Point-in-time screener screen.pit
Filter and rank US-traded securities as of any date since 2004 on 20 daily price features, computed only from bars up to that date.
PIT · Core tier · 2004-01 → today · — rows
Countries: US
Available now.
Methodology
A screen names a date, up to 32 filters and one sort over the closed price_v1 registry — last price and volume, dollar volume, 20/50/200-day moving averages, 20/50-day EMAs, momentum, realized volatility and average volumes over fixed windows — and returns at most 1,000 rows. The universe is every security with a bar in the five warehouse sessions up to that date, so it is drawn from what traded then rather than from today's listings; it is a traded-security universe that includes funds, preferreds and warrants, with no index, sector or market-cap filter. Each candidate's holder is resolved on the date, and its features are computed from its own bars across its symbol lineage, ending at its last bar; a missing session or too short a history gives null with a reason, never a filled value, and a candidate whose identity cannot be bound safely is excluded and counted by reason in coverage.exclusions. Prices are split-adjusted by default, applying only splits executed on or before as_of, and known_as_of further keeps out splits whose computed knowledge date is after the pin; returns are price returns, not total returns. Today's bar is excluded until the UTC day has ended. Bars and identity are read as currently known, so a past date is screened on corrected bars and the current symbol history — not a replay of what a feed showed that day. The complete result is frozen for 15 minutes and paged by a signed cursor, so warehouse updates never shift a walk. Screener responses keep source_revisions rather than the non-screener ETag contract. Each organization can retain five snapshots, including empty results and results complete on the first page. A sixth returns 409 ERR_SNAPSHOT_LIMIT without Retry-After. A successful DELETE /v1/screener/snapshots/{snapshot_id} frees a slot immediately. Otherwise it remains occupied until expiry. Release each result once read. An active build or exhausted quota still returns 429 ERR_RATE_LIMITED with Retry-After and can refuse a request independently after release.
Schema
| Column | Type | Role |
|---|---|---|
as_of | date | PIT key · screen date · page query |
effective_session | date | latest warehouse session read · page meta |
ticker | string | symbol traded under in the final five sessions |
public_id | string | holder's stable id · shared across share classes |
security_key | string | trading-line key · stable within one tenure |
lineage | array | symbol eras the features span · oldest first |
last_bar_date | date | latest bar on or before effective_session |
staleness_sessions | int32 | sessions since last_bar_date |
values | map | projected price_v1 features · null when not computable |
null_reasons | map | insufficient_history · missing_session · invalid_input |
splits_applied | int32 | splits applied inside the price window |
Use cases
- Rebuild a backtest's universe at each rebalance, screened as of the prior close, from the securities that traded then rather than today's listings.
- Rank a date's universe by momentum or volatility on split-adjusted closes, so a split does not read as a crash.
- Admit names into a backtest only above a liquidity floor, such as 63-day average dollar volume.
- Keep splits out of a screen until their computed knowledge date, with known_as_of.