Turn a question into a reviewed paper.
Type what you want to find out. Stemma plans the study with you, runs the experiments, writes the paper with every number traced to a logged run, and puts it in front of a double-blind review panel, in public.
Nothing runs until you approve the plan.
Can a small transformer learn modular arithmetic faster with a curriculum on operand size?
- Clarify
- Brief
- Repository
- Literature
- Experiments
- Results
- Paper
- Publish
Paper submitted for review
From a question to a paper, in eight stations
Clarify
It asks what it needs to know about your question.
Brief
A written plan: hypotheses, baselines, metrics. You approve it before anything runs.
Repository
The study gets a git repository for its code, results and paper.
Literature
Related work is found and every reference is checked to exist.
Experiments
Your method and the baselines run on the same tasks and seeds.
Results
Tables and figures are built from the logged runs, and only from them.
Paper
An IEEE-format paper is drafted, every number traced to a run.
Publish
You submit it to the library, where a double-blind panel reviews it in public.
Checked, not just written
- Every result has a run behind it
- A number in the paper must come from a run in the study's registry. One that does not is caught before submission.
- Every reference is real
- Citations are checked against arXiv, Crossref, DataCite or OpenAlex.
- The code is kept
- Code, results and paper are committed to the study's repository: yours on GitHub, or a private one Stemma keeps for you if you sign in with Google.
- Reviews are public
- Each paper shows its reviews, scores and decision next to the PDF.