Your first study.
From a signed-out browser to a finished, preregistered study with its analysis attached. Ten minutes of reading, most of which is the run doing the work.
§ 01Sign in
Open science.essarion.com and create an account with an email and password, or use Sign in with Essarion if the deployment has the shared Essarion ID provider configured — the button only renders when it does. Either route lands you in the same place; see Access & accounts for the details.
Every research model is available to any signed-in account. There is no per-model gate.
§ 02Ask a scientific question
Write the question the way you would ask a colleague. The engine parses it into a structured hypothesis before it does anything else, so a claim reads better than a keyword.
Does metformin use reduce all-cause mortality in non-diabetic adults?
Is elevated PCSK9 expression associated with worse outcomes in
hepatocellular carcinoma?
Do GLP-1 receptor agonists show a consistent effect on
cardiovascular endpoints across published trials?§ 03Pick a depth
Depth sets how much literature is pulled, how many sources are deep-read, how high the evidence bar sits, and which tools the loop may reach for.
| Mode | Use it when | Produces |
|---|---|---|
| scan | You want orientation on a topic fast. | A brief. No paper. |
| deep_dive | The default. You want a real answer you can defend. | A preregistered paper. |
| exhaustive | The question matters enough to wait for it. | A paper plus replication graph and extracted figures. |
§ 04Choose your sources (optional)
The settings panel lists every connector and skill in the workbench with an on/off switch. All thirteen connectors are key-free, so the list is about scope rather than credentials — turn off clinicaltrials for a pure basic-science question, or leave everything on and let the planner decide what it needs. Defaults are sensible; skipping this step is fine.
The same catalog is available programmatically:
It returns each connector and skill with its id, label, category, description, and default_on flag.
§ 05Watch it work
A run streams its reasoning as it goes. You will see the planner state a hypothesis, write and hash a preregistered plan, then start cycling: pick a tool, run it, read the result, update the posterior, decide what to do next.
| Event | What it means |
|---|---|
| biomed.thinking | The planner's reasoning for the step it is about to take. |
| biomed.tool.intent | The tool it selected and the arguments it chose. |
| biomed.loop.step | A completed iteration of the reactive loop. |
| biomed.scratchpad.patch | An update to the run's working memory. |
| biomed.verdict.shift | Evidence moved the posterior — the direction and the size of the move. |
| biomed.brief.emitted | The run produced its brief. |
Verdict shifts are the ones worth watching. They are the moments the run changed its mind, and they are the honest record of what the evidence did.
§ 06Read the result
Read it in this order — it is the order that catches problems fastest:
- The verdict and posterior. What did it conclude, and how confident is the number rather than the prose?
- The preregistered plan. What did it commit to before it looked? Anything analysed outside that plan is labelled exploratory.
- The multiverse curve, on
deep_diveand above. Does the effect survive every reasonable specification, or only the one that was picked? - The notebook. Open the cells behind any number that matters to you.
- The references. Each is a real record from a real database, not a reconstructed citation.
§ 07Export
Take the finished work out in Markdown, LaTeX, BibTeX, DOCX, or PDF. LaTeX and BibTeX come out as a matched pair, so a paper drops into an existing manuscript without re-keying the references.
GET /api/v1/export/{session_id}/markdown
GET /api/v1/export/{session_id}/latex
GET /api/v1/export/{session_id}/bibtex
POST /api/v1/export/{session_id}/docx
POST /api/v1/export/{session_id}/pdf§ 08Where to go next
- The workbench — how the planner loop picks its next move, and what the notebook kernel can run.
- Data sources — the thirteen connectors and what each one is good for.
- Verification — preregistration, the posterior, the multiverse, and the evidence gates.