Skip to content
Published studies are shown for demonstration purposes, not because they are good or bad research. Demonstrator · items marked Example are invented · what exists today
Search

Get started

This page takes you from installing e2er to a published study on e2er.org, with your own question and your own data.

Before you start: About e2er · a published study Demonstration

  1. Choose a template
  2. Install
  3. Your first study
  4. Review and steer
  5. Check and publish

Choose a template

A template is the sequence of steps a run follows. e2er 0.16.1 has eleven, and each fits a different kind of study. The template is named when the run starts. Templates and data sources by field groups them by the field they serve.

empirical In place Your question, data and literature in, a draft paper out. It pauses only after the steps you name with --review-at.
empirical-preregistered In place The empirical template with three pauses for you: a design review, a pre-registration frozen before any estimation, and a review of the draft.
event-study-finance In place Abnormal returns around announcements. Before estimation it also checks the estimation window, overlapping events and that every event date is a trading day.
replication In place Reruns a published study from its Zenodo replication package in a Docker sandbox and compares every reported number with the rerun.
field-map In place Maps a research field: its papers from OpenAlex, the main path of their citation network, lanes named as questions, a reading list and a short field review. It pauses for you to approve the boundary, the lanes and the draft.
descriptive-study In place Describes a sample in any field: summary statistics, distributions, associations and figures, with no causal claim. A data check stops the run on missing units, duplicate rows, unexplained gaps or missing coverage before anything is computed.
time-series-forecasting In place A series, its trend and seasonality, models and forecasts. The hold-out is frozen before any model is fitted, and every model is judged on it against naive baselines.
policy-evaluation In place What a policy changed, by difference-in-differences. The design is checked against the data before estimation; the event study, a pre-trends test and a placebo or sensitivity analysis are required after it.
spatial-analysis In place How a variable varies across regions, with maps and Moran's I. The boundaries' source, the coordinate system and the weights are checked before the analysis.
systematic-review In place A systematic review with a meta-analysis, following PRISMA 2020. It pauses for you at the protocol, a sample of the screening and the draft; every extracted number is traced to its page and quote.
text-analysis In place Counts and models of the language of a corpus, from Project Gutenberg or your own texts. The corpus design is fixed before the corpus is built, and e2er recounts every number from the corpus.

Install

e2er runs on your computer. It needs Python 3.11 or newer, and the install steps use uv, a small tool that brings its own Python. The last step starts e2er, and the browser opens its setup page, where you choose the AI provider, your literature and your data.

Install e2er: step-by-step on the Install page

Your first study

Your datasets go in data/ (.csv, .tsv, .jsonl, .parquet or .xlsx) and your PDFs in literature/. On New study in the dashboard you tick the data files and papers the study uses; in a terminal, --data and --papers name them, and without them the study takes everything in both folders. The files stay on your computer; the text of a PDF goes to your AI provider when a specialist reads it. A study can also load data from the 16 data sources e2er connects to, among them FRED, the World Bank, NASA's Exoplanet Archive and Project Gutenberg. e2er records each load with the source's terms and citation; data whose terms keep them with the source are not published with the study, and its code loads them again. The run starts with your question and the template. e2er prints the study’s id and starts a dashboard on your computer. Specialists work through the template’s steps: design, data, estimation, drafting and review. Checks sit between the steps, and a failed check stops the run, for instance when the estimation does not follow the specification declared before it. At each researcher step the run pauses until you act. When your Claude or ChatGPT plan reaches its usage limit, the run pauses too; press Resume, or run e2er resume <id>, once the limit has reset.

e2er run "<your question>" --template empirical-preregistered
e2er status <id>

A run made only to try e2er or to show how it works takes --demonstration; its paper, report and dossier then say that it is published as is and not a research contribution.

The terminal as text 
$ e2er init --defaults
e2er init --defaults — non-interactive setup
  ✓ scaffolded ~/my-study/data/ and ~/my-study/literature/ (drop your data + PDFs there)
  ✓ Wrote ~/my-study/.env
e2er install-skills: copied 189 file(s), skipped 0
  → ~/.claude/skills
  → ~/.codex/skills
  → ~/.gemini/skills
  ✓ ready — verify with `e2er doctor`, then `e2er run "<your RQ>"`
The recording was made with e2er 0.11.0 installed from PyPI; every command in it ran, the wait while pip installs is cut, e2er run is typed but not started, and the review is left with q. The study it inspects is the FOMC event study, paused at its design review. The text version is output of e2er 0.13.6; there the study folder is shown as ~/my-study and the home folder as ~.

Review and steer

At a pause the dashboard shows the files the step produced and four actions; the same actions work from the command line, and a run that has ended can be sent back to one of its steps. Every action is recorded in the study’s dossier. In the FOMC event study, the design review showed that 11 of the proposed event dates were wrong, and the researcher replaced them with the announcement dates from the data before anything was estimated.

The dashboard's review screen for the step review_design: a short explanation and the first file, paper_plan.md, in an editable box with a Save edit button.
The review screen of the FOMC event study at review_design (template event-study-finance), in the local dashboard.
The foot of the same screen: a box for an instruction to the following steps, a menu to send a step back with a remark, and the Approve and continue button.
The foot of the same screen: instruction, send back, approve.
Edit a file Opens one of the step’s files in your editor. The following steps read your version. e2er review <id> --edit FILE
Give an instruction Every following step receives it. e2er review <id> --instruction "…"
Send a step back That step or specialist does its work again with your remark, and the run continues. e2er review <id> --send-back STEP --remark "…"
Approve The run continues. e2er review <id> --approve
Rerun from a step After the run has finished, failed or stopped: that step and every later step run again with your remark, which the dossier records. e2er rerun <id> --from STEP --remark "…"

Check and publish

When the run ends, e2er export writes the study folder: paper, code, data, results, design and reviews, with a fingerprint for every file. e2er verify checks the folder offline and without any key. e2er login signs you in to e2er.org. Run e2er publish with --dry-run first; it prints the request and sends nothing. With --to https://e2er.org it publishes the description, the dossier and the fingerprint of every file; the files stay with you. e2er.org then shows the study page at e2er.org/<owner>/<project> and its dossier. A study published only to show how e2er works takes --demonstration, and its page, dossier and paper carry the Demonstration label.

The paper, the data and the code stay private unless you say otherwise; readers on e2er.org then see a button to ask you for them. --paper public --paper-url <address of the PDF>, --data public --data-url <address> and --code public --repo <repository> --commit <commit> publish them with the study. --coauthor "Name|github=<login>|role=<role>" lists a co-author, who confirms the credit on e2er.org. A study that used GitHub data is published only with its paper public: GitHub allows research use of its data only when the publications are open access.

e2er export <id>
e2er verify <folder>
e2er login
e2er publish <folder> --owner <github-login> --project <name> --dry-run
e2er publish <folder> --owner <github-login> --project <name> --to https://e2er.org
$ e2er verify showcase
2026-10-05 09:48:55 | INFO     | src.core.renderer.tables | render_tables: wrote 13 table(s) to /var/folders/61/pm7xd63s3kg1mr8hzpykx7780000gn/T/tmp4_5nclfl/tables
  ✓ [PASS] integrity  69 files hash-verified against provenance.json; 346 table cell(s) traced in the derivation graph
  · [SKIP] anchor     no e2er.json (not published): verified against the folder only
  ✓ [PASS] numbers    290 table cell(s) are source values at the precision shown; 48 coefficient(s) checked, 0 inconsistent; text (not gating): 292 of 307 number(s) are source values, 4 differ from the quantity named nearby, 11 untraced
  ✓ [PASS] tables     13 table(s) reproduce byte-identically from the sidecars; 2 not renderer-produced (rawcontrast.tex, sumstats.tex)
  ✓ [PASS] spec       estimation implements the declared identification
  ✓ [PASS] citations  33 cite key(s) resolve in refs.bib; snapshot: 33 verified, 0 unverifiable at run time

✅ Bundle verified — 4 content check(s) passed (numbers, tables, spec, citations); every file matches provenance.json.
   anchor: none outside the folder — verified against the folder only
Real output of e2er verify on the folder of the Bitcoin study, e2er 0.13.6. The study is not published from this copy, so the anchor check is skipped.

You can also publish in the browser on the Publish page, after e2er publish <folder> … --offline has prepared the folder. Everyone whose templates, specialists and skills your run used is credited in the dossier; see Credit.