How the product works

Official reports in, prepared evidence out.

The forecasting method is one step in a larger data product. It does not run inside your browser or a request-time service.

Dataset source through27 Jul 202690.3K prepared market-day records
01

Retrieve mandi reports

Fetch public AGMARKNET monthly market price and arrival reports.

02

Validate records

Check identity, dates, units, price order, duplicates, ranges, and market references.

03

Standardise fields and units

Normalise source names and publish prices per quintal and arrivals in tonnes.

04

Aggregate observations

Combine variety reports into documented market-day records without filling missing days.

05

Analyse historical patterns

Measure reporting coverage, gaps, recent movement, volatility, and seasonality.

06

Evaluate forecasting approaches

Score simple and complex methods only on later, unseen periods.

07

Publish prepared website data

Write a small manifest and state/commodity partitions for static delivery.

Key evidence

Inspect the evidence

Start with plain-language summaries, then open the exact quality, forecast, source, and methodology details.

01
SourcesAuthority, retrieval, licence, attribution, and non-endorsement.
02
Data qualityAccepted, rejected, corrected, stale, anomalous, and missing records.
03
Forecast reliabilityFuture-only evaluation, error measures, interval coverage, and limits.
04
MethodologyAggregation, time-safe features, selection, and publishing boundaries.
05
Architecture and operationsBatch pipeline, static website, refresh workflow, and failure boundary.
06
LimitationsCoverage, freshness, aggregation, forecasts, and real-market factors.