Off-the-Shelf Datasets
Licensed, ready-to-run task datasets and RL environments off the shelf. Sample the cut, run your agents against it, and license shared or exclusive, without a months-long custom engagement.
Show and tell beats a slide deck
Serious data buyers start engagements the same way: a well-understood workflow, real data samples, a running environment. That is the entire design of this catalog. Browse the cuts, request the packet, run your agents against the environment, then decide. The demo is the product.
What is on the shelf now
Live cuts come from our first occupation cohorts and published benchmarks: payroll and HRIS workflows on real software, recruiting operations including the task families behind Recruiter-Bench, and integration-engineering tasks extending ATS-Bench. Each ships as tasks, rubrics, verifiers, expert baselines, and environment access, with private holdouts available for uncontaminated evaluation.
New occupation cohorts come online continuously, sourced from verticals frontier labs cannot staff: claims adjusting, medical billing, freight dispatch, bookkeeping.
Off the shelf does not mean generic
Every cut was elicited from verified practitioners, cross-reviewed, and independently corroborated before it earned a place in the catalog. The shelf is curated ruthlessly: if ten specialists give us a hundred tasks, our job is to sell you the five best.
What ships in the box
Every cut has a sample packet, representative tasks, rubrics, verifiers, and the expert baseline, plus environment access for your own agents.
Off-the-shelf cuts are frozen, versioned, and delivered through a standard harness adapter. Integration is days, not quarters.
License shared cuts at shared pricing, or take a task family exclusive so it never appears in any other customer's corpus.
Each frontier release gets fresh variants, harder difficulty bands, and new failure corpora against the new checkpoint.
Every dataset cut maps to running environments, so the same license covers eval tasks and the RLVR loop that trains against them.
Every task traces to a verified practitioner, an authoring session, and an expert baseline run. Provenance ships with the data.
See it before you buy it.
Sample task packets and environment access for evaluation. Tell us the capability you care about and we will send the relevant cut.