Southeast Colorado alfalfa price forecasts, graded in public.

Pre-Registered Predictions — October 2026 Releases

Written: 2026-09-19 (frozen before every release it targets; the commit is the tamper-evidence). Grade against the AMS CO Direct Hay Reports of Sep 24, Oct 8, Oct 22; the CPC ENSO Diagnostic Discussion of Oct 8; NASS Crop Production Oct 9; NASS Agricultural Prices Oct 30 (September prices). Context: research/17. Rows #17–19 (NASS August prices, Sep 29) were frozen 2026-08-27 and are unchanged.

Design changes this cycle (L11): the local target prints in ~2.5% of reports, so local rows are categorical with the base rate stated, or conditional on a large print anywhere in Colorado. Weekly AMS volume is dropped (MAE 3,847 t). State-benchmark rows use the new state_price_usd_ton family at ±$30 (2× the worst measured state-row error, vs vintage-backtest h1 $6 / h3 $14); local price rows keep the ±$65 floor. Each numeric row records its mechanical anchor and the judgment adjustment.

AMS Colorado Direct Hay Report — Sep 24, Oct 8, Oct 22

#ClusterItemAnchor / base rateAdjPointRangeReasoning
28ASE CO FOB-farm per-ton alfalfa trade of ANY size prints in the Sep 24 reportprinted in 2 of 8 2026 reports (25%)NOcategoricalbase rate; nothing in the Sep 10 SE section but a per-bale ask
29ASE CO FOB-farm Good–Supreme large-square alfalfa trade ≥100 t prints in Sep 24 or Oct 86 of 242 reports (2.5%)NOcategoricalHIT if neither report has one; the 2022 analog's 1,000-t lot printed Nov 17
30AIF any CO region (non-organic) prints a FOB-farm/ranch per-ton alfalfa trade ≥500 t in Sep 24/Oct 8/Oct 22: tonnage-weighted $/ton across all such rows$260 (Sep 10 NE 2,000 t Good)+10$270$205–335Good-grade blocks dominate large lots; G/P blocks would print $290–310; NO-PRINT if none
31BIF a NE CO per-ton alfalfa trade with 'Dairy' in Freight/Use prints in those three reports: tonnage-weighted $/ton$335 (Aug 27 1,000 t delivered)−20$315$250–380Class III −$1.20 since Aug 27; NO-PRINT if none
32CDemand wording at least "Good" in BOTH Sep 24 and Oct 8last 5 reports "Good to Very Good"YEScategoricalNEAR if one report says "Moderate to Good"

NASS Agricultural Prices, Oct 30 (September prices) — family state_price_usd_ton

#ClusterItemAnchorAdjPointRangeReasoning
33ACO alfalfa, September (state avg)$225 (Jul) + rising-analog +10 = $235+5$240$210–2702018/2022 analogs +10 by Sep; cash complex $260–330 argues for the top half; Kansas says the state avg does not chase cash
34CCO all-hay, September$222 + 10 = $232+5$237$207–267tracks alfalfa −$3
35BKS alfalfa, September$126 (Jul persistence)+4$130$100–160three months of $200–260 cash have moved the KS state avg $1

CPC ENSO Diagnostic Discussion, Oct 8

#ClusterItemAnchorAdjPointRangeReasoning
36CAdvisory statusEl Niño Advisorycontinuescategorical
37CLatest weekly traditional Niño-3.4 in wksst9120.for at discussion time = the 30SEP2026 line (posted Mon Oct 5). NOT relative, NOT monthly.+2.9 (09SEP)+0.2+3.1°C+1.8 to +4.4events strengthen into Nov–Dec; ±1.3 is the family floor
38C"Historic event" probability (OND RONI ≥ +2.5 exceeds all El Niños since 1950) as stated75% (Sep 10)075%65–85%RONI outlook OND median 2.67; NO-PRINT if the sentence is dropped

NASS Crop Production, Oct 9 (October 1 hay forecast — verified to exist in the Oct 2023 and Oct 2024 reports)

#ClusterItemAnchorAdjPointRangeReasoning
39CCO alfalfa production, Oct 1 forecast (1,000 tons)2,244 (Aug 1 print)−442,2001,472–2,928NASS usually carries acreage; Aug–Sep heat and the Fort Lyon shutoff trim late cuttings; ±728 is the production family floor; NO-PRINT if the Oct report carries no hay table (check two ways)

Scoring protocol

Unchanged (frozen 2026-08-08, predictions-2026-08-13.md §protocol): HIT / NEAR (numeric: within 5% of nearest edge; categorical: adjacent label) / MISS / NO-PRINT. Conditional rows (#30, #31) grade NO-PRINT if the condition never occurs across their window. Clusters: A: #28, #29, #30, #33 · B: #31, #35 · C: #32, #34, #36–39. Scores come from scripts/scorecard.py. Rows are never edited after release. Source specs for every row are in data/current/prediction_sources.json at this commit (rows #33–35 auto-grade via grade_predictions.py, month pinned to SEPTEMBER; the rest grade by hand against the cited lines and are verified by --check).