First public release: the working prototype with one recorded, verified sample.
Run without keys
git clone https://github.com/Argona7/stampede && cd stampede
uv sync && uv run stampede demo # downloads stampede-demo.sqlite.xz from this release, replays it on :8791
uv run stampede terminal --api-url http://127.0.0.1:8791Assets
stampede-demo.sqlite.xz— the recorded sample (PONS v2, 60 min, 2026-09-10 16:29:57–17:29:57 UTC): 146,329 trades, 33,743 wallets, 87,497 sequences of the 30-minute window (331 direct, 11,833 clean, 75,333 ambiguous — ambiguous rows are listed, never counted), wallet scores from the research backtest, the alert journal. 17 MB, unpacks to 139 MB.stampede demofetches it automatically;stampede export-demobuilds it from your own store.stampede-v5-autopilot.mp4— the 44-second demo: terminal intro (real Textual renders), then one continuous browser take of the replay at 20×: 3D network build-up, autopilot over the strongest observed routes with evidence cards, RADAR, FLOW. Edit notes:docs/DEMO-V5.md.
What is in the box
stampede terminal(Textual): feed of observed sequences, hot rotations, activity sparkline, Radar screen.- Web (Vite + React + three.js): RADAR (ranked coins with every score part visible), FLOW (ego network of one coin), MAP (2D/3D scene, evidence rows per route, autopilot).
- One FastAPI server with a shared replay clock;
fixture | replay | livemodes; live needs an Alchemy key. - Docs: algorithm, coverage, verified examples, radar score, research backtest, development reference.
What it is not
Not proof of money flow between coins, not attribution of addresses to people, not a signal, not a return. Radar thresholds come from a five-hour backtest of one launchpad and describe that sample only. Nothing is sent to the chain.
Tests: 48 pytest, 8 Playwright. Licence: not chosen yet (see README).