Opening threshold

Contract 0.1, issued in publication v0.3. Not a forecast.

No scored forecast ships until every check below is true. SW-F-001 refused a software-horizon forecast because the named series had stopped and the suite was already unreliable above 16 hours. This page is the reopening test.

A draft may become Open only if a stranger can apply this page and get the same yes or no. Author intent does not count. Measured records and Argued lines do not pass through this page.

Three checks

All three must be true on the day the row opens. A later failure annuls or supersedes it.

Current series. Some publisher must describe a page as the current public series for the quantity. A frozen archive, or a “no longer actively updated” log line, fails. METR’s time-horizons page failed on 8 September 2026. This site does not designate the successor.

Named methodology. The forecast names the methodology version in the publisher’s words. Time Horizon 1.0 and 1.1 are different series. If the publisher replaces the version and the row did not already name that condition, the row is Annulled.

Quantity still published on the resolve-by date. The source must still be issuing the quantity. If it is not, the resolution is Annulled, not No. Silence is not a miss. A forecast about whether the publisher will resume is not a capability forecast, and version 0.1 does not open it.

Extra bars for software tasks

The named suite must still be reliable at the horizon the row scores. A 50 percent horizon above 16 hours fails while METR’s warning stands. The unit stays skilled-human time. A percent score on a coding leaderboard, OSWorld, or a vendor self-report does not pass.

What remains shut

As of 9 October 2026, no row passes. Open forecasts: 0. Domains 2 through 6 stay shut because no instrument is named. A new leaderboard does not open them.

Depends on Claim classes and Domains.