Where the research stands
Here is what has been run so far, and what has not.
No study tests this loop.
Its design borrows from nearby research on paper, handwriting, stepping away from a problem, what example answers do to new ideas, and how people think with AI. Together they make the loop plausible. That is a hypothesis, not a finding.
The evidence brief has the full argument, and what it does not support.
Some setups are untested.
So far the loop is verified end to end on 1 of 5 setups. The Install page lists each one in the README’s own words.
No published session reports yet.
We have run the loop end to end ourselves, on Claude Code. That shows the loop runs, not that it helps.
The evidence brief’s section for real sessions is still empty, so there are no testimonials here and no claim about how many people finish.
Ran a session? File a session report on GitHub. An abandoned session helps as much as a finished one.
Reading more will not settle any of this. Pages that come back with handwriting on them are where the record starts.