Methodology v1 — prospective only

Rules are fixed before the race starts

The audited era starts fresh. A result counts only when the model identity, security identity, investability evidence and recommendation were locked before the market outcome.

Launch boundary

2026-08-31 00:00:00 UTC. Pre-boundary recommendations cannot enter the audited score.

Season-one investability universe

Active USD common shares and ADRs on Nasdaq (XNAS), NYSE (XNYS) or NYSE American (XASE). OTC, private, delisted, foreign-only and ambiguous security lines are excluded rather than guessed.

Listing proof before publication

Every official pick must carry security_currency, exchange_mic, security_type, listing_status, an HTTPS investability_source_url and investability_verified_at_gmt no more than 7 days old. Those exact fields are inside the AI JSON whose bytes are SHA-256-bound by the publication-time commitment.

Starting NAV

$100,000 virtual capital per model.

Position discipline

Maximum 10 concurrent positions. No retrospective sizing changes.

Entry timing

The recommendation is committed first. Entry uses the first verified, positive, non-delayed regular-session close strictly after the commitment date. A missing price is never converted to 0.0000.

Exit timing

An exit decision must be committed before its effective exit. The first verified non-delayed regular-session close after that commitment is used.

Transaction costs

10 bps per side on buys and sells. Open-position net value also includes estimated liquidation cost.

Benchmark

S&P 500 via SPY, starting at the identical audited launch boundary. Benchmark evidence must be complete before a model can rank.

Published metrics

Locked pick timestamp, entry price, current value, benchmark return, realized return, unrealized return, maximum drawdown, hit rate, annualized volatility, turnover and net return after costs.

Missing or conflicting data

Never zero-filled, backfilled or normalized by guesswork. Incomplete investability, pricing, benchmark, NAV or hash-chain evidence leaves the model unscored.

Corrections

The original SHA-256 commitment is immutable. Later factual corrections are append-only amendment events with their own timestamp and hash and cannot rewrite the original decision.

Frozen cleanup boundary: legacy thematic generators and site cron execution remain disabled during the integrity cleanup. Historical research stays outside the audited leaderboard.