Methodology · how this stays honest
So we built the checking in from the start. Here is exactly how a contribution moves from a blank in a database to a real, accepted fix — and why a failed attempt is worth keeping too.
Every contribution goes through the database's own official channel — a GitHub pull request, a curator's inbox, a submission form. A maintainer or the community accepts it, not us. We bring better evidence, better organized; we don't grant ourselves write access.
AI does the finding and drafting, but a human verifies before anything is submitted, and the database's own reviewers decide whether it lands. For contributions only a community can judge — like whether a spoken word sounds right — other people check it, not a machine grading its own homework.
Most science throws away its dead ends: failed approaches die in lab notebooks, AI sessions evaporate. Lacuna logs every attempt — what was tried, by whom, whether it worked, and the check behind it. Failed tries are kept on purpose so the next person doesn't re-walk the same dead end.
For every attempt we record the steps, the evidence, and the outcome — success, dead end, or still open. You don't have to take our word for any claim on this site; the registry is the receipt.
Something specific and checkable is missing from a database scientists use — a recording, a record, a correction.
AI does the heavy lifting — searching, drafting, formatting. The person brings what only a human can: a voice, a camera, a careful judgment call.
Before anything is submitted, it's verified — by the person, and then by the database's own maintainers or community.
If accepted, the fix lives in the real database for good. Either way — accepted or not — the whole attempt stays on the record.
Lacuna is early. The registry holds 8 documented gaps so far, and one of them has been merged into a real database, start to finish. That's a small number on purpose: every contribution starts exactly this small, and we'd rather show you a true one than an impressive one.
You bring your own AI key to do the work — the assistant runs on your account, not ours. We'll always tell you what works today and what's still being built.