Retro: Late-August 2026 Hay Report
Grading: GRADED β Sep 10 AMS + ENSO rows on Sep 19; August price rows on Oct 2 (all 11 rows final) — original report: Late-August 2026 Hay Report
Status: complete. Rows #20β27 (AMS Colorado Sep 10; CPC ENSO Sep 10) graded Sep 19; rows #17β19 (USDA August state prices, released Sep 29) graded Oct 2 by script. The predictions were frozen and committed on Aug 27 with machine-readable source specs; scores are computed from the ledger by scripts/scorecard.py, never typed.
Results (graded September 19, 2026)
AMS Colorado Direct Hay Report, Sep 10
| Predicted | Observed | Grade |
|---|---|---|
| A SE farm-gate per-ton alfalfa trade prints again (YES) | None β the Southeast section had one per-bale ask and a prairie-grass lot | MISS |
| β¦its weighted price $315 ($250β380) | β | NO-PRINT |
| NE dairy trade $330 ($265β395) | none; the only NE per-ton trade was 2,000 t of Good 3x4 at $260 FOB-farm | NO-PRINT |
| Weekly confirmed volume 2,500 t (800β4,200) | 8,500 t β 4,500 t of it organic San Luis Valley contracts | MISS β high |
| Demand "Good" or better | "Good to Very Good" | HIT |
CPC ENSO Discussion, Sep 10
| Predicted | Observed | Grade |
|---|---|---|
| Advisory continues | continues | HIT |
| Weekly traditional NiΓ±o-3.4 +2.7 (+1.4 to +4.0) | +2.7 (the 02SEP line, the newest available on Sep 10) | HIT β point exact |
| "Historic event" odds 69% (59β79) | 75% | HIT |
At the Sep 19 grading, coverage across both graded sheets stood at six of sixteen, and the economically central cluster at 1-for-5.
What the misses teach
- We predicted the target would be visible, and it wasn't. A new six-year ledger of Colorado reports (242 of them, rebuilt from USDA's file server) shows the grading target β a SE FOB-farm Good-or-better large-square trade of 100 tons or more β prints in about 2.5% of reports. Local calls are now categorical with the base rate stated.
- The Aug 27 spot anchored on a print tied for the highest Southeast print in six years (88 tons at $325). The largest Colorado commercial print of 2026 came in $65 under it. The September report revises the nowcast down to $270 (first published as $280, corrected the same day after an independent review found a $170 Oklahoma lot left out of the comparables).
- Weekly volume is not forecastable at any width we'd publish. Dropped as a prediction row.
- Pinning the metric works. Last cycle the ENSO rows went 1-for-3 with the metric in prose; this cycle 3-for-3 with it specified to the file and line.
USDA Agricultural Prices, Sep 29 (August prices) β graded Oct 2
| Predicted | Observed | Grade |
|---|---|---|
| Colorado alfalfa $235 ($170β300) | $235 | HIT β point exact |
| Colorado all hay $230 ($165β295) | $231 | HIT |
| Kansas alfalfa $155 ($90β220) | $127 | HIT β but the point missed by $28 |
Interval coverage: 9 of 19 two-sided numeric ranges contained the outcome (+2 NEAR), standing record. The economically central cluster is 2-for-7.
- Judgment vs plain persistence. Each of these three rows recorded its mechanical anchor (June prices, the latest available when the rows were frozen on Aug 27) and our adjustment. Our judgment beat that anchor on average ($9.7 vs $17.0), winning big on Colorado and losing on Kansas: we bet the Kansas state average would follow Kansas cash up; it moved $1. The next Kansas row is pure persistence. Details: grading record.