Ideas that work.

The farmer with the most grievances usually has the fewest ways to be heard. A sampling frame that only reaches people who are easy to talk to will miss that farmer.

NCRB counts the crimes that were registered. A higher figure can mean more crime, easier reporting, or both, so a state's rank on it says little about safety.

The Special Intensive Revision asked people to prove they belonged on the roll with documents many poor citizens do not have. That reverses the presumption of inclusion the roll was built on.

A hand-drawn village map still identifies the village after the labels are removed, because the geometry stays. Decide first what to remove, then what to swap, and what must never be published.

Inequities hide inside an average improvement. An indicator that cannot be split by gender, caste and geography reports on a country that does not exist.

A grant is a bet on a person before it is a bet on a project. Fund the person who has the good question, even when someone else has the prettier logframe.

A pre-mortem asks how a project could fail before it has launched. Run it before the launch, because the likely cause of failure is usually already in the plan.

The people whose work a metric measures can always game it. What matters is whether the gaming produces the underlying change or only the appearance of it.

Cardboard placards go into the rain and spray-paint is walked over by Monday, so what survives is what somebody decided to keep. Every archive takes a position on what the future will remember.

The last published caste enumeration was the 1931 census. The next count is in 2027, ninety-six years later.

Power shows up in who gets time on the calendar before it shows up in the budget. Read the calendar first.

An indicator is a promise about what you will pay attention to. Choose only the ones you can afford to keep tracking.

A public archive that claims certainty can be used against the people it records. An archive that says openly what it can and cannot verify, and puts the safety of contributors ahead of a complete record, acts as a witness to what happened.

When authentication fails at a ration shop or a bank counter, the cost falls on the person in the queue, who comes back another day or goes without. A system that is free at the point of use should publish its failure rates and pay for a human fallback.

A dashboard shows what a team has decided to notice. An honest dashboard also lists what the team has decided to leave out.

A small gap between what a woman wants and what she has can mean that she reached her aspiration or that the aspiration narrowed. One number cannot tell the two apart, so they have to be measured separately.

The 2017-18 consumption survey was conducted and never released, so the published interval between rounds is eleven years. Collecting a number and withholding it leaves the same gap in the public record as never collecting it.

Resilience looks different in a Mumbai slum than in a Toronto suburb. A scale that cannot tell the two apart is measuring the assumptions of the people who wrote it.

Some tools have to ship at the speed of the news cycle that created the need for them. A civic archive built in November for a protest that ended in July arrives after the placards are gone.

A Theory of Change is a hypothesis, and a team that tests it as one can learn from it. A team that treats it as a contract will spend its time defending it.

A trial measures what reaches the field, so running one before delivery is stable can produce a null result that gets misread as the idea failing.

A training has worked only if participants do something differently on the following Monday. If nothing changes in their work, the training was a performance.

If a report needs a glossary, many readers will stop at the glossary. Write the glossary first, then write the report without using any of the words in it.

The speech gives the promise. The Revised Estimate for the year just ending shows what was spent. The gap between the two is where underspending shows.

A free printable framework redistributes something of value. It lets a practitioner run, on a Friday afternoon, the audit she could not afford to commission.

Most briefs ask for strategy, and many of them want somebody to blame later. Read the brief for the second need and price the work for the first.

GDP counts the consultant who advises on a kitchen renovation and leaves out the woman who has cooked in that kitchen for forty years. The figure measures what the economy has agreed to notice.

An RCT tells you something true about one moment, much as a photograph does. You cannot drive on it, cook with it or send a child to school on it.

India has never run a national survey of household income, so its Gini is calculated from consumption, which understates the gap at the top. Until income is measured, nobody can say how unequal the country is.

Sixteen years without a census means ration entitlements and representation are still set on 2011 numbers. Every year of delay decides who gets counted for every scheme that uses them.

India has not conducted a census since 2011. Every policy decision that rests on that data is planning for a country that has since changed.

When a country does not measure its own inequality, it has chosen not to know about it. That choice should be described as a choice.

Revising a voter roll is a political act. The names that drop out between two revisions are the finding, because nobody is counting them.

A civic tool should be judged by the things its builders decide to refuse. Decide early what you will not build, even when you could.

Twenty-four states have published final rolls. Against them the removals come to 8.87 crore, about 12 in every 100 electors, and those rolls grew by 1.22 crore names after their drafts. A draft figure and a final figure answer different questions, so a headline should say which one it quotes.

Voices at the margin stay there because the rooms are built too small for them. Publishing can make the room bigger.

A cash transfer changes behaviour for as long as the cash lasts. An incentive built into the welfare system changes behaviour for as long as the government stands.

A learning platform built for one particular first-grader has design decisions that can all be defended, because every widget is there for a real reader who needed it. When strangers start using it, it is more honest than most platforms built for a market.

In none of the 92 incidents has the archive found a written order authorising force. Every field without a source reads unknown, which makes the gaps visible.

A daily deadline enforces the discipline that a weekly one only claims to have. What you write at 6am is the story you actually had, before there was time to revise it into something safer.

Most ToC reviews are friendly because the reviewer hopes to be invited back. The useful reviews are the ones where the reviewer raises the awkward point in the first half-hour.

Read a child the same book twice in one sitting. The second reading is the one that does the teaching.

Most household surveys treat caste as a control variable. When the research question is about caste, treating it as a control means refusing to look at the answer.