What we are collecting, and why it is not answers
A system trained on conclusions learns to assert. A system trained on derivations learns to derive — and can then meet a question nobody anticipated. So what we need is not the ruling but the road to it: the reasoning shown, the sources named, in the form a teacher would use with a student who asked.
And where an answer depends on a principle scholars differ over, we want that said. A system trained on reasoning with all the ikhtilaf removed learns something worse than ignorance — fluent, confident derivation with no sense that a question is one where people disagree. The records that name what a disagreement turns on are among the most valuable we receive.
Two roles
What works best is a teacher together with one or more of their students, contributing as a pair or a small group. A teacher’s answer and a student’s working are different artefacts and we need both.
Writing records
You write the records. Each one is a single question, the answer, and the sources it rests on.
We need a spread rather than a sample — every level from muqaddimat through sath to dars-e-kharij, and every subject, because the method is what transfers and it looks different in each. Usul reasoning is not mantiq reasoning. A student’s working is as wanted as a teacher’s answer: it shows where the difficulty actually is, which is the thing a teacher stops being able to see.
- You can state a question precisely, and give an answer that shows its reasoning rather than only its conclusion.
- You are willing to cite. A claim without a source is not usable to us, however true it is.
- You are honest about the limits of your own knowledge. “I know the answer but not the derivation” is a genuinely useful thing to say, and we would far rather have it than a confident gap.
- You write in the language you study in. Translation is our problem, not yours.
You do not need to be advanced, fluent in English, or technical in any way. If you can attend a lesson and read what comes back, you can contribute.
Reviewing them
You read other people’s records before they count for anything. This is the role that decides whether the whole thing is worth having.
Reviewing is not proof-reading. You are asking three questions: is this reasoning sound, is the citation real and does it say what it is claimed to say, and would a scholar recognise this as the method rather than as a plausible imitation of it? If an answer cites a page and nobody opens the book, what the model learns is not a wrong page number — it learns that producing a confident-looking citation is what an answer does, and applies that to every book it has never seen.
- You can judge a derivation and not only an answer — realistically dars-e-kharij level, or a teacher.
- You are willing to reject, and to say why in a sentence the writer can act on.
- You are comfortable saying “this is right, but for the wrong reason” — the most valuable and least comfortable review there is.
We expect reviewers to come from among the people already writing. Somebody who writes carefully for their first batch is exactly the person we ask, so this is not a separate application.
What you would actually do
The shape of the work, and how far it has got. We would rather be exact about this than have you find out later.
- Designed, not built yet
You record your own lesson. Entirely optional, it beeps audibly when it starts, and it shows clearly that it is running.
- Designed, not built yet
It is transcribed, in the language of the class.
- Designed, not built yet
Translated, if you want it in another language.
- Designed, not built yet
A draft record is prepared from the transcript, mostly filled in already — so you are correcting rather than writing from a blank page at eleven at night.
- Working now
You complete the form and check it. Nothing counts until you have: a record is never made from your lesson without you, and you decide what it says.
- Working now
Someone else reviews it — approved, held for an unverified locator, sent back for changes, or refused, always with a reason you can act on.
We are not giving a date for the second half. This project has given a date before and missed it, and we would rather say nothing than do that twice. What exists today is the form and the review, and they work.
Who owns what
Your recording is yours.
It lives on our servers so you never have to keep it safe, and you can download it whenever you like — for life. And if daftar ever winds down, you get your material out before it does. Whether or not you are still contributing. Scheduled maintenance aside, it is always there.
Your transcript is yours, on the same terms.
Downloadable at any time, in the language of the class and in any translation we have made of it. These are your class notes, written for you.
The questions and answers you write are yours.
You wrote them. Your name stays on them permanently — on every record — and they stay yours for life. Your access to daftar lasts as long as daftar does, and if it ever winds down you get them out before it goes.
What we own
The assembled, reviewed dataset — the corpus as a whole, in the form a model can be trained on. That is the one thing that is ours, and it is not downloadable by anyone. If another institution in the hawza wants access to it, that is a conversation we want to have.
What you get
- Your own lessons back — recorded, transcribed, translated if you want them, yours to keep and to download, for life.
- Your name on your work, on every record, permanently.
- A tool built to help you first. If this feels like a favour you are doing us, we have built it wrong: it is supposed to fall out of a lesson you were attending anyway.
What we would ask at the start
Roughly ten records, so we can look at them together and talk about what makes a good one. This is a conversation and not a test — the first batch from anyone is where we learn how to explain ourselves better, and it is at least as much about our instructions as your work.
After that, as much or as little as suits your study. There is no quota and no deadline. Twenty careful records are worth more than two hundred hurried ones. We also check work at random, and everyone is told so at the start — not because we expect dishonesty, but because a dataset nobody checks is one nobody can vouch for, and vouching for it is the entire point.
What we are not asking for
- Fatawa, or anything that reads as a ruling.
- Anyone’s opinion presented as the position of a marjaʿ.
- Work that competes with your studies.
The limits this is built inside
There are six things this system deliberately does not do — it issues no fatwa, replaces no hawza study, and declares no winner where scholars differ. They are design constraints written into the material itself rather than disclaimers added afterwards, and a record that breaches one is refused at review. They are set out in full on the About page and worth reading before you decide.
Taking part
daftar is invite only, and there is no sign-up. If you teach or study in a hawza and this sounds like something you would want to do, tell us who you are. We are not onboarding widely at the moment, so this may sit for a while.