Skip to content
Demonstrator · items marked Example are invented · what exists today
E2ER

About E2ER

E2ER is open infrastructure for publishing, verifying, reproducing and reusing AI-enabled research. The software is open source under the MIT licence.

Your research can live entirely on your machine

You can run, export and check research without an account. Publishing is optional and adds a public description; the files stay where you keep them. The AI models are hosted services, so the text sent to them leaves your machine. Every connector states what it sends out.

StepCommandStatusLocal or not
Run a templatee2er runIn placeWorkspace and run database stay on your machine. Prompts go to the AI provider you choose.
AI providersanthropic, openrouter, claude_code, codex, geminiIn placeAll five are hosted. A provider that runs on your own computer is not supported yet.
Literature librarye2er corpusIn placePDF files and the library database stay local. A passage is sent to the AI provider when a specialist reads it.
Export the resultse2er exportIn placePaper, results, code and a fingerprint of every file, written to a local folder.
Check the resultse2er verifyIn placeRuns without an account or API key.
Check citations onlinee2er verify --onlineIn placeOptional. Sends the reference list to OpenAlex, Crossref or Semantic Scholar.
Prepare publicatione2er publishPrototypeWrites a description of the study to a local file. On a development branch.
Publish on E2ERPrototypeOptional. Adds the description to the public list; the files stay where you keep them.
Record AI usage in the exportPlannedUsage per specialist is in the run database and the publication description. The exported folder does not carry it yet.
Fingerprint the raw input dataPlannedThe export holds the data dictionary and summary statistics. Fingerprints of the raw files are planned.

A run without enforced checks

On 5 August 2026 E2ER ran one research question with its checks recording problems but not stopping the run. Full record.

53 / 110
table values that differed from their result file (20 critical)
14 / 110
table values with no result file
11 / 18
references the automatic check could not confirm
284 → 2
prose numbers flagged before and after a fix to the check

Governance

Changes to the formats are proposed and discussed on the proposals page. How a steering group is formed and when the formats move to a neutral body are still open.

Background: Brunsson and Jacobsson, A World of Standards, Oxford University Press, 2000; O'Mahony and Ferraro, "The emergence of governance in an open source community", Academy of Management Journal 50(5), 2007, 1079–1106.

Risks

RiskCurrent response
Checks favour work that a machine can check, such as numbers.Each template lists which checks apply to its method; the others are marked not applicable.
Researchers hand over judgment to the specialists.A run can pause for the researcher at chosen stages (e2er run --review-at).
Dependence on a few AI providers.Five providers are supported, and one question can be run on several and compared (compare runs).
Usage figures invite gaming.The site shows no scores or rankings of people. Only published studies count in figures.
A study can pass every check and still be weak."Checked" means traceable. Quality is judged by readers and reviewers.

Data protection

E2ER does not upload your data. The datasets in Discover are public deposits on Zenodo and Harvard Dataverse. Approval and consent for research with personal data are not recorded yet; a field for them in the study description is planned.

Credit

Published studies list their contributors with CRediT roles and cite the templates, specialists, skills and datasets they used. A DOI for each published study, deposited on Zenodo, is planned.

Formats

Source code on GitHub