R-02In progress
Marketing Production Index
What marketers actually build, edit, publish, and measure.
COOL / Lab · South Florida
data.cool
The research lab for human work.
We study how people actually get things done, and turn permissioned records of that work into datasets, benchmarks, evals, and environments for advanced AI. Business first. Not business only.
Exchange open
The rest of this page is the blueprint. We're building the lab now, in the open, one dataset at a time.
- 01Work
- 02Decisions
- 03Corrections
- 04Outcomes
- 05Intelligence
The loop every dataset is built from.
01 / Thesis
The internet taught models what people said. Human work can teach them how things get done.
Models have read most of the public web. What they have rarely seen is the work itself: the workflow behind a deliverable, the decision made under real constraints, the tool reached for next, the correction a senior person makes, the exception that breaks the rule, and what happened after.
The densest record lives inside operating businesses, so that's where we start. But the same signal runs through all human work: the clinician's call, the tradesperson's fix, the editor's red pen, the parent running a household.
It is generated every day, it is mostly unused, and it is the next frontier of training signal.
Historical data tells us what happened.
Workflow data tells us how.
Outcomes tell us what worked.
02 / Research
We measure what actually happens.
R-01Read ↗
RL environments: the 2026 market
What environments are, who is paying for them, and why real business history makes them work. With sources.
R-03In progress
Stock vs. Flow
Why historical business data is only the beginning.
R-04In progress
The anatomy of a valuable operational dataset
What makes business history useful for AI.
R-05In progress
Human correction as intelligence
What edits and overrides reveal about judgment.
03 / Areas
Human work, from the desk to the job site to the kitchen table.
A-01Live
Business work
Email, documents, CRM, tickets and calls: how companies actually operate, licensed through the exchange.
A-02In build
Tasks
Discrete, verifiable tasks drawn from real jobs, with graders built from real outcomes.
A-03Queued
Computer use
Traces of real people driving real software, step by step, to finish real work.
A-04Queued
Expertise and craft
Clinicians, lawyers, engineers, tradespeople, creatives: the judgment calls that never get written down, captured with the experts and paid for.
A-05Horizon
Robotics
Physical work: demonstrations and sensor records from real job sites and operations.
A-06Horizon
Everyday life
Learning, caregiving, running a household, making big decisions. Opt-in records of how people run their own lives, owned by them.
Listed in build order. Business is where we start, not where we stop.
04 / Data
The most valuable data is created while work is happening.
S-01
Operational workflows
The actual sequence of steps a person or team follows to get something done.
S-02
Decision trails
What was chosen, what was rejected, and the context at the moment of choice.
S-03
Human edits and corrections
Where an expert changed the output, and exactly how.
S-04
Tool-use sequences
Which systems were opened, in what order, to finish a real task.
S-05
Exceptions
The cases that broke the process, and how people resolved them.
S-06
Outcomes
What shipped, what converted, what failed, tied back to the work.
S-07
Longitudinal history
The same work repeated over months and years. Patterns, not snapshots.
Standard
Permissioned.
Rights-aware.
Provenance-first.
05 / Supply
From raw operations to proprietary intelligence.
Pillar 01
Live workflows
Marketing production intelligence.
Briefs, drafts, edits, publishes, and performance captured from our own live content products.
Pillar 02
rl.data.cool
The exchange.
Vetted, de-identified business data from owners and brokers, licensed to buyers.
Pillar 03
Jottings.ai
Real-world business data supply.
Permissioned operational records from companies, prepared and licensed. Personal identifiers removed or replaced.
data.cool turns those signals into datasets, benchmarks, evals, and environments.
Datasets
Curated, provenance-tagged corpora of real human work.
Benchmarks
Scored against how professionals actually perform.
Evals
Built from real corrections and real outcomes.
Environments
Replayable workflows agents can act inside.
06 / Work with us
For AI teams
Commission a dataset, eval or environment.
Tell us the human work you need models to learn: domain, task types, volume, and how you'll evaluate. We scope it, source it with permission, and deliver it with provenance.
For businesses
Your operating history is worth something.
The workflows, decisions and outcomes your team already produces can be licensed. Permissioned, de-identified, on your terms.
For brokers
Bring the businesses.
Represent owners on the exchange, stay broker of record, and get paid at close.
Better intelligence needs better evidence of how the real world works.
That's what we're building, from South Florida.
data.cool
Escape the frontier · Delray Beach, Florida