A selected later state
An excerpt from a later draft, not the published paper.
From research to shared understanding
The paper is only
part of the story.
Our aim: AI that helps develop papers people can understand, scrutinize, and build on.
A writing trajectory links drafts, working notes and revisions over time. This is our first example of a broader direction: learning from how research and development projects unfold.
ImitSAT / writing historyFrom the manuscript
The note records a writing request. Later prose makes the approach explicit. The history connects the two.
The craft of making
research understood.
research understood.
Understand
Make the contribution recognizable and the explanation clear.
Scrutinize
Keep claims accountable to what the evidence supports.
Build on
Aim for an account others can examine and use.
01 / Follow the evidence
Open the history.
Follow the writing decision.
Start with a writing question. Follow how the draft develops, then open the passage or its data to look more closely.
How to follow the example
The situationA learned model suggests choices inside a problem-solving program. The example is about explaining that research, not teaching the solver.
The historyUse Next and Previous to follow the writing. Some questions examine different aspects of the same draft.
The evidenceRead the passage first, then its explanation. Open the full excerpt for context or its data for technical inspection.
The writing situation
This is a selection of drafts, not every edit. Highlights draw attention to wording; they are not tracked changes.
02 / The craft in the records
One paper.
Different kinds of judgment.
Choose a writing skill to see the passages that illustrate it. A passage can involve more than one skill.
03 / Open possibilities
Your question. Your way to use the records.
Explore the histories as context, learning material or evaluation examples. You choose the question; we want to understand which records and preparation would help your team.
These are possible uses, not measured results.
Explore the data behind the storiesExplore three possible writing tasks
Newly written tasks show how the records could be used. They are examples, not completed model evaluations.
04 / Explore the data
The story opens
into real data.
Explore the records behind the stories. Choose a passage or writing question, inspect its details, or download the complete sample as JSON.
Historical evidence
Manuscript passages, working notes and recorded draft revisions.
Writing commentary
Explanations of writing decisions, relevant skills and possible value for readers.
Possible tasks
Newly written instructions and review criteria, kept separate from the manuscript history.
Draft passages, writing questions and possible tasks
JSON format
JSON format
1Research paper
3Writing histories
21Manuscript excerpts
7Writing skills to explore
24 writing questions explore these passages. Some use the same evidence, so they should not be counted as independent examples.
The broader direction
Writing is the first chapter.
We plan to extend from writing into broader AI research and other scientific fields, connecting questions, experiments, results and revisions. Software and game-development histories are potential extensions.
Writing is the supplied example. Broader coverage is planned, not present here. Game development means designing, building, testing and revising games—not gameplay traces.
Our aim is better human–AI collaboration. The histories do not identify who or what made each edit.
Questions & approachesIntended broader coverage
Experiments & resultsIntended broader coverage
Interpretation & next stepsIntended broader coverage
Writing developmentSelected evidence in this sample
A concrete sample. An open conversation.
Interested in the data?
Get in touch to discuss availability, pricing, and a possible data partnership.
A short introduction is all you need.