QuantSquawk accuracy ledger: grading methodology v1
====================================================

This document fixes the rules used to grade every score in the public
accuracy ledger (https://quantsquawk.com/accuracy). Its SHA-256 digest is
timestamped in Bitcoin through OpenTimestamps, so the rules cannot be
changed after results are known. Any change will be published as a new
version (v2, v3, ...) with its own timestamp; earlier versions stay online.

1. Which headlines enter the ledger
   - Only headlines scored by the AI model (no fallback heuristic scores).
   - Only headlines from a named real source.
   - Only event tier "Tier 1: Systemic" or "Tier 2: Major".
   - Each headline text is recorded once. Rows are never deleted.

2. What is recorded at scoring time (and hashed)
   - headline, source, tier, score (-100 to +100), instrument, scored_at.
   - Leaf hash = sha256(JSON.stringify([headline, source, tier, score,
     instrument, scored_at])), lowercase hex.
   - Leaves are rolled hourly into a Merkle root (sha256(left + right) over
     hex strings; an odd node pairs with itself) and the root is
     timestamped with OpenTimestamps.

3. Which instrument a score is graded against
   - If the headline names a mapped company, that company's ticker.
   - Otherwise a proxy by category: Technology -> QQQ, Energy -> XLE,
     everything else -> SPY.
   - One instrument per headline. Headlines that affect several assets are
     graded on this single instrument only; this is a known limitation.

4. Prices
   - Last traded price from Yahoo Finance at the time of each check.
   - Entry price: taken at the first check after scoring. If no entry
     price is captured within 15 minutes of scoring, the row is EXCLUDED.
   - 30-minute price: taken at the first check at least 30 minutes after
     scoring. If none is captured within 90 minutes of scoring, the row is
     EXCLUDED.
   - End-of-day price: taken at the first check at least 15 minutes after
     the close of the US cash session (treated as 21:00 UTC) on or after
     the scoring time.
   - Checks run every 5 minutes. Missed prices are never backfilled.

5. What counts as a hit
   - move = (later price - entry price) / entry price.
   - Hit if the sign of the score matches the sign of the move.
   - Unmeasurable (not counted either way): a score of exactly 0, or an
     absolute move under 0.01%.
   - Graded separately at 30 minutes and at end of day.

6. Published statistics
   - Rolling 30-day window, using the 30-minute result.
   - Bands: all scored rows; |score| >= 60; |score| >= 80 and Tier 1.
   - A hit rate is not displayed for a band until it has at least 30
     resolved rows.
   - Excluded rows are not counted in any statistic.

7. What this does not measure
   - Directional agreement only. Not returns, not tradable after spreads,
     slippage or timing. Not a track record of trading performance.
   - Scores are AI-generated estimates. Not investment advice.

Effective for all rows graded from the time this document was first
timestamped.