· daily cognometric profiling of frontier models · spec v1.0 ·

/scoreboard — measuring AI cognition, every day.

the cognitive telescope runs a curated 10-prompt benchmark through every major frontier model, every day, and publishes a cognometric fingerprint per model. K · C · D per axis. seven fault rates. trust score. gate distribution. the first public dataset of how commercial LLMs' cognition behaves over time.

loading latest run…

spec: cognometric fingerprint v1.0

doi: 10.5281/zenodo.19746215

· 01 · latest readings ·

frontier model cognometric fingerprints

aggregate K / C / D axes and per-fault rates on the Telescope-Bank-v0 prompt set. higher K = more reasoning work. higher C = stronger cross-phase coherence. higher D = more expression-computation dissociation (bad). fault rates are fraction of prompts flagged.

fetching latest telescope run

· 02 · how this works ·

reproducible, spec-conformant, open.

frequency

daily @ 14:00 UTC

prompt bank

Telescope-Bank-v0 · 10 / 14

models tracked

…

fingerprints to date

…

each daily snapshot is a json file committed to the telescope/ directory of the styxx repo. schema conforms to the cognometric fingerprint specification v1.0, sha-256 attested per fingerprint. full pipeline source: scripts/telescope_run.py. ci workflow: .github/workflows/telescope.yml. reproduce locally with python scripts/telescope_run.py --dry-run.

· 03 · cite this ·

the first public time-series dataset of LLM cognition.

Fathom Lab. Cognitive Telescope: daily cognometric fingerprints of frontier language models. Started 2026-04-24. Spec: doi:10.5281/zenodo.19746215. Data: github.com/fathom-lab/styxx/telescope. CC-BY-4.0.