Aggregated representative price
A market-day aggregate of variety modal prices; it is not a statistical mode.
Key evidence
Observed data stays separate from estimates.
The method validates and aggregates source variety rows before any historical analysis or forecasting.
Dataset source through27 Jul 202627 Jul 2026
Aggregated representative price uses arrival weights only when all variety arrivals are available; otherwise it uses the median.
Features use only information available by each forecast origin.
One common comparison date prevents unfair forecast rankings.
Reporting consistency
How regularly the market reports during its measured active period.
Expected range
The interval is calibrated from earlier out-of-time errors. Historical coverage is not a guarantee.
Estimated amount after entered costs
Transport, commission, handling, quality, grade, taxes, spoilage, and payment timing are not assumed unless you enter them.
Statistical flag
Anomaly flags remain visible and are not automatic proof of an error.
Technical report
Detailed notes, interpretation, and operational boundaries
This section documents what the published evidence means, how it was produced, and where the interface deliberately avoids making a stronger claim than the data supports.
Acquire and identify
The pipeline reads official AGMARKNET 2.0 market price and arrival reports responses and records source identity before transforming any market row.
- The prepared window covers 2024-07-01 through 2026-07-27; the current refresh contains 168 source files.
- State, district, market, commodity, variety, unit, date, price, and arrival fields remain traceable through the prepared artifacts.
- Source retrieval time and dataset source-through date are separate so users can distinguish data recency from application build time.
Validate and normalize
Records are parsed into stable types, checked for valid price order and bounds, deduplicated, and normalized into consistent market identifiers and units.
- Prices are expressed as rupees per quintal (100 kg) and arrivals as metric tonnes in the public contract.
- Invalid records are excluded with reason codes; a statistical anomaly can remain published because unusual is not equivalent to incorrect.
- Missing values remain nullable and reporting gaps are measured rather than silently interpolated into observed history.
Aggregate, model, and calibrate
Variety rows become market-day observations, time-safe features are built, candidate point methods are evaluated, and residual intervals are calibrated.
- Every forecast target uses an explicit origin date, target date, lead time, and common comparison date.
- Selection folds and the locked holdout are chronological; random train/test splits are not used for the published time-series claim.
- Forecast point and interval methods are versioned with the history artifact that produced them.
Publish and explain
The final export creates a manifest, partitioned market JSON, evidence summaries, checksums, and static routes consumed by the web application.
- Observed data, forecasts, quality evidence, and definitions are kept separate so the interface can state which claim comes from which layer.
- Market pages expose source partitions and detailed tables; high-level charts never replace the downloadable evidence.
- Product copy uses qualified language such as represented, prepared, estimate, and empirical coverage to preserve the method's boundaries.