Vertical Industry Data
Vertical B2B data built for industry deployment. Tasks are grounded in a full document world — mail, contracts, calendars, working records — and scored against explicit, criterion-level rubrics, with compliance bounds made explicit. This is the format frontier agent evals use, delivered to your domain.
Coverage
- Legal
- Finance
- Healthcare
- Manufacturing
- Enterprise workflows
- Grounded document worlds
- Business rules
Deliverables
- Grounded world bundle
- Task definitions
- Gold responses
- Scoring rubrics
- Compliance bounds & audit record
From source to acceptance
We don't hand-label a pile and ship it. Every category moves through a closed, instrumented loop — generated to a brief, checked by machines, adjudicated by experts, and traceable end to end — but the path each data type takes is its own.
- 01
Grounded World Construction
A coherent document world is assembled — mail, contracts, calendars, working records, and application surfaces — so every task is anchored in a realistic, internally consistent enterprise setting.
- 02
Expert Task Authoring
Domain experts author tasks against that world, each shipping with a gold response and an expected output that reflects how a competent practitioner would actually resolve it.
- 03
Rubric Design
Each task is decomposed into criterion-level rubrics, so a candidate answer is scored against explicit, defensible checkpoints rather than a single holistic impression.
- 04
Compliance-Bound Review
Domain reviewers verify that tasks, gold responses, and rubrics respect the regulatory and business constraints of the vertical, with those bounds recorded alongside the data.
- 05
Two-Pass Quality Audit
A second independent audit re-checks grounding, rubric fidelity, and scoring behaviour, resolving discrepancies before anything is admitted to the deliverable.
- 06
Acceptance & Provenance
Every accepted task ships with an audit record tracing it back to its world, its rubric, and the reviews it passed, making the deliverable defensible to a customer.
Every run emits a learning signal that feeds back into the source set — the pipeline tightens itself, batch over batch.
See the data itself
One real, trimmed sample from this category — the scenarios it serves, why it matters for training, and the shape of the data as delivered.
Where it’s used
- Training and evaluating domain agents on realistic enterprise workflows
- Rubric-scored evaluation of legal / financial analysis quality
- Grounding long-context reasoning in a real document world
Why it matters for training
Critical
Enterprise value lives in the verticals, and this is the only data here with a full grounded world plus criterion-level rubrics — the hardest and most defensible to build.
Notable features
- tasks/
- case_002_covered_parts_counterargument_and_client_confirmation_analysis.json
- case_003_authorized_channel_and_notice_record_analysis.json
- tasks_and_rubrics.delta.json
- world/
- world_project_alloy_notice.zip
# APEX-Agents-Like Law Cases Delivery ## Overview This delivery package contains a curated set of high-difficulty law cases in an APEX-Agents-like format. It includes one complete law world bundle and three task definitions grounded in that world. ## Domain and Environment - Domain: Law - World type: outside-counsel matter workspace - Application surfaces: calendar, chat, code execution, spreadsheet, filesystem, mail, PDF, presentation, word-processing ## Contents - world/world_description.json — world id, domain, application surfaces - world/world_project_alloy_notice.zip — full grounded world bundle - tasks/tasks_and_rubrics.delta.json — all tasks: prompt, gold_response, gold_response_type, rubric, expected_output, world_id, domain - tasks/case_*.json — standalone task records for case-by-case use
Need a sample or a custom build?
Tell us your spec and scale — we deliver to order.
← Back to Data