# Course corpus → Cosmos Luminosity platform blueprint

Use when Margarita sends a course folder containing many audio/video lessons, homework and supporting PDFs and asks to derive a business/platform system from the whole corpus.

## Intake and source preservation

1. If Margarita says more material is coming, collect it until her explicit `конец` or equivalent completion marker.
2. Preserve originals in one stable project directory with separate `downloads/`, `out/`, and `analysis/` folders.
3. Extract text from Google Docs/DOCX/PDF; if a PDF is image-only, run OCR rather than accepting an empty extraction.
4. Inventory every media file with `ffprobe`, recording duration, size and format.
5. Transcribe sequentially in a tracked background process; for a large CPU batch avoid simultaneous ASR jobs.
6. Completion requires process exit plus last-timestamp verification against every source duration. Write a manifest with duration, last timestamp, gap, file sizes and status.

## Status communication

When Margarita asks «сколько осталось?», inspect the live process and transcript timestamps. Report:

- fully completed files / total;
- currently active file and percentage;
- audio hours processed / total;
- estimated remaining ASR time;
- separate estimate for synthesis.

Do not infer progress from transcript file count alone because the active transcriber writes partial checkpoints.

## Analysis pattern

For a 20+ hour corpus, split synthesis into bounded thematic reports:

1. **Resource/author system** — source claims, repeated principles, exercises, algorithm, risks, founder operating model.
2. **Audience/journey** — confirmed observations vs hypotheses, JTBD segments, language/barriers, free value, transition triggers, testing plan.
3. **Business/platform** — value ladder, recurring revenue, content automation, founder/AI/assistant roles, metrics, MVP and bilingual architecture.

Require every report to distinguish:

- direct source claims;
- cross-source synthesis;
- proposed Cosmos Luminosity adaptation;
- decisions still requiring Margarita’s approval.

If a delegated analysis times out, inspect the requested output path first. A full or partial report may already exist. Continue only the missing report or suffix; never restart the entire corpus blindly.

## Final synthesis

Create one reader-friendly blueprint in the Portal rather than leaving three reports disconnected. It should include:

- core architecture and correctness criterion;
- Margarita’s protected creative role;
- what AI and assistants handle;
- scalable core vs living laboratory;
- confirmed audience evidence vs international hypotheses;
- Magic Cloud’s standalone free-value loop;
- one short public customer path;
- LumoSide recurring-value hypothesis;
- revenue engines and P&L separation;
- one-source-to-many-assets content pipeline;
- English-first / full-Russian localization architecture;
- MVP exclusions;
- staged interviews, paid pilot, concierge beta and development gates;
- explicit “what we cannot claim yet” section;
- the first 3–4 founder decisions, handled sequentially.

Link the blueprint from `content/marketing-content-method.md` and `content/index.md`. Keep it marked as a strategic draft until Margarita approves decisions one by one.

## Strategic standards captured from the Universe corpus

- Resource and pleasure choose direction; they do not replace action, consistency or economics.
- Worldview precedes product, but the product still needs a concrete user result.
- Build loyalty depth: awareness → trial → first purchase → repeated use → loyalty → evangelists.
- Scale the authorial principle, not one delivery format.
- The platform is successful when value and revenue grow while Margarita’s manual work per user falls.
- Default architecture: free standalone value → one first paid result → recurring membership → bounded deep routes → limited premium support.
- Build custom platform infrastructure only after real payment, completion, repeated use and renewal signals.
- English-first means one canonical content object with culturally adapted English and full Russian representations, not producing everything twice.

## Pitfalls

- Do not call international product–market fit proven from Russian client stories.
- Do not make the free product a disguised sales hook; it must create a complete micro-result.
- Do not expose a new user to the whole product constellation at once.
- Do not sell a membership as “access to a library”; define one repeated result and verify renewal.
- Do not let an AI symbol, energetic interpretation or quiz result pressure a purchase.
- Do not turn a course’s metaphysical or money claims into established causal facts.
- Do not start editing the live site from an unapproved strategic blueprint.