Methodology & data lineage

We curate 27 institutional fund managers — a deliberately small set of value, activist, Tiger-lineage, multi-strat, and systematic shops whose 13F filings are most heavily watched as smart-money signal. The full list (with style tags and AUM) is on /universe.

Every quarter, each fund discloses long US equity positions on Form 13F-HR within 45 days of quarter end. We mirror those filings, compute per-quarter diffs (new / increased / decreased / exited), and roll them into cross-fund consensus, entry/exit feeds, and backtested portfolios.

• 13F filings & fund metadata: SEC EDGAR directly (no third-party aggregator). Filings are fetched on demand and cached server-side; snapshotted funds are persisted into src/data/backtest/<cik>.json so the dashboard renders instantly.
• Fundamentals: SEC XBRL companyconcept API — the same source EDGAR uses for its own financial summaries.
• Prices: Yahoo Finance for daily close prices and sparkline series.
• CUSIP → ticker mapping: a seed file built from OpenFIGI, supplemented by a manual fixups list for the most heavily-held names where the OpenFIGI mapping is ambiguous or stale. Coverage is good for large-cap US equities; thinner for ADRs, REITs, and recent IPOs.

• Fills are at the filing-date close— the day the 13F became public, NOT the as-of date (quarter end). This is conservative: a real follower can't trade until the data is filed, often 45 days after the fund actually bought.
• T+45 reality.A position labeled “new this quarter” was opened sometime in the three-month as-of window and disclosed up to 45 days after quarter end. Treat all timestamps as “disclosed” timestamps, not real-time.
• No slippage by default. Backtests are stylized — no transaction costs, no market impact, no borrow costs for shorts. A real implementation will lag.
• Survivorship. Snapshots only include funds that are currentlyon our marquee list. Funds we've since dropped, or that wound down, are not present in the historical cohort — backtests overstate the universe's skill against a true any-fund-of-its-era control.
• See /backtest/follower and /backtest/conviction for the strategies these snapshots feed.

19 of 27 funds are snapshotted across their full available 13F history (back to ~2013 in most cases). The remaining 8are “8Q” funds — too large and high-turnover (Citadel, Millennium, D.E. Shaw, etc.) to be backtested across decades without prohibitive snapshot cost; their window is the most recent eight quarters.

Practical impact: aggregate stats across “all funds” blend two cohorts of different historical depth. The 52Q-cohort numbers are comparable to each other; the 8Q-cohort numbers are NOT comparable to the 52Q ones outside the most recent eight quarters. The fund pages annotate which mode they're in.

• Trian:the 13F parser doesn't yet handle their atypical filing structure cleanly. Trian shows on /universe for completeness but its snapshot is not included in cross-fund consensus or backtests.
• Starboard:CUSIP→ticker coverage on Starboard's portfolio is incomplete (several mid-cap activist targets aren't in the OpenFIGI seed). Tickerless rows still appear with the CUSIP.
• Short positions, options exposure, swaps:not disclosed on Form 13F. A fund “exiting” on 13F may have replaced the position with options or swaps that aren't disclosed at all. Treat 13F as a partial view of the long book.