The boundary is the asset. Everything else is a query against it.
A field technician walks one plot six times a season. That walk was bought and paid for the moment the lifecycle capture was made mandatory. What comes out of it is not a map, it is a table of observations with a device class, a timestamp and a shared edge on every row. Seven product lines read that table differently. None of them needs a new field visit, and each one fails in a different way if the table is wrong.
Yield estimation. A forecast is area times yield. Multi-stage capture attacks the area term directly, and germination coverage tells you how much of the forecast is evidenced rather than assumed.
| Attribute | |||||||
|---|---|---|---|---|---|---|---|
| Areaplot.stageAreaAcres[0..5] | |||||||
| Timecapture.capturedAt | |||||||
| Devicecapture.deviceClass, capture.sigmaM | |||||||
| Topologyplot.sharedEdgeIds, plot.sharedBoundaryPct | |||||||
| Cropplot.cropMix, crop.contractRateInrPerQtl | |||||||
| Provenancecapture.hdop, satellites, method, flags | |||||||
| Identityplot.id, plot.surveyNo, plot.farmerId | |||||||
| Synccapture.syncSessionId, syncLagHours |
1. Yield estimation
Decision. How many quintals is this book going to deliver, and how sure are we. A forecast is area times yield. Multi-stage capture attacks the area term directly, and germination coverage tells you how much of the forecast is evidenced rather than assumed.
At 0 the plots are independent and the book band collapses to 0.10%, which is the number a naive portfolio model prints. Shared handsets, shared technicians and shared satellite geometry on the day all push it up, so nothing here assumes independence for you.
Applies only to plots with no germination capture. At 100% the forecast assumes they all came up.
| Stage | Plots with evidence | Area used | Forecast | Book band | Mean plot band |
|---|---|---|---|---|---|
| Contract only | 0 | 408.4 ac | 5,800 qtl | 6.64% | 11.9% |
| Onboarding | 159 | 396.1 ac | 5,597 qtl | 1.94% | 4.6% |
| Sowing | 177 | 393.3 ac | 5,580 qtl | 1.09% | 2.9% |
| Germination | 179 | 393.3 ac | 5,581 qtl | 0.87% | 2.3% |
| Pest / Disease | 179 | 393.8 ac | 5,592 qtl | 0.74% | 2.0% |
| Pre-Harvest | 179 | 392.2 ac | 5,569 qtl | 0.65% | 1.8% |
| Post-Harvest | 179 | 392.2 ac | 5,568 qtl | 0.59% | 1.6% |
Forecast quintals are measured area times the fixture’s own contracted yield per acre and crop mix. The per-plot band is the engine’s one-sigma area error, divided by the square root of the captures folded in and multiplied by that plot’s shared-edge topology factor. Plot bands are combined under equicorrelation, so the book band is sqrt((1-rho)·sum sd^2 + rho·(sum sd)^2) at the correlation you set, not silently at zero. The contract-only row applies the same combination to the programme’s published contracted-versus-measured RMS of 11.9%, because a paper area has no capture behind it. The systematic over-statement beside the chart is measured directly and does not depend on the correlation at all, which is exactly why it is the harder problem. Establishment is a scenario input, not observed data. All captures in this world are simulated.
A contract-farming operator does not get to choose whether the technician walks the plot. The lifecycle capture is mandatory, the visits are already budgeted, and the technician is already standing there. That spend is the platform’s fixed cost, and it is sunk before the first application is written.
What the seven product lines add is not more field work. Each one is a different read of the same rows, so the marginal cost of the next application is engineering, not logistics. That is also why quality at the point of capture compounds: a handset refresh does not improve one report, it improves the input to all seven at once. The reverse holds too. A capture stream with a one-sided bias in it poisons every application downstream, and only the shared-edge check can see it.
The sequencing follows from the clocks rather than from ambition. Input recommendation decides at sowing with two captures. Procurement decides weekly and cares more about freshness than precision. Settlement and insurance decide once, at the end, and can afford to wait for all six. Build in clock order and each application ships against evidence that already exists.
SYNTHETIC DATA. Every company, village, club, farmer, plot boundary, capture, contract, rate and settlement in this file is fabricated for a product study. No real person, holding or agreement is represented. District and state names are real Indian administrative names used only to anchor the belt; village centroids are approximate synthetic points inside it and do not correspond to any actual settlement. Figures on this tab come from that world and from the deterministic engine in src/lib/traverse. The five claims the study takes as inputs rather than results - satellite separability at 10 m and 3 m, dual-frequency accuracy bands, handset availability from 2022, the 15 to 25 percent variance band, and the benchmark rupees per acre - are carried as study inputs in the fixture and are never rendered here as engine output. External regulatory and scheme facts are sourced inline on the module that uses them.