osmosMind

Data/Vertical B2B

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

The pipeline

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.

  1. 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.

  2. 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.

  3. 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.

  4. 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.

  5. 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.

  6. 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.

A specimen

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

  • Grounded document world
  • Criterion-level rubrics
  • Gold responses included
  • APEX-Agents-like format
apex_agents_like_law_hard_cases_delivery_20260603Delivery package
  • tasks/
  • case_002_covered_parts_counterargument_and_client_confirmation_analysis.json4.2 KB
  • case_003_authorized_channel_and_notice_record_analysis.json5.4 KB
  • tasks_and_rubrics.delta.json13.6 KB
  • world/
  • world_project_alloy_notice.zip8.8 MB
README.mdmarkdown
# 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.

Get in Touch

← Back to Data