howclose.to
CHANGELOG

Site notes · recorded changes

What changed, and why.

A factual record of changes to the evidence, the page structure and the links that carry it.

7 recorded changes

Site

Shared problem-page design shipped

The 20 technology pages now use shared Plain and Engineer sections, with figures selected from each record's data.

Observed and target states stay explicitly marked, and source links can open the relevant evidence, figure, history or builder record.

Open a problem page ↗

Site

Problem pages use shared sections

The site moved to a modular page contract so the same evidence sections can serve every technology.

The AGI page no longer has a problem-specific section fork; section numbers derive from the sections a record actually renders.

Open a problem page ↗

Data

Steward figures became re-pullable

Steward-owned numbers are regenerated from their source files so a cached estimate cannot quietly become the new record.

The ingest registry now wires the METR steward source through its own ingester; other sources remain explicitly separate.

See the sourced ruler ↗

Data

The frontier claims register opened

Model claims now carry their benchmark, conditions, verification state and primary source in one register.

A hollow mark means the lab reported the number. A solid mark means a named steward checked it.

Inspect the frontier ↗

Site

Problem URLs were shortened

Technology pages moved from /problems/name to /name, with permanent redirects kept for shared links.

The shorter address is canonical while the old address remains safe to follow.

Browse the register ↗

Site

Rulers became a shared section

Independent benchmark rulers can now appear on any problem carrying a benchmark record.

The section is data-selected, so adding a ruler set does not require a page-specific component branch.

View a ruler set ↗

Site

The first public register opened

The site launched with the fusion programme fully researched and five further programmes stubbed.

The launch record described progress estimates as editorial judgments rather than measurements from a universal formula; the methodology was still forthcoming.