For CSR teams, NGOs and assessment agencies
What did the programme actually change?
Most CSR reports show a number going up after a programme started. That number usually would have gone up anyway. Counterfact compares the districts, schools or villages the programme reached with similar places it didn't reach yet, and reports what the programme itself changed, with an honest range and a public, tamper-proof record.
Why now
Impact assessment is required
Companies with an average CSR obligation of ₹10 crore or more over the three preceding years must have an independent impact assessment of each project with an outlay of ₹1 crore or more, at least a year after it finishes. The reports go to the Board and are attached to the annual CSR report.
Before-and-after isn't evidence
Enrolment, yields and incomes rise for many reasons: monsoon, prices, other schemes. Without a comparison, a report can't separate the programme from the trend, and boards increasingly ask.
Phased roll-outs are natural experiments
Programmes usually start in some districts first. The places that start later are the comparison. Counterfact turns that into an estimate in minutes, using public data where possible (UDISE+, HMIS, crop and price data).
What you get
- A clear answer per outcome: what the programme changed, a likely range (9 in 10), and a verdict in plain words: worked, hurt, or can't tell yet.
- Every check written out: fake start dates, leaving out each comparison area, and what was compared with what.
- A verified receipt for every result in Counterfact's public record, anchored on the Bitcoin blockchain, so the Board and auditors can see it wasn't changed after the fact.
- Register the next programme before it starts in the change registry, so next year's assessment is measured against goals fixed in advance.
Counterfact is the analysis engine. Rule 8(3) asks for an independent agency: we work with your appointed agency, or provide the analysis for your agency's report. Ranges are honest: on 1,200 real test cases our 9-in-10 ranges held 90% of the time, and they say "can't tell yet" when the data can't tell.