Tailfifty desk
424,069 observationssnapshot 2026-10-15Registry

The dataset

424,069 observations from 11 keyless public sources. Every observation carries four timestamps: when it happened, when it was publishedPublished time When a number became knowable to anyone. Different from when it happened (event time) and when the desk fetched it (ingested time). Futures closes publish at the session close; NOAA observed degree days about 30 hours after the day. glossary, when the desk ingested it, and its revisionRevision A later version of the same data point. The old print stays; the new one is knowable only from the moment it arrived. Reads as of an earlier date still see the old print. glossary number. Every read filters on the published time, so a test as ofAs-of (point in time) Reading the data as it looked on a given date: only observations published by then, and the version that was current then. Backtests that read today's cleaned data are reading the future. glossary a date sees only what was knowable then, which is how look-aheadLook-ahead bias Using information that was not available at the time. A close stamped at midnight is 'known' 17 hours before the market closed; a seasonal average that includes next year leaks the future. Every read here filters on the time a number was published. glossary is made impossible rather than discouraged. A changed print becomes a new revision; the old one stays.

SourceWhatSeriesObservationsCoverageMean lag (h)Mean publication lag On average, how many hours after an event its number became knowable. Zero for a price at its own close; about 30 for observed degree days; days for weekly reports. glossaryPublished whenPublished time When a number became knowable to anyone. Different from when it happened (event time) and when the desk fetched it (ingested time). Futures closes publish at the session close; NOAA observed degree days about 30 hours after the day. glossary
NOAA CPC forecasts
cpc_forecast
cooling and heating degree-day forecasts for 15 states, every issue kept as its own revision30210,9392023-12-31 → 2026-10-19-67.6published at the issue time; forecast revisions are first-class
Kalshi
kalshi
hourly implied probabilities for NFL game markets and ladders215105,7542026-08-24 → 2026-10-1458.3at the candle
NOAA CPC observed
cpc_observed
observed degree days for the same states3030,4802023-12-31 → 2026-10-1130.0published about 30 hours after the day
Weather stations
open_meteo
temperature, precipitation and wind for Corn Belt and Gulf Coast sites2829,2072023-12-31 → 2026-10-14119.2published five days after the day, conservatively
Prices
yahoo
daily closes for futures (WTI, Henry Hub, RBOB, corn, soybeans, wheat), energy equities and ETFs, BTC and ETH2821,3612023-12-31 → 2026-10-143.1published at the session close (17:00 ET futures, 16:30 ET equities, 00:30 UTC next day crypto); rows ingested before 2026-10-02 carry the older midnight+1h stamp
Scores and lines
espn_scores
NFL and MLB results, per-team margins, pre-game spreads and totals23713,5842024-06-26 → 2026-10-196.0results four hours after kickoff (synthetic for backfilled seasons); lines when fetched
Injury reports
espn_injuries
per-player status and per-team counts677,2922026-02-09 → 2026-10-144.7when fetched; the earlier backdated stamp was survivorship-biased and is gone
CFTC positioning
cftc
Commitments of Traders for five futures152,1752024-01-01 → 2026-10-0592.0Tuesday positions published Friday 20:30 UTC
Crypto spot
coingecko
BTC, ETH, SOL price and volume92,1402025-09-18 → 2026-10-141.5at the print
Crypto derivatives
okx
funding rates and open interest41,1352026-03-23 → 2026-10-149.3funding at its timestamp; open interest one day later
Hurricanes
nhc
active Atlantic storm maximum wind122026-09-18 → 2026-09-220.0at the advisory

What the miner may test against what

Weather features may be tested against commodities, equities and games; commodity features against commodities and equities; crypto against crypto; market features against commodities, equities and crypto; game features against games. Physically absurd pairs are never run, which keeps the multiple-testing burden honest. Raw third-party series are not redistributed; only derived figures are published.