STAGING BUILD · OCTOBER 4, 2026

Market-first interface; forecast capability remains incomplete

Regional weather and soil correction — local, October 4

Colorado, Nebraska and South Dakota grain assessments now use 76 native NASA weather grid cells, each assigned only to its own historical crop-support cells. The maximum assignment distance is 38.41 km; over 99% of each state's historical grain weight is covered. Local crop stages and stress windows are calculated before aggregation. The map and outlook consume the same weather. NASA root-zone relative wetness is separate from field observations.

Actual USDA 2012–2025 harvest outcomes were checked using prior-year training only. For the September 25 / June 8 planting scenario, average errors changed from 4.44 to 4.34 bu/ac in Colorado, 3.91 to 4.06 in Nebraska, and 5.10 to 5.03 in South Dakota. The changes are small and uneven. Soil did not consistently improve on corrected weather alone. Colorado's current estimate is outside the training envelope; Nebraska fails the improvement gate. South Dakota remains experimental.

USDA's June 30 release estimates 500,000 U.S. proso acres planted in 2026. Current harvested acreage, yield and production are blank in that report. Historical map shapes are not 2026 field boundaries. Crop evidence · Actual-outcome checks. Revised outcomes and fixed historical crop geography limit the backtest; no prospective or field-scale accuracy claim is made. Other countries' yield models were not recalibrated. No deployment or access change.

Map presentation repair — local candidate

Rejected filled weather rectangles removed. Compact sample markers preserve readable geography, source data and North American access. Thirty current gridded samples and twelve retained reference comparisons are available; 150 additional cells remain pending provider capacity. Local international yield calibration remains unavailable.

Earlier international integration

Official area, production and yield estimates for 46 countries (2024–2026, September 11 source vintage); national comparison map; twelve regional ERA5 weather references in nine countries; January–September 2026 weather replay with same-period 2021–2025 comparisons; dated regional crop assessments. The existing U.S./Canada detail is preserved. Country boundaries are not crop polygons, regional weather samples are not acreage-weighted national assessments, and no new independently calibrated international yield model is claimed.

Local browser checks covered country selection, missing country weather, Namibia without a production estimate, historical-year health withholding, timeline playback, weather layers, U.S./Canada controls, and controlled 503 failures. The 390px mobile and 675px app-pane layouts were inspected. A hidden-map canvas initialization defect was repaired and the affected workflow rechecked. Source access, field-level health, complete global coverage and an advancing automated refresh remain unproven. This is a local candidate; publication requires separate approval.

Preserved U.S./Canada capabilities

Market question first, explicit uncalibrated 30/60/90-day probabilities.
State/province selections replace numbered clusters in the dropdown. The underlying historical-support geometry is preserved and remains subject to agronomic validation.
Stage-matched NASA POWER rain and heat comparisons for Colorado, Nebraska and South Dakota, using the existing weather support and weights.
Existing date-specific experimental yield checks, uncertainty ranges and seven-day model revisions are surfaced next to regional exposure.
Calendar reset corrected when moving between Canadian and U.S. selections; missing evidence clears prior charts and forecasts.

Earlier U.S./Canada evidence and verification

Local browser checks exercised horizon selection, regional selection, preplanting date withholding, latest-date return, Canadian missing calibration, country calendar reset, planting delay, unspecified cultivar, evidence-map expansion and additional-region disclosure. The 390px layout had no horizontal document overflow. A simulated HTTP 503 on required weather produced a clear unavailable-data state with disabled controls.

Focused Node checks pass: seasonal threshold logic, missing weather, no future weather inclusion, maturity exposure stop, thermal-progress matching, planting-delay sensitivity and region-surface clipping. These checks demonstrate software behavior; they do not establish agronomic or price-forecast validity. The older check-app.mjs adapter is stale and is not used as proof of this interface.

Source provenance

Historical comparison method and source hashes · Release asset hashes

Not proven

Forward market probabilities, a validated observed crop-health classification, causal crop damage, current-season national/global supply, current stocks/demand/trade balances, calibrated Canadian yield and independently validated local growing-region geometry remain unfinished. Weather comparisons reconstruct earlier seasons from recovered data; they are not a reconstruction of what was published in real time. U.S./Canada weather can be replayed through September 25; international reference weather through September 28. Grower quotes and price-report assets are excluded from this general site build. A verified U.S. farmer access layer is not yet implemented.

No domain switch, audience expansion, scheduled update, billing or trading capability is included. The page remains a staging prototype.

Observed station context — October 4

The existing shared IEM ASOS collector was used across all 31 U.S. and Canadian networks intersecting Millet support; the previous per-network cap is not used. It found 633 stations. 610 passed the configured 60% temperature-coverage and pinned GISit plausibility screens (588 U.S.; 22 Canadian); 23 failed coverage. Screened stations map within 80 km of 252 of 255 historical growing-region outlines. Three U.S. other-use areas in Oklahoma, Montana and Wyoming have no nearby screened station.

The optional map layer shows one hourly observation per UTC day; it is not daily maximum/minimum temperature or a rainfall total. Canadian hourly precipitation is omitted. Station records are map context only, not proof of field conditions and not yield-model inputs. IEM says the archive receives very little quality control; Millet screens temperature plausibility only and does not verify sensor calibration or field representativeness. View station coverage and source limits.

Global NOAA station context — October 4

The optional map overlay applies the pinned GISit sampling and daily screen to NOAA GHCN reports. The screen found 213 eligible sites globally; the international payload contains 88 passing sites outside the U.S. and Canada, associated with 42 historical modeled regions in 13 countries. The U.S./Canada IEM station layer remains separate. Only 33 of those 88 international sites passed all three variables (precipitation, minimum and maximum temperature). It is limited to one selected SPAM growing-region support area, follows the exact map date, and uses only actual variables reported that day. The sampled candidates do not cover 746 of 805 support polygons; absence is not evidence of normal weather. I checked the remainder of the top-40 inventory pool for those international gaps: five additional unscreened candidates were found, all in Russia, but the NOAA daily query failed because DNS resolution was unavailable and a direct browser request was blocked. They remain excluded; coverage did not change. No other unique candidates appeared in that pinned top-40 pool; these five still await a successful NOAA screen. This does not prove no other stations exist outside that pool. The source query ends July 8, 2026. SPAM polygons describe modeled 2020 support, not verified current plantings. Station reports are not inputs to crop-health, yield or market estimates.

Local browser verification: Argentina / Chaco, July 1 shows one NOAA station report and opens its dated temperature values and method limits; October 1 shows the July 8 cutoff and no marker; disabling the overlay clears markers. No browser console errors were recorded. This proves the local preview only. No Site publication or GitHub change was made. Global station coverage and integrity index.