Nota, for judges
Nota enters all three tracks. Each section below stands on its own: what to open, what it proves, and the rubric line it answers. Every link goes to live data, never to a mock-up.
Every track (common 40)
| Rubric | Open this | What it shows |
|---|---|---|
| Combining sources · 20 | receipt 0d9469f115d1, Evidence provenance | Five RYO tools plus seven independent checks (exchange prices, recomputed RSI/ATR, OKX positioning, DefiLlama, prediction markets, narrative voices) in one evidence pack, argued over by three agents and a judge. |
| Coping with failure · 20 | 2ae531ec2c9b (RYO answered 401) and
c60acf3ba3b1 (RYO deep_analysis down for three days) |
The run still completes, records which section failed and why, and refuses to trade. A missing value stays
null; nothing is filled in. The ledger kept locking and settling through both outages. |
Track 1 · Autonomous Agents
- Run it yourself. In the dashboard, "Run the council live" gathers fresh RYO evidence and convenes the council on any symbol (one to two minutes, capped per visitor).
- Real thinking (25). Open any receipt: each agent cites dotted paths into RYO's answer, and the value shown is read back out of the evidence, not retyped by the model. Citations to paths that do not exist are dropped in code.
- Cause and effect (20). "What changed" ranks every evidence move since the previous receipt by impact; the derivatives gate withholds RYO fields that are not about the token (see a receipt's Warnings).
- Repeatability (10). Press Verify replay: the decision is rebuilt from the stored evidence and prints
identical: true. Offline:uv run nota replay <id>. Each receipt is also anchored in Bitcoin (OpenTimestamps). - Advanced (5). Practice trades sized from ATR with stops and targets, a twice-daily ledger cycle in CI, and every agent scored by Brier against the token's own base rate.
Track 2 · Dashboards & Interfaces
- Easy to read (25). Open the dashboard: each receipt starts with a five-line summary (now, biggest change, position, what RYO said, next step).
- Sorted by importance (25). "Needs attention" sits on top; "What changed" is sorted by impact; the scorecard puts the next settlements first.
- Works for everyone (10). Keyboard only: j/k, Enter, p, /, ?. Skip links, checked contrast, reduced motion, light and dark, and a Japanese overview.
Track 3 · New Skills
- Fills a gap (25).
positioning_check: RYO reports open interest and funding, but some of those fields carry the same value across unrelated tokens. The skill checks RYO's derivatives against OKX for the token and withholds what is not token-specific. - Follows the spec (15). Every skill returns RYO's envelope (
status,data_mode,availability,warnings) at RYO's own path shape:GET /api/skills/,POST /api/skills/{name}/invoke. Try one in the dashboard's "Run a skill". - Could we use it (20).
contrib/positioning_check/in the repository is a drop-in: no Nota imports, a FastAPI router RYO can include, and its own tests. The same skills are an MCP server at/mcp.
What only Nota does
- It audits RYO itself. The Verdict Scorecard locks RYO's own verdict and trade plan for 25 majors every day and settles each one on OKX candles at 24 and 72 hours, every miss counted.
- It found a defect in RYO's data: one open-interest figure repeated across unrelated tokens, now withheld by a gate.
- Every lock and receipt is anchored in Bitcoin and every pushed ledger snapshot is attested with Sigstore.
Limits are listed plainly in LIMITATIONS.md.