# Source Map — The Careful Bitcoin Prediction Lab

| Source | Why it is used | Claim boundary |
|---|---|---|
| `docs/architecture/btc-polymarket-5m-prediction-architecture.html` | Proposed system flow and separation of data, ML, and deterministic gates | Design target |
| `docs/planning/btc-polymarket-5m-15m-research-backtest-replay-prompt.md` | Data contracts, model ladder, cost-aware replay, and no-live-trading rule | Design target |
| `docs/planning/btc-polymarket-5m-15m-research-backtest-replay-redteam.md` | Risk and red-team constraints | Design target |
| [Referenced Polymarket BTC Up/Down market](https://polymarket.com/event/btc-updown-5m-1785678300) | Settlement wording: Up when end Chainlink value is greater than or equal to start value | Source-specific; re-check for every live market |
| [Polymarket API documentation](https://docs.polymarket.com/api-reference/introduction) | Public market metadata and CLOB data interfaces | External documentation; verify version before implementation |
| [Polymarket market WebSocket documentation](https://docs.polymarket.com/market-data/websocket/market-channel) | Order-book, trade, and price update concepts | External documentation; verify version before implementation |
| [CatBoost paper](https://proceedings.neurips.cc/paper_files/paper/2018/hash/14491b756b3a51daac41c24863285549-Abstract.html) | Ordered boosting and categorical-feature treatment | Literature-derived |
| [DeepLOB](https://arxiv.org/abs/1808.03668), [TLOB](https://arxiv.org/abs/2502.15757), [LiT](https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1616485/full) | Order-book representation and directional-model concepts | Literature-derived; no BTC production claim |
| [SimLOB](https://arxiv.org/abs/2406.19396) | Self-supervised order-book representation-learning concept | Literature-derived; no BTC production claim |

## Claims that must remain design targets

- That historical Chainlink settlement values and historical Polymarket level-2 books will be available with sufficient quality.
- That CatBoost, the encoder, or the directional head beats Polymarket’s implied probability after costs.
- That any model finds a tradable edge.
- That a replay remains valid in live execution.
- That a risk rule prevents all losses.