DATAFOODING한국어

TEAM ATTENTION / INDEPENDENT RESEARCH INSTITUTE

Datafooding

We turn real work into RL environments.

Team Attention is an independent research institute, in the spirit of Artificial Analysis. Our research turns raw data from real work into tasks, graders, and environments for reinforcement learning.

Join session2env and get $200 every month.

Join session2env →

DATA AND QUALITY

Partners own the data. We own the quality.

Partners bring the raw data. They own it and are responsible for its rights. We synthesize the tasks, graders, and environments, and we own their quality.

CONTRIBUTOR PROGRAM

session2env

session2env is our program for contributors. You send three things:

  1. Your real agent session logs, plus the files involved.
  2. A screen video of at least 5 minutes in which you explain the workflow.
  3. Your profile.

REWARD

$200 every month.

If your profile shows relevant experience, the reward is multiplied by 1.5.

RIGHTS

  • We work only with your explicit consent.
  • Raw data stays private.
  • Only reviewed, sanitized projections become environments.
  • You can revoke your consent.

JOIN SESSION2ENV

Agents you use

THESIS

From workflow evals to a company model, and back to evals.

  1. Enterprise workflow evals
  2. RL data
  3. A model specialized for that company
  4. Its operating data feeds new evals

↺ Then the loop repeats.

The moat is not accumulated data. It is the pipeline that keeps producing graders and environments cheaply.

UNIT ECONOMICS / MEASURING NOW

We measure human time per graded task.

OUR TARGET
≤ 1 hour≈ $100 per graded task
LAB PRICES
≈ $200–2,000per task

We are measuring this now. We do not report a measured value yet.

SESSION2EVAL / WHAT WE LEARNED

What we learned from our pipeline.

  1. We map each session to the format of the nearest authoritative benchmark: Terminal-Bench-style coding, AutomationBench-style app workflows, and AA-Briefcase-style knowledge work.
  2. We calibrate each task with gold, no-op, near-miss, alternative-correct, and mutant solutions.
  3. We run each task on 9 models across 3 tiers.
  4. We audit failure traces. In our first round, every defect that changed a verdict was found in failure traces, not in text review.
  5. After the fixes, one workflow task went from 2/9 to all strong models passing, while weak models still failed.

These are internal numbers from synthetic rewrites.

CONTACT

For questions, removal, or correction requests, email bong@team-attention.com.