For the complete documentation index, see llms.txt. This page is also available as Markdown.

Public catalog data freshness

How fresh is the public catalog data — refresh cadence, "last updated" timestamps, methodology versioning.

The Stratix Public catalog is refreshed continuously. This page documents how to interpret freshness signals on the site.

What refreshes when

Object
Refresh trigger

Models

New models added when a provider releases or when LayerLens runs the first public evaluation

Benchmarks

New benchmarks added on a roughly quarterly cadence

Public evaluations

Continuous — new runs published as completed

Leaderboard

Updates immediately when a new evaluation lands

Quarterly reports

Q1, Q2, Q3, Q4 — published shortly after the quarter close

"Last updated" timestamps

Every model page, benchmark page, and evaluation page shows a "last updated" timestamp. The most useful field for deciding whether the score you're looking at is current.

Methodology versioning

Benchmarks evolve. When LayerLens updates a benchmark's methodology (sample expansion, harness change, scoring tweak), the methodology version increments and old scores are versioned alongside the new one. The benchmark page surfaces the current methodology version and links to prior ones.

Stale-warning conventions

  • A score is fresh if it's within the last quarter

  • A score is dated if it's >1 quarter and <1 year old

  • A score is stale if it's >1 year old — the page shows a warning chip

Stale scores are still useful for context but should be re-run before being used to make a current decision.

Where to next

Last updated

Was this helpful?