Everything Vintage can answer, with the arguments each field needs.

Call this first. It returns every field prefix, whether it needs an entity, whether as_of can be enforced on it, which ones need a key, and an example call per prefix that runs as written, so field names never have to be guessed. discover searches inside this surface.

Turn any identifier into the entity key every other verb accepts.

Accepts a ticker ("AAPL"), a CIK ("CIK0000320193" or "320193"), a FRED series id, or a French dataset name. Start here when unsure what an entity is called.

Search every source's catalog for fields matching a plain-English query.

This is how breadth is reached. There is no per-source tool. With entity set, it also searches that company's reported XBRL concepts, which is the reliable way to find the exact field name for fundamentals.

Fetch any field from any source, as point-in-time rows.

Field forms: "us-gaap:Assets" (needs entity), "price:close" (needs entity), "fred:CPIAUCSL", "french:ff3", "openap:Mom12m" (or "openap:*" for all 331 published claims), "crypto:close" (needs an entity like BTC-USD), "short:short_ratio" (needs an entity), "ape:all-stocks", "13f:value" (needs a manager like BERKSHIRE), "bls:CUUR0000SA0", "bea:T10101". Run discover first if unsure.

as_of is the point-in-time switch: it drops every row that was not public on that date, so you see what a researcher on that day saw, restatements and all. quarter picks a reporting period for 13F holdings ("2024-12-31" or "2024"); without it you get the latest.

Timeline of what happened to an entity, with exact public timestamps.

Currently the SEC filing stream, 8-K material events, 10-K/10-Q, Form 4 insider transactions, 13D/G stakes. Every row carries the minute it became public, which is what makes event studies possible.

Run a cross-sectional backtest. Point-in-time and costed by construction.

At each rebalance the engine sees only data knowable strictly before that date. Costs are always charged on turnover; there is no zero-cost mode.

The validation block leads with purged cross-validation across every held-out combination of blocks. Its multiple_testing sub-block carries the deflated Sharpe and the count of specs run since the last reset. That correction assumes the specs were attempts at one question; if this run starts an unrelated line of enquiry, pass reset_trials=True to zero the count first.

Signals: momentum_12_1, momentum_6_1, reversal_1m, low_volatility, trend_200d.

Score a backtest against the published Fama-French factors.

Answers the question that decides whether a result is interesting: did you discover something, or rebuild a factor that has been public since 1993?

Datasets: ff3, ff5, momentum, ff3_daily, industry49.

Cache size, configured keys, sources available, and specs tried so far.