Care in every
production step.
The practical workflows that turn a source into a readable, measurable and reviewable lesson.
Source mapping
Extract source material while preserving page, slide, paragraph and CSV row references.
Reject scanned pages with no selectable text rather than substituting a guessed transcript. Preserve the original bytes, SHA-256 and extraction segments. CSV content mapping must be explicit; roster columns never become training text. Normalize whitespace only when checking evidence, retain the original quoted values.
Download skill ↓Evidence binding
Attach exact source evidence to narration, facts and knowledge checks.
Verify every quotation against its numbered source segment. A numeric value must occur in its quotation and retain its unit. Distinguish actual, proposed, forecast and model figures in both narration and visible labels. Reject unsupported claims before paid narration.
Download skill ↓Teaching design
Convert report scope into observable learning objectives and explanatory chapters.
Build a section inventory before writing. Give each chapter an observable objective, a causal explanation, a takeaway and a misconception check. At two minutes, teach the main relationships and link detailed tables to the source. Ask for a shorter scope or more material when the source cannot support the requested lesson; do not pad.
Download skill ↓Scene direction
Choose fixed 3D scene templates to explain each narration beat.
Select geometry according to the relationship: comparison for two compatible values, chart for labeled series, document flow for clauses, globe for regional context. Preserve citations, speech and IDs when changing visual treatment. Use only fields allowed by motion.schema.json; never execute generated JavaScript or shell commands.
Download skill ↓Narration timing
Fit complete measured male narration to the approved course timeline.
Keep one Charon voice. Generate manageable narration sections, cache it by text, voice and model. Measure WAV duration before assigning frames. Leave reading holds, keep scenes 6–30 seconds, use 24 fps boundaries, and reject speech that cannot fit. Never speed up or cut training content to meet duration.
Download skill ↓Data illustration
Build readable, source-bound 3D numbers and charts.
Bind labels and chart geometry to the same fact record. Comparisons require matching units and comparable periods; mark forecasts explicitly. Bar charts include zero, preserve sign, use a nonzero range and never clip values. Display source precision rather than invented rounded labels. Keep visible fact counts within preset limits.
Download skill ↓Motion continuity
Maintain camera, transitions, palette and reading rhythm across scenes.
Use absolute-time transforms so seeking backwards produces the same frame. Limit camera choices to locked, push_in, pull_out and lateral. Use short entrances with long readable holds; match depth between related concepts. Dispose previous geometry and CSS3D panels. Reduced-motion web previews use a still reading frame.
Download skill ↓Caption alignment
Create synchronized captions using measured speech and exact approved wording.
Align to acoustic speech, not character count. Check word confidence and missing tokens; preserve uncertainty for review. Caption intervals must be ordered, positive and inside the complete audio, allowing only the documented 20ms PCM boundary tolerance. Normalize known acronyms for independent ASR comparison, never use the target script as its prompt.
Download skill ↓Release control
Freeze approved sources, storyboards and finished assets into an immutable course release.
Require a matching storyboard digest before rendering and finished-media approval before publication. Store asset hashes and source lineage in each release. Later draft changes cannot replace an assigned release. Authorize asset access by workspace membership or assignment; issue short-lived URLs. A playback resume position does not prove watched time.
Download skill ↓Production verification
Verify rendered media, recovery and costs before course delivery.
Probe resolution, frame rate, exact duration and audio. Decode the whole file and compare independent speech with approved meaning. Inspect representative scene frames for clipping and labels. Reserve costs before submission; quarantine lost acknowledgments and reuse unchanged assets. Retry durable handoffs without duplicating paid work. Do not publish while source hashes or approvals differ.
Download skill ↓