Release information · v1.1.0 · updated 2026-07-28
Context pack.
A compact, machine-readable release layer for readers and tools that need the Atlas’s scope, boundaries, and source trail without treating it as an unbounded capability assessment.
Claims
22
Concepts
18
Sources
33
Scope
LLM mechanisms and scaling, post-training and inference-time reasoning, tool use and agents, benchmarks, construct validity, evaluation, and the limits of forecasting.
Exclusions
Forecast scenarios, local audit artifacts, headline benchmark leaderboards, price trajectories, quantum computing, and unreviewed capability forecasts are outside this public release.
Release endpoints
Use this pack as a bounded orientation layer. Attaching it to a model does not make the model’s answers authoritative; retain the linked source record, status, and limitation when reusing a claim.