overview

what a research room is.

a folder you copy and then work in. it is a workspace for sustained mathematical research with an agent, arranged so that understanding accumulates in a form you can come back to — while you keep hold of the questions, the interpretations, and where the effort goes.

the shape of the problem it addresses: a long research collaboration produces a great deal of material, most of it provisional. some of it is load-bearing. some of it was believed for a while and then wasn’t. without somewhere for each kind to live, it all collapses into one undifferentiated transcript, and the next session has to re-derive what the last one knew.

how it divides the work

seven places, each with one job.

conventions.md holds definitions, notation, and standing assumptions — read first, every time. goals/ holds the current questions, stated concisely and edited as they sharpen. journal.md is chronological: the reasoning, and the changes of direction. notes/ holds arguments, examples, computations, and informative dead ends. canonical/ holds the claims currently load-bearing, with their support. review/ holds exact review units, reports, responses, and outcomes. proposals/ holds the optional proposal and panel forms.

method/ sits beside these and describes how the room is meant to operate — culture, execution, reporting, artifacts, review, proposals, adjustable resources, and an optional handoff policy. a method file is read when its subject applies, not on every turn.

what the room is careful about

your ability to steer. a direct instruction to start or proceed authorizes that work; the agent does not answer it with a proposal panel instead. small examples, source checks and toy computations do not need a proposal at all. and an exploratory exchange is not silently converted into a mandate to finish a theorem.

the difference between kinds of evidence. a proposed argument, a finite computation, an imported theorem, cleaned exposition, and an independently reviewed argument are distinguished, and a dependent conclusion is never recorded as more secure than the input it rests on.

independent certification. blind referees see an exact, self-contained review unit — no author confidence signals, no prior votes. review proceeds by natural claims and applications, so an accepted conditional calculation is not mistaken for verification of its hypotheses.

the ordinary register. confidence language appears where it informs a mathematical claim. it does not get appended to every conversation as a compulsory ledger.