Nota

An AI trading opinion you can audit. A council of specialised agents debates live RYO market evidence and records the practice trade it would make as a replayable decision receipt: nota replay <id> rebuilds it from the stored evidence and prints identical: true, every cited number is read back out of the evidence rather than retyped by the model, and independent sources audit RYO's own price and indicators before anything is sized. Built for the RYO-CHAN Hackathon 2026.

Tracks entered: 1 Autonomous Agents (council, receipts, watch loop, a derivatives gate the council cannot argue past), 2 Dashboards (diff-first receipt dashboard, and the RYO Verdict Scorecard), 3 New Skills (narrative_convergence, news_verify, price_crosscheck, technicals_crosscheck, positioning_check, move_base_rate, verdict_track_record, liquidity_check, crowd_odds, all in RYO's envelope).

RYO Verdict Scorecard

Every deep_analysis answer carries a verdict and a trade plan (a stop 1.5 ATR away, a target at +1R). RYO never reports what became of either. nota lock stores the whole answer for 25 majors once a day (never twice within 20 h); nota settle later asks OKX's public candles which level was touched first within 24 h and 72 h: 1-minute candles from the lock to the first full hour, hourly after, and an hour that touched both levels is split on its own 1-minute candles. RYO does not grade itself: the bracket is re-anchored to OKX's price at the lock, and a lock where the two prices are more than 2% apart is kept as basis_mismatch and never settled. A window with an hour OKX never published is retried for a day, then kept as unsettleable. Verdict pairs, confluence states and RYO's momentum gate are each compared on the target-first rate and on the return at the horizon (plans that touched neither level included): within each lock day, combined with Mantel-Haenszel weights so a lopsided day cannot flip the sign, with a p-value that shuffles labels within clusters of overlapping windows. Nothing is called distinguishable before 10 shared days and p < 0.10; a simulated null in the tests checks that rate. Every failed lock is a counted row, one per token and day however often retried, and a dead key stops the run after one call with an aborted row. The page also reports how often each lane of RYO's answer came back available, per RYO's own words. /scorecard shows it, most urgent first; /api/scorecard?horizon=24|72 is the raw record and /api/scorecard.csv the same table as CSV (one row per lock and horizon, failures included). The page carries a schema.org Dataset description (JSON-LD) naming both downloads.

Anchored in Bitcoin, not in Nota's clock. "Locked before the move" is only worth something if nobody, Nota included, can backdate a lock. nota stamp submits the SHA-256 of every good lock row and every receipt, exactly as stored, to three public OpenTimestamps calendars and keeps the .ots proof; a later run swaps in the Bitcoin attestation once the calendar has one, after checking that block's merkle root against mempool.space. The scorecard and the dashboard say "Bitcoin block N" or "timestamp pending (submitted …)" from that stored status and nothing else. To check one yourself:

curl -o lock.json https://nota-ryo.vercel.app/api/scorecard/locks/<id>.json    # the row as stored
curl -o lock.json.ots https://nota-ryo.vercel.app/api/scorecard/locks/<id>.ots
ots verify lock.json.ots        # pip install opentimestamps-client; or drop both files on opentimestamps.org

Receipts work the same way with /r/<id>.json and /r/<id>.ots.

Day 0 (2026-09-18): 25 of 25 locked, 4 CONFIRMED and 21 MIXED, and 7 plans long while RYO's own verdict was cautious (XRP, DOGE, ADA, LINK, SUI, BCH, WIF).

What the council may not cite

deep_analysis.data.derivatives reported funding 0.0 and a null long/short ratio for every token probed on 2026-09-18, and one 24 h open-interest change repeated exactly across unrelated tokens (-10.44 for ETH, SOL, WIF and ONDO). positioning_check rules each field without guessing RYO's undocumented definitions: the same value on two other tokens locked that day is not_token_specific, an opposite sign to OKX's coin-terms OI change of 3% or more is conflicts_with_venue, BTC funding of exactly 0 while RYO's own sentiment tool reports BTC funding is conflicts_with_ryo. A withheld field reaches the prompt as withheld: <reason> and a citation of it is dropped. scripts/gate_ab.py <id> re-runs a receipt's council without the gate; the first run (docs/gate-ab/cd5ada18b179.json) changed nothing, and is published anyway.

Screenshots are generated from the shipped demo ledger by scripts/screenshots.py.

Watch it instead: nota-ryo.vercel.app/demo — a narrated 3:04 walkthrough with a clickable transcript, or the bare file at /demo.mp4.

How a decision is made

gather ──────────► council ──► judge ──► risk (pure ATR math) ──► receipt ──► ledger ──► notify
 5 RYO tools        macro       weighs      entry / stop / target      id = hash(evidence,   Telegram
 + own skills       technician  opinions    or Blocked when price       model, prompts)      Discord
 (price_check,      narrative   by Brier    or ATR is missing
  voices, news)                 weights