Pre-Registered Predictions — September 2026 Releases
Written: 2026-08-27 (frozen before every release it targets; the commit is the tamper-evidence). Grade against the AMS CO Direct Hay Report (Thu Sep 10), the CPC ENSO Diagnostic Discussion (Sep 10), and NASS Agricultural Prices for August (~Sep 30). Context and reasoning: research/13.
Calibration compliance (L7): every numeric range below uses half-width ≥ the calibration.json floors measured from our own 2/10 record — price ±$65, volume ±$1,694 abs, anomaly ±1.3°C, probability ±10 pts. The ranges look embarrassingly wide. That is the point: they are as wide as our measured error says they must be.
New this cycle (judgment-layer grading, HANDOFF §12): each numeric row records its mechanical ANCHOR (persistence/base rate/last cash print) and the judgment ADJUSTMENT separately, so over time we can grade whether the judgment layer beats the anchor at all.
NASS Agricultural Prices, ~Sep 30 (August prices)
| # | Item | Anchor | Adj | Point | Range (HIT if inside) | Reasoning |
|---|---|---|---|---|---|---|
| 17 | CO alfalfa, August (state avg) | $210 (June print, persistence) | +25 | $235 | $170–300 | Two straight +$20 months through June; cash prints Aug ran $285–335; state avg lags cash 4–8 wks (L2) but momentum is strong |
| 18 | CO all-hay, August | $207 (June, persistence) | +23 | $230 | $165–295 | Tracks alfalfa −$3/5 |
| 19 | KS alfalfa, August (state avg) | $125 (June, persistence) | +30 | $155 | $90–220 | KS cash grinding $246–260 delivered; state avg closing the gap slowly |
Note: rows #14–16 (July prices, grading Aug 31) were frozen 2026-08-08 in predictions-2026-08-13.md and are unchanged by this file.
AMS Colorado Direct Hay Report, Sep 10 (week ending ~9/11)
| # | Item | Anchor | Adj | Point | Range | Reasoning |
|---|---|---|---|---|---|---|
| 20 | A SE CO FOB-farm alfalfa trade on a per-TON basis prints again (any tonnage) | base rate: printed 1 of last 4 CO reports | — | Yes | categorical | Demand pull (OK "calling all hay producers") + the 8/27 $325 print shows SE sellers now transact |
| 21 | IF a SE CO FOB-farm per-ton alfalfa trade prints: weighted-avg price | $325 (8/27 print, 88 t) | −10 | $315 | $250–380 | Small-lot premium may fade; regional premium complex $287–335; NO-PRINT if no such trade |
| 22 | NE CO dairy channel per-ton alfalfa trade (FOB or delivered), weighted avg if prints | $335 (8/27, 1,000 t delivered) | −5 | $330 | $265–395 | Milk futures fell ~$1 → some ceiling softening; freight up; NO-PRINT if none |
| 23 | Confirmed hay volume (tons, statewide) | ~1,900 (median of last 4: 1,306/1,338/2,415/2,734) | +600 | 2,500 t | 800–4,200 | Harvest-season cadence alternates; demand "Good to Very Good" pulls volume |
| 24 | Narrative demand word for hay | "Good to Very Good" (last 3 reports) | — | at least "Good" | categorical | Drought + early feeding pressure unchanged |
CPC ENSO Diagnostic Discussion, Sep 10
| # | Item | Anchor | Adj | Point | Range | Reasoning |
|---|---|---|---|---|---|---|
| 25 | Advisory status | El Niño Advisory (100% JAS per IRI) | — | continues | categorical | — |
| 26 | Latest weekly traditional Niño-3.4 anomaly in wksst9120.for at discussion time (metric: OISSTv2.1 weekly, NOT the relative anomaly, NOT the monthly index) | +2.6 (wk of 8/19) | +0.1 | +2.7°C | +1.4 to +4.0 | Three weeks plateaued at +2.6/+2.7; events typically keep strengthening Aug→Oct; the ±1.3 floor is our own measured miss |
| 27 | "Historic event" probability (OND RONI ≥ +2.5°C exceeds all El Niños since 1950) as stated in the discussion/slides | 69% (Aug 13) | 0 | 69% | 59–79% | No contradicting data in the Aug 24 deck; subsurface steady, weekly anomaly plateaued |
Scoring protocol
Unchanged from predictions-2026-08-13.md §protocol (frozen 2026-08-08): HIT (inside range) / NEAR (numeric: within 5% of the nearest edge; categorical: adjacent label) / MISS / NO-PRINT (segment absent from the report — an ask without a trade is a MISS only where the row predicts a trade). Conditional rows (#21, #22) grade NO-PRINT if their condition doesn't occur. Clusters: A economically central: #17, #20, #21 · B channel spreads: #19, #22 · C regime checks: #18, #23–27. Report interval coverage and MAE from scripts/scorecard.py, never hand-typed (L8). Never edit prediction rows after release.
Source specs
Every numeric row's machine-readable source spec is frozen in data/current/prediction_sources.json at the same commit as this file (L8). Rows #17–19 auto-grade via grade_predictions.py (esmis_agpr, month pinned to AUGUST). Rows #20–27 are graded by hand against the cited report lines, then verified by grade_predictions.py --check.