You've done the reading. There are forty papers in a folder, most of them annotated, and a document that opens with "Smith (2019) found… Jones (2020) argued… Patel (2021) showed…" — one paragraph per study, marching down the page. It's thorough. It's also not a literature review. It's a reading list with sentences.
The gap between the two isn't effort, and it isn't more reading. It's direction. A stack of summaries reads across papers one at a time — down a row, paper by paper. A review reads down — the same question asked of every study at once. That second move is where an argument comes from, and almost nobody does it on purpose.
Rows tell you what happened. Columns tell you something new.
Lay your papers out as a grid. Each row is a study. Each column is a question you ask all of them — the method, the sample, the finding, the limitation. Now do the thing that feels slightly unnatural: pick one column and read straight down it, ignoring the rows.
Reading a row gives you a summary you already had. Reading a column gives you a sentence you didn't: every study used undergraduates, or the effect shrinks the more realistic the setting gets. That sentence — the pattern across studies, not inside any one of them — is what a reviewer actually writes. The gap you eventually claim as "understudied" is just a column with a hole in it.
Choosing the columns
The columns are the whole game, and they depend on what kind of review you're writing. A few that earn their place in most of them:
- Method — how the evidence was produced. Grouping by method often reveals that the strongest claims ride on the weakest designs.
- Sample / context — who or what was studied. This is where "nobody has looked at ___" hides in plain sight.
- Finding — not just the direction, but the size. Effects that all point the same way but shrink across studies tell a more honest story than a vote count of significant results.
- Limitation — read down this column and the field's blind spots stop looking like isolated caveats and start looking like a pattern you can build on.
You don't need many. Three or four good columns across fifteen papers will surface more structure than fifty summaries ever will.
From grid to prose
Once the matrix is populated, the writing gets easier because you're no longer staring at a blank page — you're narrating patterns you can see. Each column becomes a paragraph or a theme: here's what the field agrees on, here's where it splits, here's the question nobody's answered. The papers become evidence for claims instead of the subject of the sentence. "Three separate studies, across lab and field settings, converge on X — but all three rely on undergraduate samples" is a review sentence. "Smith found X" is not.
The synthesis matrix is also the honest antidote to a certain kind of AI summary. A model will happily write you three fluent paragraphs about your topic. What it can't do is stand behind which specific papers support which specific claim — and that traceability is the entire point of a review. A grid keeps every cell attached to its source, so the argument you build is one you can defend line by line.
Where Folio fits
This is what the synthesis matrix in Folio is for. You pick the papers, you choose the columns, and Folio drafts the grid from what's actually in each source — then you read down the columns and write the argument. When you're still finding the papers, the Synthesizer in Folio Search does the same move live: it reasons across the results you retrieved, cites each claim back to a specific source, and refuses to assert anything it can't ground. In both cases the tool builds the scaffold. The thinking — deciding what the pattern means — stays yours, because that's the part worth your name on it.
Forty papers isn't a problem to be summarized. It's a table waiting to be read the other way.