How it works
Built for work that has to survive a client pushing back on it — a consultant’s memo, an analyst’s note, a journalist’s draft. Every claim traces to a source. Every objection raised against it is on the record. What the panel could not settle is flagged as contested, not smoothed over.
A single confident answer is most dangerous exactly where it is most fluent. Giving the researcher a dedicated adversary — one that may only attack how a claim was reached, never whether the conclusion is agreeable — surfaces the weak joins a single pass papers over.
- 1
Interviewer
Asks a few questions, then writes the brief the other three run on: what is in scope, which sources count, what the critic must check, and who the report is for. With the interview off, a single silent scoping call does the same job from your question alone.
- 2
Researcher
Searches within that brief and returns claims with provenance. Every claim cites a source that exists, and what a source states is kept separate from what the researcher inferred by combining sources.
- 3
Devil's Advocate
Attacks how each claim was reached — methodology, sourcing, unstated assumptions, missing evidence, statistics. It may not attack the conclusion, and every objection has to say what would settle it.
- 4
Journalist
Writes the report, and cannot resolve anything the debate left open. What survived critique is reported as agreed; what did not is reported as contested, with both positions and the evidence that would settle it.
Why the line matters
Research and advocacy both stand or fall on the same distinction: what is actually established, and what is still contested. Most tools write over that line — a summary reads as settled whether or not anything in it is. Panel keeps the line visible. A claim that survived critique and a claim nobody could defend do not come out looking the same, and the report says which is which even when that makes it less quotable.
The two colours
Green marks what survived the critique. Red marks what is disputed. They are never used for anything else, so a report can be scanned for where the disagreement is without reading a word of it.
The presets
Modes change what gets asked, what gets hunted, what gets attacked, and how the report speaks — never what counts as evidence. The same panel, the same critique discipline and the same referee run underneath every one of them. Type / in the composer to switch.
- Source check
- One link or citation: what it actually says, how it was made, who paid for it, and whether the claims about it survive the text.
- Claim check
- One claim in circulation, and what stands behind it.
- Paper review
- A study read the way a methodologist reads it.
- Case prep
- Format, side and the resolution verbatim. Builds the opposing case too, and writes a case file.
- Bill analysis
- A bill or policy: what the text requires, who bears the cost, and what is genuinely contested — or the strongest case on each side.
- Compare
- One specialist per entity, and a comparison matrix.
- Any question
- No preset. The panel configures itself to what you ask.
The modes
- Interview me first
- Three to six questions to sharpen the brief, or skip them and start straight away.
- Deep run
- The researcher has to answer the critic and the report says who won. Off, one critique pass runs and its objections all reach the report unanswered.
- Sources
- Which literature the researcher reaches for first — academic, news, primary documents, or a mix.