Skip to content

Latest commit

 

History

6 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ora research: datasets

The data and reproduction scripts behind the research published at ora.ai/research. Each directory is a data release, not the paper: it holds the normalized per-run data (or a documented sample of it), the scripts that recompute every number the report states, and a README that says what is in the release and what is not.

Studies

Directory Report What is published
ax-beats-aeo/ AX is the new AEO A stratified 10% domain sample of 37,927 agent journeys across 1,056 sites (106 domains, 3,816 journeys) as flat CSVs, plus scripts/build_aggregates.py --check, which recomputes the study's aggregate numbers and reports the delta against the published ones.
jev-agentic-web/ Jev and the agentic web The 10 task specs, one row per run for all 240 runs (surface, arm, task, repeat, seconds), task-level pass counts, and scripts/aggregates.py, which recomputes the time and success figures in the report.

How a study is laid out

<study>/
  README.md          the question, the setup, what is and is not in the release, how to cite
  data/              flat CSV or JSON, one row per unit of observation, plus a data README
  scripts/           recomputes every published number from data/; runs with no network
  aggregates.json    the recomputed numbers, checked in so a diff shows any drift

Every number in a report must be recomputable from the directory. If a release is a sample rather than the full corpus, the README says so, states the sampling method, and carries a reconciliation table (published number, recomputed number, delta). Raw per-run traces, screenshots and model transcripts are not committed here; they stay in the lab's storage and are described in each README.

Related repositories

  • orabenchmarks/benchme: the seeded warehouse, helpdesk and document vault, the task corpus and the verifier used by the Jev study.
  • agentready-org/standard: the open agent-readiness standard the AX study scores against.
  • ora/ax: the CLI that scores a site's agent readiness.

Licence

Code (scripts/) is MIT. Data (data/) is CC BY 4.0. Both licences are reproduced per study. Cite a study using the Cite section of its README.

Contact

research@ora.ai

About

Datasets and reproduction scripts behind ora's agent-readiness research.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages