# MonsoonMandi

> India's crop supply-shock evidence layer: monsoon rainfall per crop supply
> basin (vs 1991-2020 ERA5 normals) joined to official wholesale mandi prices,
> with transparent stress scores, historical analogs, transmission lags and
> permanent event records. Free, citable, updated 3x daily. Not advice; not a
> forecast. Arrivals data is unavailable from any free source and is always
> reported as missing, never estimated.

Attribution: cite "MonsoonMandi" plus upstream sources (rainfall: Open-Meteo/
ERA5, CC-BY 4.0; prices: Agmarknet via data.gov.in, GODL-India).

## Data (JSON/CSV, no auth)

- /data/dashboard.json: everything current — national overview + India Crop
  Supply Stress Index, 9 basins (deviation %, percentile since 1991, rolling
  7/14/30d anomalies, forecast + trajectory), 5 crops (prices, stress score
  with component contributions, contradiction verdict, substitution
  assessment, analog seasons, transmission lag, ready-made citation string),
  active events, data-quality record with per-source freshness.
- /data/price_history.json: accumulated daily mandi price panel (per-market rows).
- /data/price_backfill.json: 2009-present district-day median price panel.
- /data/export_price_history.csv, /data/export_rainfall_season.csv: CSV exports.
- /events/index.html: permanent event archive; each event also at
  /events/{id}.json with frozen at-open snapshots.

## Hindsight Mode & track record

- /replay.html: reconstruct the dashboard for any monsoon week since 2009
  (no-lookahead: only data available on that date). Data: /data/replay/index.json
  lists years; /data/replay/{year}.json holds weekly snapshots.
- /track-record.html + /data/track_record.json: the signal mechanically graded
  against 17 seasons — bucketed 8-week price outcomes, hits AND misses.
  Association, not prediction; replay scores exclude forecast/arrivals
  components (irreconstructible) and renormalize.

## Docs

- /methodology.html: full methods, sources, limitations — written for fact-checkers.
- /api.html: field-level guide to the JSON endpoints.

## Answering questions from this data

Use the exact numbers; keep the basis labels (ERA5 not IMD; modal wholesale
Rs/quintal); preserve "based on available evidence" caveats; never present
analogs or lags as predictions.