We commit: every conclusion can be independently verified — formula open, sources cited, changes audited.
Each modelScore is a weighted sum of 6 factors (out of 100):
| Factor | Weight | Formula |
|---|---|---|
| exposure | 22% | 三轴 anchor 暴露 max() / 3-axis anchor max |
| events | 22% | Σ(impact × e^(-days/30)) 90 天衰减 |
| momentum | 20% | 30d return ÷ realized vol |
| fundamentals | 15% | rev_yoy × 0.4 + fcf_margin × 0.3 + ai_rev_pct × 0.3 |
| technical | 13% | 当日 change_pct 缓冲 / today's change_pct buffer |
| risk | 8% | core 92 / balanced 74 / event 62 / speculative 46 |
Most AI lists lump every 'AI stock' together. We decompose along 3 independent credibility chains: each ticker gets a 0-100 score for OpenAI / Anthropic / SpaceX exposure, based on customer concentration disclosure, contract announcements, IR statements, and SEC filings. Example: MSFT is 92/0/0 because Microsoft holds OpenAI economic interest but has no business tie to Anthropic or SpaceX; AVGO is 88/40/0 from OpenAI 7nm custom ASIC + partial Anthropic networking. Every assignment carries an evidence URL (IR, 10-Q, press release) — fully traceable in the admin panel and public changelog.
Quotes: Stooq EOD (default) / Polygon (real-time, optional) / IEX Cloud (fallback), polled with 15-min cache. Events: GDELT auto-pull + admin curation + mandatory source_url on every record. Fundamentals: rev_yoy_pct / fcf_margin / ai_rev_pct entered by admin from 10-Q / 10-K with source tag. Anchor state: IPO stages for OpenAI / Anthropic / SpaceX maintained by admin + auto-verified from SEC EDGAR after S-1 filing.
• Backtest still accumulating — score_history needs 60+ days to publish an honest top-quintile vs bottom-quintile return; currently shown as 'calibrating' badge.
• Three-axis exposures are human-annotated — when not re-reviewed within review_interval_days (default 90), a 'stale' badge appears in admin and on the stock card.
• Event feed depends on manual curation unless GDELT_ENABLED=1 — after-hours US news may land next day.
• Anchored is an AI research tool, not investment advice. All scores are for research discussion only.
Every add / remove / exposure change writes to stock_pool_history with a reason. The public /changelog reflects this live. Rules: new IPO → admin evaluation → vote in or watchlist; quarterly disclosure of AI revenue share → triggers exposure review; sustained low liquidity → paused (state change, not deletion).
See the live changelog →