What is known, and what goes unread.
That is the problem. It sits in a district paper nobody clipped, in a working paper nobody cited, in the ninety minutes of a podcast nobody transcribed, in a thread that went past at two in the morning and was gone by breakfast. The work has been done. It is just scattered across a hundred publications, in a hundred formats, under a hundred bylines, and there is nobody whose job it is to go and put it together.
So I am going to try.
There is so much about this country that could be counted and never is. There is so much that is counted, and published, and argued over on television every night. And then there is the third kind, which is the one that got me: the thing that was counted, and written up, and put out into the world, and read by nine people.
I keep walking into it. A figure that would change how you think about an entire state, sitting in table 4.2 of a PDF on a ministry site that goes down every other Tuesday. A piece of reporting from a district nobody sends reporters to, filed once, never followed up. A researcher who gave four years to a question and got one paragraph in one newspaper on a Sunday. Somebody who actually knows what they are talking about saying the true thing at minute sixty-three of a podcast, where it will die.
None of this is a failure of research. India is researched, patiently and well, by more people than you would guess. It is a failure of assembly, and of attention. That is a much smaller problem, and it is one that one person with enough stubbornness can actually work on.
What it gets built from
An article has a beginning and an end. You write it, you file it, it goes up, it does its numbers for two days, and then it rots quietly in an archive while the thing it described carries on moving without it. I have read thousands of those and written my share. I would like to try something else.
A dossier here is an ongoing document. It opens when there is enough to say and after that it is never finished, only more or less current. When a new survey lands, it changes. When somebody publishes the paper that knocks out half of it, it says so, and says what it used to say, and when it changed and why. Every part of it carries its own date and version. You can watch the record think.
The point is not to be first. The point is to still be right in four years, and to show my working while I get there.
Every dossier answers
condition · dossier
A record of how India measures women's participation in the labour force, what that measurement leaves outside itself, and what the numbers in circulation actually rest on. It opens on one article and one disagreement inside it.
32.8%
of working-age women counted as in the labour force
International Labour Organization, 2023, as given by Ideas for India
Read updated 5 Sep 2026
Some of the most important things about this country have no number attached to them, because nobody has gone and counted. Not disputed, not contested, not measured badly. Simply absent.
Those absences are published here as records in their own right, each one saying what it would take to fill it and who would have to go and do it. A gap only goes up if knowing the answer would change how you understand the subject. Absence is the finding.
Almost none of this is my reporting. It rests on somebody else’s labour — a reporter who went to the district, a researcher who gave years to one question, someone who made a podcast nobody paid them for — and they get their names on it, beside the words, with the way back to the original. I quote a little and link a lot, and where a summary will do I summarise and send you to the real thing, because the original is better than my summary and always will be.
I build it with language models, and I would rather tell you than have you work it out. They read more than I can read and they draft. They do not decide what is true. Mechanisms may come from a model, particulars never do, and a figure with no source you can open yourself does not go up.
The attribution rules in full, including what gets quoted and what gets summarised, and what a model is allowed to touch.
What that means in practice
1 open 2 seeds 2 edges
Started 5 September 2026 at a desk in Bangalore, with nothing in the record. On day one the whole thing is backlog. That is the honest place to begin.
The letter. What opened, what changed, what turned out to be wrong, and which gap closed. Fortnightly, and short.