Synthesis13 min read
Research synthesis
See what your papers support, where they conflict, and what you can actually claim.
Carrel is a research synthesis tool that extracts claims, methods and findings from your papers with their source passages, maps what supports, contradicts or qualifies what, and turns those relations into sections you can draft from.
Summary is not synthesis
Reading forty papers produces forty summaries. Synthesis is the different job of saying what they establish together: which finding holds across studies, which two results cannot both be true, and which condition explains why they differ.
That job fails for a mundane reason. The unit you need while writing is smaller than a paper, a claim or a finding or a limitation, and the relations between hundreds of them will not stay in anyone’s head. So the draft falls back to one paper per paragraph, and the review lists instead of arguing.
How it works
What Carrel does, in the order it happens
01Extract the claims with their passages
Each paper gives up its claims, methods, findings and limitations, each with the exact quote behind it. Accept, edit or reject them; the model’s original is kept for audit, and the passage is one click away.
02Link what supports, contradicts or qualifies what
Carrel proposes relations between claims and you judge them. What emerges is the synthesis matrix drawn as a graph: agreement, conflict, and the conditions that explain the conflict.
03Find the gap the evidence points to
Where the claims run out, where two findings conflict with nothing to resolve them, where a method has never met a population. Candidate gaps are flagged from the structure rather than announced in the last paragraph.
04Turn the structure into sections
Background, what the field agrees on, where it contradicts itself, where the gap is: Carrel proposes the sections and fills each with the claims and saved answers behind it. Rename them, merge them, throw out what is wrong, then draft.
Guides
The method, without the product
The method, with before-and-after examples of a paragraph that argues instead of lists.
Research writing13 min read
How to Start a Literature Review When You Don’t Know Where to Begin
If you are still collecting papers and do not yet have a working set.
Research writing13 min read
Thesis Writer’s Block: What to Do When You’ve Read Everything but Still Can’t Write
If you understand the themes but still cannot get words onto the page.
Questions people ask
More in the full FAQ.
- What is the research graph for?
- Claims, methods and findings are extracted from your papers with their exact quotes, then linked by what supports, contradicts, qualifies or extends what. Those relations are what make a review paragraph argue instead of list — when you draft, a contradiction comes through as contrast rather than as another item in a sequence.
- What happens if my papers do not answer the question?
- You are told. That is a deliberate design choice and it costs us better-looking demos: an answer built from a model’s training data is unciteable and unverifiable, and it is exactly the sentence you cannot defend later. “None of your papers speak to this” is also useful information — it is either a gap in your folder or a gap in the field.
- Can it be wrong?
- Yes, and we would rather say so. Extraction can misread a table or a figure caption, a retrieved passage can be the wrong one, and a model can misinterpret a passage it is looking straight at. What the design buys you is that every one of those is one click from being checked.
- Will it catch me if my writing drifts from the source?
- It will flag it. Because a citation is a link to a passage, the sentence and the evidence can be compared — so when a paragraph you have tightened no longer says what its source says, Carrel tells you and you can open the passage and decide.