Pre-Registered Predictions — Aug 12 USDA Print
Written: 2026-08-08 (four/five days before the releases; §"Aug 12" added late evening 2026-08-08 after the retraction in research/10 §3, still pre-release). Grade against NASS Crop Production (Wed Aug 12), the AMS CO Direct Hay Report (ams_2905, Thu Aug 13), and the CPC ENSO Diagnostic Discussion (Aug 13). Context and reasoning: research/10 §5–7.
NASS Crop Production, Aug 12 — first official 2026 hay forecast (survey basis Aug 1)
These test the production thesis directly. Calibration from the stored analog files: in deteriorating droughts the Aug print runs HIGH vs the eventual final (Aug 2022: 2,001k printed, 1,769k final, +13% — NASS carried June's acreage while abandonment accrued); in recovering years it runs low (Aug 2025: 2,145k vs 2,356k final, −9%). Our estimate of the 2026 final is 1.5M tons; 2026 is a deteriorating year, so we expect the print above our final estimate.
| # | Item | Point prediction | Range (HIT if inside) | Reasoning |
|---|---|---|---|---|
| P1 | CO alfalfa production, forecast | 1.75M tons | 1.55–2.00M | 1.5M final est. × ~1.15 deteriorating-year overshoot; acreage stickiness pushes up |
| P2 | CO alfalfa harvested acres in the print | 620k | 580–660k | 2022 carried June intentions (690k) exactly; 2025 cut below intentions (620→550); 2026 damage is visible enough to force some cut from 660k |
| P3 | CO alfalfa yield, forecast | 2.85 t/ac | 2.60–3.10 | Farmer Aug-1 reports track conditioned yield well (2022 Aug yield 2.90 = final exactly); NE CO normal-water share props the state number |
| P4 | CO all-hay production, forecast | 2.40M tons | 2.15–2.70M | P1 + other-hay ~600–650k (dryland devastated: pasture 76% VP+P) |
| P5 | US all-hay production vs 2025 | down y/y | direction only | >40% of US hay acres in D2+ |
P1 is the headline test. Scoring per the protocol below; P5 is directional (HIT/MISS only, half weight). Grading note: the Aug 12 print and our 1.5M final estimate are different targets — if the print lands ~1.7–1.8M and January later lands ~1.5M, both the prediction and the thesis are right; if the print lands below 1.5M, our thesis is likely too optimistic, not vindicated.
Scoring: each row is HIT (inside range), NEAR (within one notch), MISS. Point = single most likely value.
AMS Colorado Direct Hay Report (week ending ~8/14)
| # | Item | Point prediction | Range (HIT if inside) | Reasoning |
|---|---|---|---|---|
| 1 | SE CO 3x4 alfalfa — does the $250 FOB ask convert to a printed trade? | Yes, $250 | trade prints $240–260 | Floor rising under it (KS grinding +$50, OK re-dry); asks at import parity get paid in a rising market |
| 2 | SE CO Premium/tested small-lot channel | $300 | $285–315 | $270 (Jun) → $300 (Jul 30); heat is degrading the marginal supply of tested hay; retail channel sticky |
| 3 | NE CO dairy trades (Prem/Supreme) | $305 | $290–320 | $300–310 printed Jul 30; W Nebraska +$20–40 in 2 wks pulls the corridor up; $17 milk caps it under ~$320 |
| 4 | NE CO new-crop delivered-dairy offer stack | $235 | $225–245 | Was $220–230; offers chase trades in a rising tape |
| 5 | Confirmed hay volume (tons, statewide) | 3,000 t | 1,500–4,500 t | Jul 30 printed 2,415 t; biweekly cadence alternates heavy/light; demand "Good to Very Good" |
| 6 | Straw floor | $150–160/ton FOB | holds ≥$150 | Report said floor "getting set at $150"; short supply |
| 7 | Narrative demand word for new-crop hay | "Good to Very Good" | at least "Good" | Unchanged conditions; drought + early feeding pressure |
| 8 | Narrative drought %: CO D3–D4 coverage cited | 45–55% | 40–60% | Was 50.76% (Jul 28 map); SE improved on monsoon but NW/metro deteriorated; Aug 1–7 statewide dry+hot |
| 9 | Any feedlot-scale (500+ t) alfalfa contract, if one prints | $240 delivered | $230–250 | June benchmark was $225 delivered; KS-landed parity moved ~+$15–25 since |
CPC ENSO Diagnostic Discussion (Aug 13)
| # | Item | Point prediction | Range | Reasoning |
|---|---|---|---|---|
| 10 | Advisory status | El Niño Advisory continues | — | 100% JAS odds per IRI |
| 11 | Latest weekly Niño-3.4 anomaly cited | +1.4°C | +1.2 to +1.6 | +1.2 on Jul 9; events strengthen Jul→Oct; 23/26 models head ≥+2.0 at peak |
| 12 | "Very strong" OND probability | 85% | 78–92% | Was 81% and building |
| 13 | Persistence through NH spring 2027 | ≥95% | ≥90% | Was 97% |
Bonus: NASS Agricultural Prices, Aug 31 (July prices)
| # | Item | Point prediction | Range | Reasoning |
|---|---|---|---|---|
| 14 | CO alfalfa, July (state avg) | $220/ton | $212–230 | Two straight +$20 months ($170→$190→$210); momentum intact but base rate says sequential jumps shrink; cash quotes support continued climb |
| 15 | CO all-hay, July | $217/ton | $209–228 | Tracks alfalfa −$2/3 |
| 16 | KS alfalfa, July (state avg) | $140/ton | $130–155 | Cash grinding hit $200–210; the lagging state avg ($125 Jun) starts closing the gap |
Honesty note: #1 and #9 are the economically meaningful calls (they test the floor-rising thesis); #2–4 test channel spreads; #10–13 test the spring-2027 relief prior. AMS SE coverage is thin — a week with zero SE prints voids #1–2 rather than counting as a miss (record it as NO-PRINT).
Scoring protocol (fixed 2026-08-08, before any release)
- Outcomes: HIT (observed value inside the stated range), NEAR (numeric: within 5% of the nearest range edge; categorical: adjacent label), MISS (anything else), NO-PRINT (the report contains no observation of that item at all).
- NO-PRINT applies only when the market segment is absent from the report. If the report shows SE CO activity but the predicted event doesn't occur (e.g., asks listed, no trade), that grades #1 as a MISS, not NO-PRINT.
- One-sided ranges (#6, #13) can only score HIT or MISS — they are weak predictions and are labeled as such; they carry half weight in any summary.
- Correlated groups are graded but not summed as independent tests: report results in three clusters — (A) economically central: #1, #9, #14; (B) channel spreads: #2, #3, #4, #16; (C) regime checks: #5–8, #10–13, #15 — plus overall interval coverage (% of ranges containing the outcome) and mean absolute point error for the numeric rows. No single "accuracy %" headline.
- Grading record: for each row, cite the exact report line/figure used. Never edit the prediction rows after a release; grades go in a separate section appended below with the release identifiers.
- Freeze: this file is committed to git before the releases; the commit hash is the tamper-evidence.