The user, twice, after four review agents reported: "I asked the agents to do some research,
but I don't see any of them writing a report" and then "It needs to be clear that whenever
agents are dispatched to do some research, they should write a fucking report."

WHY THE DISPATCHER WRITES IT. A subagent cannot write the journal — it has no ledger unless
one is granted, and the grant is for work, not for reporting. So the report is the
orchestrator's job: it has all four answers in front of it and nobody else ever will. The
subagents' findings live in a transcript the user cannot read.

WHY TO-DOS ARE NOT ENOUGH. A to-do says what to do next; it does not say what was asked,
what was checked, what was found to be FINE, or what the reviewer refused to flag. Eight
briefs scattered across a list is not an answer to "what did they find" — it is the raw
material of one. The report is what the user reads; the to-dos are what the agent works.

WHAT COUNTS AS RESEARCH. Any dispatch whose product is FINDINGS rather than a change: a
review, an audit, an investigation, a comparison, "go and see whether…". A dispatch that
edits files and reports that it did so is work, not research.