What a district actually thinks — with the evidence attached.

JanPulse interviews residents in their own language, maps where opinion groups agree and where they split, and writes a brief in which every factual sentence is tied to the respondent evidence behind it and machine-checked before delivery. It is built for policy buyers who have to defend a finding in public, not just circulate a report.

What you receive

A map, not an average. Opinion groups, the statements that bridge all of them, and the statements that divide them.
Confidence on every figure. Agreement rates carry a credible interval, sample size and the group-level breakdown — no bare percentages.
A cited brief. Every factual clause carries one to five evidence references, and a checker recomputes every number.

A coverage report first. What your question set already has an answer to, how old that evidence is, and what genuinely needs fieldwork.
A reproducibility record. Snapshot, run, codebook and instrument versions, so a reviewer or auditor can re-derive the result.
A consent register. Who agreed to what, per respondent, including whether their words may be quoted or reused.

What it costs

Engagement	Indicative fee	Time	What it covers
Coverage diagnostic	₹1.5–2.5 lakh	2 weeks	Your question set run against the existing corpus and public data: what is already answered, what is stale, what fresh fieldwork is required, and a costed fieldwork plan.
Single study	₹3–15 lakh	6–8 weeks	Instrument design, fieldwork management in one or more languages, opinion map, cited brief, methods annex. Fee varies with sample size, districts and languages.
Multi-wave programme	₹12–20 lakh / year	3–4 waves	Repeated waves with matched cohorts, trend analysis, and a subscription to query your own corpus between waves.
API / assistant access	₹75k–1.5 lakh / yr	from week 2	Your analysts query the corpus through their own AI tools; answers return with citations. Metered at roughly ₹2,500 per 1,000 queries.

Indicative fees for discussion, not a quotation. Fieldwork costs — respondent incentives, enumerator time, translation — are billed at cost with the invoices attached. Fees are per defensible report, never per interview: we are not selling volume, we are selling a finding you can stand behind. The avoidable cost we remove first is duplicate fieldwork — in the demonstration corpus roughly half of a proposed twelve-question instrument already had prior evidence, some of it stale.

How the evidence is checked

2,031

participants in the public conversation our method was demonstrated on

896

statements clustered into groups, bridging and divisive

3 of 3

planted errors caught by the citation checker (consent, numbers, causal claim)

Method demonstrated on the Bowling Green conversation released as open data by the Computational Democracy Project (CC BY 4.0), 226,148 votes. A public dataset, not a JanPulse client study.

What we do not claim

Not a random sample. Respondents are recruited through partner communities and opt in. Every report names the sampling frame and where it may not generalise.
No map below the floor. Under roughly 30 respondents per group we report that no stable group structure exists instead of drawing one. Small groups are suppressed, not merged away.
No invented figures. Numbers in a brief are copied from evidence units and recomputed by code. Where a scorer has no measured accuracy, claims resting on it are labelled thin.
No synthetic people. Synthetic data is used only to demonstrate interfaces, and always labelled as such.
Consent, then erasure. Consent is captured per respondent — quoting and reuse included — re-checked at the point of use, and erasure is honoured with a receipt. Respondents are pseudonymous and their answers are held apart from contact details.

Next step

One question set, two weeks, ₹1.5–2.5 lakh: what is already known, what is missing, what a full study would cost. If the diagnostic does not change your fieldwork plan, do not commission the study.