Ledger

Contract 0.1, issued in publication v0.4. Open forecasts: 0.

This is the book. A row not listed here is not a published claim. Rules: Claim classes, Domains. Retrieval date: 9 October 2026.

SW-M-001 · Measured

In METR’s Frontier Risk Report dated 19 May 2026, the public frontier in February–March 2026 is given as about 12 hours at 50% and about 1.5 hours at 80% on Time Horizon 1.1.

“Public frontier” and the intervals are METR’s, from Table 1. The 50% figure is printed as ~12h [5h–61h]. The 80% figure is printed as ~1.5h [50m–2h40m]. This row does not pick a point inside those intervals.

Source: METR, Frontier Risk Report, 19 May 2026. Opened 9 October 2026. Not scored. Status: open as a record.

Known bias. The report’s own note says the Time Horizon 1.1 suite cannot reliably measure horizons above 16 hours. The 50% interval crosses that line.

SW-M-002 · Measured

METR’s time-horizons page warns that measurements above 16 hours are unreliable with the current task suite. The update log says that warning was added on 8 May 2026.

Source: metr.org/time-horizons. Opened 9 October 2026. Not scored. Status: open as a record.

SW-M-003 · Measured

METR’s time-horizons update log, retrieved 9 October 2026, records an 8 September 2026 note that the page is no longer actively updated. This row does not claim that METR will never publish a successor.

Source: metr.org/time-horizons, update log. Opened 9 October 2026. Not scored. Status: open as a record.

SW-F-001 · Extrapolated · not opened

No software-horizon forecast is opened. A dated forecast against Time Horizon 1.1 would need a source that is still publishing that series. SW-M-003 says the page is no longer actively updated. SW-M-002 says the suite is unreliable above 16 hours. Opening a horizon date would forecast the source, or would resolve only by annulling.

The 196.5-day doubling time from the 29 January 2026 note was considered and not used. It is not a forecast.

Opened 9 October 2026, as a refusal. Status: not opened.

D-A-001 · Argued

The measured software horizon and the two bounds do not share a unit. No single gap, percent, or date is published. For one more training run at today’s utilization, data movement binds before the CMOS ceiling. Neither bound is the instrument for task horizon. Instrument empty. Not scored. Issued in publication v0.7.

SD-A-001 · Argued

No resolution source for novel, pre-registered scientific claims. Instrument empty. Not the instrument: citation count, co-authorship, lab blog posts. Not scored.

LH-A-001 · Argued

No resolution source for agency whose success condition is not an automatic grade. METR software horizons are not this instrument. Instrument empty. Not scored.

PM-A-001 · Argued

No scheduled external series of real-world manipulation in skilled-human time. Instrument empty. Not scored.

IC-A-001 · Argued

No series that measures time from a measured capability to normal use inside organizations that did not build the model. Instrument empty. Not scored.

SI-A-001 · Argued

No source that defines the share of frontier-model improvement work done by models, with a numerator and a denominator, on a schedule. Instrument empty. Not scored.