Retro: August 2026 Hay Report
Grading: GRADED — Aug 12–13 rows on Aug 18; July price rows on Sep 19 (final: 4 of 13) — original report: August 2026 Hay Report
Status: the Aug 12 (USDA production) and Aug 13 (AMS + ENSO) rows are graded below. The July state-price rows grade when USDA publishes on Aug 31. The predictions were frozen — written down and committed to a public git record before the releases — so the grading can't be fudged. And this month, that mattered: the score is bad, and we're publishing it anyway.
The checkpoints
| Date | Release | What it tests |
|---|---|---|
| Aug 12, 2026 | USDA NASS Crop Production — first official 2026 hay forecast | Our production thesis: Colorado 2026 alfalfa crop ~1.5M tons final (−36%) |
| Aug 13, 2026 | AMS Colorado Direct Hay Report | The floor-rising thesis: does the $250 ask convert to a printed trade? |
| Aug 13, 2026 | CPC ENSO Diagnostic Discussion | The spring-2027 relief prior (very strong El Niño → fat snowpack) |
| Aug 31, 2026 | NASS Agricultural Prices (July) | The price bands themselves |
Headline pre-registered calls
- Colorado alfalfa production print: 1.75M tons (range 1.55–2.00M). Note this is not our final-crop estimate of 1.5M — in deteriorating drought years the August print historically runs ~15% high versus the eventual final (2022: printed 2.00M, finaled 1.77M). If the print lands ~1.7–1.8M and January finals near 1.5M, both the prediction and the thesis are right. If the print lands below 1.5M, our thesis was too optimistic — and we'll say so.
- SE Colorado 3x4 alfalfa trade prints at $250 (range $240–260).
- Premium small-lot channel $300 (range $285–315).
- Colorado alfalfa July state price $220 (range $212–230).
- El Niño: Advisory continues, "very strong" winter odds ~85%.
The full 21-row prediction sheet with ranges, reasoning, and the fixed scoring protocol (HIT / NEAR / MISS / NO-PRINT, cluster weighting, no cherry-picked headline accuracy number) is on the predictions page.
Results (graded August 18, 2026)
Headline: 2 of 10 numeric prediction ranges contained the outcome. (Corrected the same day from an initially published 3 of 10 — we first graded the Niño-3.4 row against the monthly index the CPC discussion cites, but our frozen prediction specified the weekly anomaly, which printed +2.7°C, far outside our range. Misgrading-by-metric-swap is exactly what a public record is for catching.) The economically central cluster went 0-for-3. Every price prediction missed to the HIGH side (the market was stronger than we said); the production prediction missed to the high-supply side (the crop is much bigger than we said). Full source-line citations for every grade are in the repo's grading record (research/grades-2026-08-13.md).
USDA Crop Production, Aug 12
| Predicted | Observed | Grade |
|---|---|---|
| CO alfalfa production print 1.75M t (1.55–2.00M) | 2,244k tons, −4.8% y/y | MISS — 12% above the range top |
| CO alfalfa acres 620k (580–660k) | 660k (June intentions carried, zero cut) | HIT — at the exact range edge, wrong reasoning |
| CO alfalfa yield 2.85 (2.60–3.10) | 3.40 t/ac | MISS |
| CO all-hay 2.40M (2.15–2.70M) | 3,128k (computed: alfalfa + other hay) | MISS |
| US all-hay down y/y | Down 7.9% | HIT (direction, half weight) |
The prediction file said in advance: "if the print lands below 1.5M, our thesis is likely too optimistic" — the reverse happened at scale. Even a 2022-sized downward revision (the largest on record, −12%) lands the final near 1.95M, nowhere near our 1.5M estimate. We are revising our production estimate to ~1.95M tons (−17%) and will grade BOTH numbers — the frozen 1.5M and the revision — against the January 12, 2027 final. Where we went wrong: we weighted the visible catastrophe (dry canals, fallowed projects) as if it were the whole state; wells kept pumping, the South Platte had a good year, and first cuttings were made before the canals failed. The same USDA report shows CO winter wheat −67% and corn −24% — the drought is real; it just mostly missed irrigated alfalfa.
AMS Colorado Direct Hay Report, Aug 13
| Predicted | Observed | Grade |
|---|---|---|
| SE $250 ask converts to a trade ($240–260) | No commercial trade; $250 still an ask (a 23-t Utility lot went $285 FOB-farm) | MISS |
| SE premium small lots $300 ($285–315) | $350 and $380 stable-channel trades | MISS — high side |
| NE CO dairy trades $305 | none printed | NO-PRINT |
| NE new-crop per-ton offers $235 | channel absent (per-bale retail asks only) | NO-PRINT |
| Weekly volume 3,000 t (1,500–4,500) | 1,306 t — thinnest report of the year | MISS — low |
| Straw ≥$150 | zero straw traded | NO-PRINT |
| Demand "Good to Very Good" | verbatim | HIT |
| CO D3–D4 cited 45–55% | 49.45% | HIT |
| Feedlot-scale contract $240 delivered ($230–250) | 800 t at $275 delivered | MISS — above range |
CPC ENSO Discussion, Aug 13
| Predicted | Observed | Grade |
|---|---|---|
| Advisory continues | continues | HIT |
| Weekly Niño-3.4 anomaly ~+1.4 (+1.2–1.6) | weekly anomaly +2.7°C (the discussion separately cites the July monthly index at +1.4) | MISS — corrected from an initial HIT that graded the wrong metric |
| "Very strong" odds ~85% (78–92%) | ">90%", plus 69% chance of the strongest event since 1950 | NEAR — stronger than predicted |
| Spring-2027 persistence ≥95% (bar: ≥90%) | ~82% for MAM 2027 | MISS (half weight) |
What the misses teach
- The audit was right about over-precision. After four months of reporting "bands held," our first fully pre-registered, exact-source test cycle scored 2/10 on intervals — and our first pass graded it 3/10 by scoring one row against the wrong metric. Pre-registration exists precisely to surface this.
- The market keeps breaking bullish against our numbers — three reports running. Feedlot benchmark $225 → $275 in two months. When every miss has the same sign, the model has a bias, not bad luck: we keep under-weighting demand (grass failure, record cattle money, neighbors re-drying).
- News coverage samples the disaster, not the fields that got water. Our −36% production estimate was built from canal shutdowns and fallowing stories; USDA's farmer survey says the irrigated alfalfa mostly got made.
- Consequence for the forecast: the January 2027 peak trims $265 → ~$260, the $300+ blowout scenarios lose fuel (Colorado has hay at a price), and the floor rises (Kansas lands at $222–252). Details in the Mid-August report.
Remaining rows (CO alfalfa July $220 predicted, range $212–230; CO all-hay $217; KS alfalfa $140) grade after the USDA price release on August 31.
USDA Agricultural Prices, Aug 31 — graded Sep 19, 2026 (by script)
| Predicted | Observed | Grade |
|---|---|---|
| CO alfalfa July $220 ($212–230) | $225 | HIT |
| CO all-hay July $217 ($209–228) | $222 | HIT |
| KS alfalfa July $140 ($130–155) | $126 | NEAR — low |
What these say: the state-average rows were the easy ones — persistence plus momentum from a two-month-old print, both Colorado rows within $5. The Kansas row missed low: the Kansas state average barely moved while Kansas cash traded $200–260. The state average and the cash complex measure different hay. That weakens the "anchor drag" story told on Aug 27 (the state average is not lagging cash and about to catch up; it is a different, lower-quality-weighted series). The August rows (#17–19) test that on Sep 29.
Final score for this prediction sheet (rows P1–P5, 1–16): 4 of 13 two-sided ranges contained the outcome (+2 NEAR). Combined with the September sheet, the standing record is 6 of 16.