versions open without a web server.
• The original 2D animation remains available and shares canonical rite data with 3D. Scenario compilation clones baseline data before applying adaptations.
• Small server teams combine practical duties; incense and ringing the bell at the elevation are not assigned simultaneously to the same person. This is a teaching simplification, not a staffing rule from the Missal.
• Communion under both kinds uses a deacon in this simulation. This is an app constraint, not a claim that the rite always requires a deacon for that option.
• Furniture presets only change the sanctuary arrangement within a fixed church shell. They are not measured plans of real churches. Nearby destinations are deformed with the furniture; there is no collision-avoidance solver.
• The action camera prioritizes concrete gestures, then speakers/movements. Elevations and processions have explicit framing overrides.
• Speech waits for each cue and skips stage directions or absent translations. Imported language packs do not translate the rubric UI.
• Live browser speech uses only voices marked local. The static pack uses four character voices and a layered stereo mix for the congregation, with slight timing/speed differences and two quiet echoes.
• The first congregation clips mixed six voices. After listening, the user found that too many, so new generation defaults to four (two male, two female), configurable with --group-size. Regeneration is tracked in issue #1.
• Audio files were first named by a 20-character content hash. At the user’s request they now have readable names in Mass order, per language folder, with an 8-character fingerprint, plus a CSV index to support later translation.
• Kokoro has no native Latin model here. Latin phonemization with English-trained voices is explicitly marked experimental; pronunciation has not been certified.
• AI speech is synthesized, not a recording of clergy or a congregation. No person-specific voice cloning was requested or used.
• The generator normalizes ONNX input dtypes because kokoro-onnx 0.4.9 types speed as an integer for this export. Generation uses speed 1, then applies modest resampling locally. Four inference threads avoid CPU oversubscription.
• Temporary Python environments and AI model/voice downloads live outside the repository. Only scripts, dependency requirements, generated recordings, manifests and documentation belong in the progress snapshot.

Verification status

Automated checks cover staffing combinations, sampled positions/prop owners, layout transformations, training cues and language/speech scheduling. Browser checks exercised the original and 3D views, imported packs, saved layouts, camera targets, live speech and cancellation.

This does not constitute independent verification of every Missal paragraph, approval of the translations, human listening review of every generated clip, or collision-free validation of every editable arrangement.

Holy Mass simulator game (15 October 2026)

The user asked for a playable simulator built on this explorer, with decisions recorded for later review. See GAME-DECISIONS.md for every choice, the alternatives and the source.