Hackathon build · 8–9 Aug 202628 states · 8 UTs · 165 districtsSeeded demo data

Nobody decided this.

Every season, Indian farmers choose what to sow with almost no information about what the market will want at harvest. Last year’s high price pulls everyone into the same crop. Six months later they arrive at the mandi together, holding a perishable product, and the price collapses. Six months after that, the consumer pays triple.

Sowline is the demand-side mirror that does not exist yet: it normalises what the country will buy into kilograms per district per week, subtracts what is actually coming, and shifts the difference backwards through the crop calendar into a sowing date a farmer can act on — capped, so the advice cannot itself become the next glut.

The cobweb, drawn from the seeded mandi record

Every peak is a farmer ruined. Every trough is a consumer paying triple.

Price and arrivals for tomato, three years, weekly. They move against each other with almost mechanical reliability — and the mechanism is simply that thousands of people made the same reasonable decision at the same time, alone.

202420252026
Modal price, ₹/kg — national mean of reporting mandisMandi arrivals, kg — total across reporting mandis
19.36 Mt
shortfall found

across 4,066 district-weeks under 0.8 coverage

3,571
glut warnings

district-weeks above 1.2 coverage — farmers told to stay out

15,892
acres released

148 opportunities, each capped at 80% of its gap

1,670
gaps skipped on purpose

sowing window closed or agronomically impossible — shown, not swallowed

Three surfaces, one engine

Who this is for

What this is not

The limits, stated before you find them

  • Not a buyer commitment

    An indicative price band is not an offer. Nothing here is binding on anyone, and the card says so.

  • Not an ML forecast

    Demand projection is last year's same week × a growth factor × a hardcoded festival table. A rule a judge can read beats a model a judge must trust. Labelled v0 baseline throughout.

  • Not a supply registry

    Declaration coverage is a few percent. Everything else is inferred from mandi arrivals, and the coverage number is shown on every screen rather than buried.

  • Not identity infrastructure

    Phone number only. No Aadhaar, no land records. Where AgriStack Farmer ID would plug in is marked, not faked.

Consent, privacy and the asymmetry problem