# Brief: merge Paradigm's token generator visual with TensorCash's branding and idea

You have curl but no browser. Work in this directory. Deliver a self-contained HTML animation.

## 1. Research first, and save what you used
- https://www.paradigm.xyz — find the generative "token generator" visual on the site (the
  animated grid / ASCII / particle piece) and describe in your own words how it looks and
  behaves: shapes, motion, palette, typography. Follow at most 3 links on the same host and at
  most 3 script or stylesheet files, to see how the visual is made. Time-box the research to
  about 12 shell commands in total; what you cannot establish in that budget, note as unknown.
- https://tensorcash.org and its /docs/proof pages — the branding (colours, type, tone) and
  the idea: proof of inference as proof of work, draws tied to the latest block, 256-token
  windows, verification by replay.
- Save `research/notes.md`: every URL fetched, the date, what you observed about the Paradigm
  visual, and the TensorCash sentences you rely on, quoted. Do not invent facts about either
  site; if something is unreachable, say so.
- `receipt.json` is a real TensorCash inference receipt. Fields: `pow.proof.chosen_probs`,
  `pow.proof.sampling_u`, `pow.proof.chosen_tokens` (256 values each), `pow.proof.block_hash_hex`
  (byte-reversed vs explorers; explorer order 0xb967cbf2…6446, height 31536),
  `pow.proof.tick`, `pow.proof.hash_hex`, `pow.proof.model_identifier`, `token_counts`.
  Load it with json; do not cat it.

## 2. Build
- `animation.html`: one self-contained page (inline CSS, SVG or canvas, vanilla JS; no external
  fonts, scripts, images or network calls), 1200×675 stage, looping, with a Pause control.
  The visual language comes from the Paradigm generator as you observed it; the content,
  colours and words come from TensorCash. Show the 256 window steps being drawn from the chain
  (block hash, VDF tick), the chosen-token probabilities and draws from `receipt.json`, and the
  window becoming a proof. Respect `prefers-reduced-motion` with a static fallback.
- `build_animation.py`: Python 3, standard library only; reads `receipt.json`, writes
  `animation.html`. Run it and check the output parses with `html.parser`.
- `README.md`: what the piece shows, which Paradigm observations and which TensorCash sentences
  shaped it, and how to regenerate.

Write `research/notes.md` before you start building. Use only numbers from `receipt.json` and
facts from `research/notes.md`. When all three files
exist and the script runs clean, summarise what you built and stop.