For AI teams & data partners
Human reasoning,
tied to real work.
People record their own computer work and explain their decisions as they make them. You hear what caught their attention, what they thought it meant, and what made them change course. You can see what they did next, then hear what they remember the next day with the work still closed.
01 / Why this data
More than a recording of the steps.
Reasoning we ask for
Our guide asks members to explain why they took a step, what they trusted, and what experience told them to look for. Those explanations are part of how we assess the recording and decide what it earns.
Thinking across sessions
Follow real work across recorded sessions, with pick-ups when the work spans days and a recall session the day after the project ends.
Work people already do
Members choose projects they were already going to do. Each project comes with the contributor’s résumé, with name and contact details removed, so you can see what experience they bring.
What teams could use it for
Training computer-use agents
Recorded actions alongside the reasons people give for key decisions, including the options they weighed and the sources they chose.
Evaluating reasoning
A person’s account of a decision, made at the moment, to compare with what a model says and does in similar work.
Long-running work
How people plan, lose the thread, pick work back up and judge it later, across sessions and days.
02 / The collection method
The questions members answer out loud.
These come straight from our Member Guide. Members answer them in their own words as the moments come up, while the work is happening. The reasoning is spoken by the person doing the work, not written afterward or generated by a model.
Changes of plan, uncertainty and dead ends stay in the recording.
- Before you act
- What are you about to do, and why this step now?
- Weighing options
- What are your options? Why are you picking this one over the others?
- What your experience tells you
- What have you seen before that makes you do it this way? What tells you this is right, or that something is off? If this is new to you, what are you guessing, and what is that based on?
- Before you look something up
- What do you hope to find, and what is it for? Why this source over another one?
- While you read
- Which lines matter? Read them out loud. What do they mean to you?
- After you read
- What are the key points? What made sense, what didn’t, and how will you use it?
- When you change your work
- What are you changing, and what made you change it?
- When something goes wrong
- How did you notice? Why do you think it happened? How will you fix it?
- After you stop to think
- What were you thinking about, and what did you decide?
And at five moments in every project
- Start introFirst session
- What are you working on, and what is it for? What do you already know? What does “done” look like?
- Pick-up introStart of each later session
- Before opening your work: what do you remember about where you left off? What did you learn or think about while away, and which ideas stuck? After opening it: what looks different, and what had you forgotten? What’s your plan today? If you worked off the record, what did you change and why?
- Sign-offEnd of every session
- Where did you stop? What will you do next? What’s still unresolved or bugging you?
- Project wrap-upEnd of the last work session
- How did your plan change from start to finish? What was the biggest turning point? What would you do differently next time?
- Recall checkThe next day, no notes, work closed
- What was the project, and how did it end? Which three things do you remember most clearly? What fact or number do you remember exactly? What do you now think you got wrong?
03 / One project, start to finish
What the reasoning sounds like.
An example from someone closing the month’s books for their own small shop, followed across its sessions. Notice where experience sends them first, what turns out to be wrong, and what they still remember at the end.
- Start introDay 1
- “I’m closing out October for my shop. The sales report has to match what actually reached the bank. I’ve done this every month for about six years, so I know roughly where it goes wrong. Done means the two numbers match and I can explain anything left over.”
- What experience tells themDay 1
- “This total looks off. I’ve done enough of these to know shipping often gets counted twice, so I’m checking that line first.”
- Weighing optionsDay 1
- “I could work from the payment processor’s report or the bank statement. I’m going with the bank statement, because the report shifts when refunds post late. The bank is what actually came in.”
- Pick-up introDay 2
- “Yesterday I got it down to about a $240 gap, and I blamed refunds. On my walk this morning I remembered we changed processors mid-month. Opening it now… the gap is $243, not $240. Today I’ll split the month at the changeover.”
- Catching a mistakeDay 2
- “Wait, this refund is dated the 31st, but it didn’t clear until November. I put it in the wrong month. That’s why this wouldn’t tie out. Moving it.”
- Sign-offDay 2
- “Stopping here. The two numbers match except for $12 I can’t explain yet. Next I’ll go through the fee report once more.”
- Project wrap-upLast work session
- “I expected a shipping problem, and there was one, but the bigger issue was the processor change. The turning point was splitting the month in two. Next time I’d check for changes like that first.”
- Recall checkThe next day
- “It was October’s books, and it ended with everything matching except a small fee. I remember the $243 gap and the refund that crossed into November. I now think I was too quick to blame refunds.”
Illustrative example written for this page. It is not a recording from a delivered dataset.
04 / The planned deliverable
A project, with its context.
The reviewed recordings, recorded events and contributor résumé belong together. Each pilot defines the export format, processing and acceptance standards before collection.
- Screen & voice
Video from all displays and the contributor’s spoken narration across the recorded sessions. No webcam footage.
- Actions & timing
Keystrokes, mouse clicks, scrolling, app and window context, session timing and pause gaps.
- Project progression
Start and pick-up introductions, session sign-offs, the project wrap-up and next-day recall, kept with the work they describe.
- The actual résumé
Work history, education, skills and qualifications, with name, phone number, email, street address and personal links removed. The buyer agreement will allow background selection and AI training on this copy.
A résumé is contributor-provided background, not independently verified credentials. Removing direct identifiers does not guarantee anonymity.
- Planned processing
Transcripts with word timing, explanations aligned with actions, and review information. Each pilot will specify what is provided.
Technical details from prototype tests
Video: all displays as one wide H.264 MP4, about 8 to 9 frames a second on two 1080p screens, with small text readable.
Voice: FLAC, 16 kHz mono.
Actions: time-stamped clicks, keystrokes and scrolls, with app, window and interface element details where available. Mouse moves are kept at up to 10 a second.
Timing: recorded video frame timestamps help line up speech, video and actions.
Size: about 0.8 to 1.3 GB per recorded hour.
Measured on one Windows PC in short tests. Windows only for now.
What the package does not include
The package covers recorded sessions. Members may work off the record and explain those changes when they return. Final work files and replayable environments are outside the current offer, and we do not assign tasks.
05 / Quality & rights
Built so the data can be used.
This is how the service is designed. Pilot agreements will set the standards you can hold us to.
How we plan to keep quality high
- Selection. We accept applicants who look able to produce good work and fit the backgrounds a collection needs.
- Pay follows reasoning. Narration is what members are paid for, and strong reasoning is what moves them up a level.
- Whole-project review. We review each project after Submit. Made-up tasks and wasted time are not paid and are deleted unused.
- Consequences. Members hear what went wrong, and those who keep having issues are removed.
How rights and privacy are handled
- Consent first. Members agree to terms before recording, and we say plainly that recordings are licensed to AI companies to train AI.
- Clear ownership. The terms are intended to give CommonWhy ownership of submitted recordings and the right to license them. Projects can’t be withdrawn after Submit.
- Right to share. Members confirm they have the right to record and share what they capture. Client, customer, patient and confidential company information stays out.
- Private information. Pause stops all capture. We remove private information we find in review, but we can’t promise to catch everything.
- Voice. Licenses prohibit cloning or impersonating a member’s voice, while allowing general AI training on narration.
06 / Who contributes
Broad backgrounds.
A collection that fits.
CommonWhy is designed for everyday people and experts across occupations and experience levels.
You tell us which occupations and backgrounds you want and don’t want, and that decides who we accept for the collection. The company’s wider scope remains broad.
Members choose real work they were already going to do. We do not assign tasks or ask them to act out a workflow for the recording.
07 / Founding pilots
Shape the first paid collection.
We’re looking for buyers whose data needs can guide the first collection. Scope, price and funding come from that conversation.
What founding partners get
- A say in who we recruit, the data format and delivery
- An early review of the first material against the agreed criteria
- Exclusivity on their collection, if they want it, defined and priced in the agreement
- A first look at related future collections, where agreed
Define the need
Tell us what you want to train or evaluate, the people you need represented and the data your team can use.
Agree the pilot
Set the collection size, contributor criteria, format, acceptance standards, funding, delivery milestones and license. We can discuss a rough estimate as scope takes shape.
Prepare and collect
Complete the required member and delivery setup, then recruit, collect and review. Agree an early review checkpoint to assess the first material.
Evaluate and decide
Review delivery against the agreed criteria. Further paid batches can follow if the collection meets your needs.
Where we are today
A working prototype.
A service in development.
Our Windows recording prototype works. Member onboarding, uploads, review, delivery and final agreements are still being built. We are not yet collecting from outside members.