October Deep Dive — October 2, 2026

Status: CURRENT. Supersedes the research/18 grid. Companion files: research/grades-2026-10.md (grading), research/predictions-2026-11.md (rows #40–45, frozen today), research/21-confidence-review-2026-10-02.md (how sure we are, and why), OctoberHayReport.md/.html (share-ready), review_2026-10-02.md (the pre-publication review, unedited).

Method. Three parallel pulls (AMS hay reports for all seven states; water, rain, drought and ENSO; NASS and macro), all raw files saved under data/2026-10-02/ (ams/, wx/, macro/). Every load-bearing number was re-checked by the operator by a second method (§9). Then an outside reviewer read the draft against the saved files (L11 rule 9). It found 17 problems, three of them HIGH; all are fixed in this version and listed in §11. Pipeline update_weekly.sh ran; its only failures were the expected "grades file must state the new coverage" checks, fixed by grades-2026-10.md.

1. Bottom line

Southeast Colorado Good/Premium large-square alfalfa: about $275/ton at origin today, range $210–340, LOW confidence. No grading-target trade printed (no Southeast Good-or-better large-square sale of 100+ tons). Eight defensible ways of averaging the regional sales give $271–308, median $277 after the Oklahoma report posted that evening (§13; the morning calculation in §4, $262–308 / median $273.5, is superseded); we round to $275. That is $5 above the Sep 19 estimate, which is noise inside the range. The number is most sensitive to geography: northern sales (Colorado, Nebraska, Wyoming) average $283, Kansas and Oklahoma $239.

What changed is the weather, not the price. Lamar Airport had 3.18" in July, 4.47" in August and 4.50" in September (normal 1.03"; that includes 2.17" on Sep 30), then 0.96" on Oct 1. Bent County went from 61% severe-or-worse drought (Sep 8) to 0% (Sep 29 map, before the Sep 30 storm). The Fort Lyon Canal, dry Sep 7–9 and Sep 11–20, is running at 371–405 cfs. We have no measured skill at turning weather into price (L10). We still cut the winter lift because of it; that is a weather-driven judgment, logged as an exception (§7).

Grades: Colorado's August state prices came in at our points ($235 alfalfa exactly, all-hay +$1); Kansas missed its point by $28 but stayed inside a wide range. Standing record 9 of 19 ranges contained the outcome (+2 NEAR). We also missed a categorical call: we said no Southeast per-ton farm sale would print on Sep 24, and a 25-ton Fair lot did, at $250.

2. Grading (details in research/grades-2026-10.md)

Judgment vs persistence. The rows were frozen Aug 27; the only persistence baseline available then was June (210/207/125, the anchors written into the rows). Against it, judgment won: MAE $9.7 vs $17.0 — it won on Colorado (errors 0 and 1 vs 25 and 24) and lost on Kansas (28 vs 2), where we bet the state average would chase cash and it moved $1. July prices (225/222/126, published Aug 31, after the freeze) would have scored $6.7; that is a later-information benchmark, not a fair test of the call. (A first draft and the pre-publication review said July persistence was the fair baseline and "judgment lost"; a second review (codex) caught the date error.)

Partial windows (graded later): Sep 24 had no SE ≥100-t Good-or-better trade (#29), no Colorado ≥500-t farm alfalfa trade (#30; the 4,320 t was millet), no NE dairy alfalfa line (#31), and demand "Good to Very Good" (#32, first half).

3. The market, report by report (primary files in data/2026-10-02/ams/)

NASS August prices (released Sep 29), alfalfa $/ton: CO 235 (+10), NM 285 (+5), WY 230 (+10), TX 230 (0), OK 193 (+5), KS 127 (+1), NE 104 (+1), US 206 (+3). US all-hay $190 (+$6). July not revised anywhere.

4. The nowcast, built in the open

Governing rule (research/18): conventional alfalfa; completed per-ton trade; large squares (3x4/4x4) only; ≥100 t; price at origin (FOB, FOB-farm, FOB-dairy, FOB-stables as printed); Good grade or better (Fair/Good lots are a lower grade class and are shown separately); rounds, delivered prices and organic excluded. L11(2)'s older wording ("≥500-t lots in NE CO, WY, W NE, KS") is shown as a variant. Computed by scripts/comp_set.py, which reproduces both Sep 19 sets exactly (G/P $269.8 on 1,840 t; Good $268.7 on 4,050 t) before being trusted on new files.

(Morning calculation — superseded by §13 the same evening: spread $271–308, median $277.45; estimate unchanged.) Estimate: $275 = the rounded median of the eight ($273.5). It is not a directional judgment; the spread is $262–308. Good ~$270. Fragility: the governing anchor is $299.6 without the Oklahoma $170 lot, and the rolling set is $276.0 without it. By region, northern lots (CO/NE/WY, 6,450 t) average $283 and Kansas/Oklahoma lots (2,340 t) $239; which of those markets Southeast Colorado hay trades with matters more than any rule above. Lot-to-lot spread (tonnage-weighted SD) is $32, which by itself supports a ±$65 range for any single sale. Like-for-like, western Nebraska G/P slipped $18 and Wyoming's G/P rose $35 on a different bale and basis: no trend we can claim.

Fair/Good large squares (mostly Kansas, Sep 15): wa $203 on 4,260 t, $125–245. Including them, all large squares average $249.

Import parity into Bent County (market_parity.csv, $0.157/t-mi derived from DAT flatbed $2.59/mi linehaul, wk ending Sep 25, plus fuel at $6.38 diesel): Oklahoma 4x4 lands $211–287; Kansas dairy P/S $273; western Nebraska G/P $321–351; Wyoming G/P $382. No export netback is computed (no delivered price with a known destination printed).

Regime-change rule check (forecast_policy.json): no directly comparable trade printed, so it does not fire.

5. Weather, water, drought (primary files in data/2026-10-02/wx/)

• Rain, Lamar Airport (USW00003013, ACIS, no missing days): Jul 3.18", Aug 4.47", Sep 4.50" (normal 1.03"; includes 2.17" on Sep 30), Oct 1 0.96". La Junta AP Sep 6.11" (normal 0.84"); Rocky Ford Sep 4.02"; Las Animas Sep 3.63" plus 4.50" logged Oct 1 (single source).
• Fort Lyon Canal: dry Sep 7–9 and Sep 11–20; FLCC's own 1884 priority came into the river on Sep 21; telemetry 78 cfs Sep 21 → 160 Sep 30 → 405 cfs Oct 1 (daily mean); FLCC page this morning: headgate 371 cfs, "Run #5", Las Animas, Limestone and Lamar divisions open. Jun–Sep diversions ~26,000 AF (26,445 CDSS monthly; 25,969 telemetry) = 25–26% of the 2010–25 average (102,502 AF). Gotcha: CDSS daily "Total" records carry repeated placeholder values for Sep 7–9 (33.8 cfs) and Sep 11–Oct 2 (37.3 cfs); the monthly 26,445 includes them, so the two methods share inputs and their agreement is partly luck. Use telemetry for September.
• Reservoirs: John Martin 22,978 AF (CDSS telemetry, Oct 2 08:15; USGS lake elevation 3802.41 ft agrees) and rising, up from 19,412 on Sep 18. The USACE morning report, unreachable at first and retrieved later the same morning (wx/usace_johnmartin_2026-10-01.html), shows 20,228 AF at the end of Sep 30 (+578 AF that day): an earlier observation of the same rise, not a conflict. Pueblo 157,315 AF (Oct 1), down from 161,265.
• Rivers today: Purgatoire ~1,050–1,900 cfs (median 9); Arkansas at Lamar 1,800 cfs (median 17).
• Drought (USDM map valid Sep 29, before the Sep 30 storm): Bent County D1 48%, D2+ 0% (D2+ was 61% on Sep 8). DSCI: CO 277, KS 139, NE 151, NM 253, WY 270, all improving; OK 372 and TX 288, both worse (OK 92% D2+, 24% D4).
• Pasture, very poor + poor (Sep 27 vs Sep 20): CO 46 (52), KS 37 (39), NE 43 (57), NM 55 (57), OK 65 (67), TX 59 (60), WY 65 (67); 48 states 51 (53; 35 a year ago).
• ENSO: weekly traditional Niño-3.4 +2.9 (09SEP), +3.0 (16SEP), +3.1 (23SEP); the 02SEP line was later revised +2.7 → +2.8 (row #26 stays graded on the value available at discussion time, per its spec). El Niño Advisory; 75% "historic event" odds (Sep 10). Next discussion Oct 8; the 30SEP line posts Oct 5.
• Outlooks: 6–10 and 8–14 day lean dry in Bent County; October no lean; OND 40% wet; Seasonal Drought Outlook: removal in Bent County, improvement at Pueblo, Dodge City, Guymon, Amarillo, Oklahoma City.
• Snowpack: six Arkansas-basin SNOTEL sites at 0.0–0.4" SWE. Nothing to read until Nov 1 / Dec 1.

What the rain does and doesn't do (inference, not measured): it adds little 2026 hay (most districts are on their last cutting; first freeze at Lamar is typically mid-October). It does (a) green local pasture and wheat, which cuts local cow-calf hay demand this winter, (b) refill soil and start refilling John Martin for 2027, and (c) lower the grade of hay that was cut or stored uncovered. Oklahoma and Texas moved the other way. Net for SE Colorado price: two-sided.

6. Demand and costs (primary files in data/2026-10-02/macro/)

• Cattle: Sep 1 on feed 11,163k (101%); August placements 1,617k (91%), the lowest August since the series began in 1996; marketings 97%. Futures (Yahoo delayed quotes saved Oct 2 ~14:20 UTC, not settlements): feeders Oct $335.43, live cattle Oct $218.60.
• Dairy: Colorado milk cows 218k (211k a year ago), production +3.9%. Class III Oct $15.08, Nov $15.80, Dec $16.22 (same source). Weak milk, growing herd.
• Grain: Dec corn $5.00 (499.75¢, same source), down from $5.27½ on Sep 19.
• Fuel: US diesel set a new record $6.529 on Sep 21 (the $6.285 "record" cited Sep 19 was passed), then $6.382 on Sep 28. Rocky Mountain $6.407, a series high. L1: marked to market, not forecast.
• Fertilizer PPI (WPU0652013A): August 144.547 (prelim), −9.6% m/m.
• Trade (secondary, DTN): the US–China truce was extended from Nov 10 to Jan 10, 2027. US hay exports H1 −8% (secondary).

7. The grid (data/current/premium_forecast.csv, version "research/20")

L10 honesty note. Weather has not passed the evidence gate, and it still moved the Nov–Jan midpoint by about −$11 and the spring roll-off. The amended policy permits that only as a logged exception, and it is logged (forecast_adjustments.json). It is judgment, not a measured effect. forecast_preflight.py: OK. State-benchmark grid unchanged (Sep $240, Oct $245, Dec $255, Jan–Mar $260).

Falsifiers. Down: a Colorado ≥500-t farm sale at or under $240; Kansas/Oklahoma wheat-pasture reports after the rain; NE/WY tone going from "moderate" to "light". Up: any ≥500-t Good-or-better print at or above $300 in Colorado; Oklahoma's Oct 2 report still "calling all hay producers" with prices up; Colorado's September state price ≥$260 (Oct 30).

9. Verification table (claim | method 1 | method 2 | status)

10. What we could not get

The Kansas report expected Sep 29; the Oklahoma and Texas reports due today (not posted at 08:49 MDT); the USACE John Martin morning report (retrieved later the same morning; see §5); the 30SEP Niño-3.4 line (posts Oct 5); Horse Creek storage. Local private asking prices for the Arkansas Valley: searched, none published (Front Range asks $340–375). Kansas and Oklahoma are the two that could move the nowcast; research/21 sizes that.

11. Pre-publication review (review_2026-10-02.md) — what it found, what changed

12. Second review (codex, same day, after commit 358a693, before push)

Third review (codex, same day, after 7ecc22a): (1) the regenerated signals.html said "0 cfs in canal" — signal_dashboard.py summed the FLCC page's division PM column, which reads 0 until afternoon; it now reads the page's "Headgate N CFS" line (371 cfs) with the page's own update stamp (10/2/2026 7:15 a.m.). (2) "August-only rain" summed Aug 1 → today (20.30") and judged timing by day-of-month; now bounded to Aug 31 and timed by full date (6.67", two-station sum). Both were pre-existing generator bugs that this run republished. (3) Figure 08: marker size capped so large lots no longer hide the next row's label; local-sale label shortened. Archive revision 3; earlier revisions kept as -rev1/-rev2.

Fourth review (codex, same day, after 73533de): (1) the FLCC fallback summed the division "AM/PM In Use" table (lateral deliveries, not headgate flow) — removed; no headgate line now means "unknown". (2) the dashboard stamped the USACE John Martin storage (20,228 AF, end of Sep 30) as today; it now shows the report's own period date; that also explains the gap with the 22,978 AF CDSS reading (Oct 2 08:15). (3) the SE AMS price tile silently vanished because it only looked in data/<date>/hay/; it now finds ams/ams_2905_<date>.txt saves too and shows the $250 Sep 24 lot. (4) research/21 said both "0" and "about two" comparable local outcomes; it is zero. New tests/test_dashboard_parsers.py (7 tests) pins all three parsers to the saved pages.

Fifth review (codex, full dashboard audit, after 55686ef): nine findings, all in pre-existing dashboard code except one stale line in this report. Fixed: the dashboard price chart now plots statewide all-hay actuals under the v2 all-hay forecast (it plotted alfalfa); winter snowpack tile implemented (6-station index vs its 2015–25 median, labeled a proxy for the NRCS number that decides); ENSO parsed with entities decoded, status/issue date from the page, newest saved discussion as fallback, no hard-coded August text; rain tiles refuse to publish when >10% of days are missing; forecast chart limits come from the full low/high range; Bent County, diesel and Pueblo verdicts use one threshold function each, live and cached; CDSS call responses are validated and the "most senior" detail is built from the records; the John Martin limitation above is corrected (USACE retrieved later: 20,228 AF end of Sep 30).

Sixth review (three parallel audits: website, charts, price-math/grading scripts; after c871ae0): 35 findings across the three. Fixed: (1) a past grade value — row #2's recorded $365 missed a $300 lot across a page break; the true value is $342.3, grade still MISS; row #9's "contract vs trade" reading questioned. Past records are NOT edited: both are noted in the August retro only; (2) the production-estimate code checked a looser window (−15%..+2%) than LESSONS L4 states (−12%..0); code now matches L4, and a new estimate row statewide_rev2 = 1,975k (the L4 central) replaces rev1 (1,950k, now flagged outside L4; it and the 1,500k row are still graded as published on Jan 12); (3) NASS state rows #14–19 are calibrated as the state-price family by their data source, so easy state-average hits can't release the local cash-price floor; (4) market parity now lists every ≥100-t Good-or-better Kansas lot and the Wyoming line once (L11 rule 5); (5) the website's "inside our bands for all four graded months" line replaced by the ledger's 9 of 19; homepage "next checkpoint" now comes from the watchlist; frozen reports now cite dated copies of their charts; figures land under the heading they belong to; (6) the dashboard no longer shows the v2 model's runaway path (to $376) with a ±$7 band; (7) chart scripts no longer crash or mislabel on the next sale or release; (8) tests now pick the newest file by date, not file time. Still prose, not code: the regime-change and two-same-sign rules.

13. Evening update, Oct 2 (Oklahoma and Texas reports posted after the morning version)

• Oklahoma, Fri Oct 2 (data/2026-10-02/ams/ams_3095_2026-10-02.pdf, sha256 d711fe62…): "Demand for hay continues to be high. We have received some rain, but unfortunately it has been too late in the year… Oklahoma is still seeking rain." Alfalfa: Central Good/Premium 4x4 80 t $240 FOB (under 100 t); Northwest Premium/Supreme rounds 840 t $240 (rounds excluded); asks: NW Premium 4x4 $210, NW Good/Premium 4x4 $170 (ask), Central Fair/Good 4x4 $180. No ≥100-t Good-or-better large-square trade. Next report Oct 16.
• Texas, Fri Oct 2 (ams_2707_2026-10-02.pdf, sha256 e2627783…): "Hay prices and demand remain good"; drought "continued this past week across most of Oklahoma and Texas… Much of Texas and Oklahoma need to receive 20 to 25 inches of rainfall over a 3-month span to break out." Conventional alfalfa: asks only (ground $350 delivered-feedlot; Premium small squares $375 delivered).
• Kansas: still Sep 15 at 6:47 p.m.

Nowcast recomputed by comp_set.py with OK Oct 2 as Oklahoma's latest report:

Estimate unchanged at $275 (rounded median; spread $271–308). The governing variant jumping to $307.5 is a mechanical artifact of one state's thin report (two lots), not a price move; the rolling set, which still holds every lot, is unchanged at $271. The OK $170 lot reappearing as an ask says nothing about whether that lot was cheap hay or a one-off. Southern demand evidence strengthened (OK "continues to be high", TX drought intensifying, TX feedlot ground-alfalfa ask $350); local SE Colorado pasture is recovering. Two-sided, as in §5; the grid is unchanged.

Seventh review (codex, evening): stale morning numbers ($262–308 / $273.5) left in the site blurbs and research summaries — now show the evening $271–308 / $277 with the morning figures labeled superseded; the statewide chart's right edge and y-ranges now come from the data (survives the next release); the standalone October HTML embeds its dated chart copies; the website's NEAR count uses the scorecard's two-sided rule; prediction adjustments #40–#42 and #44 added to forecast_adjustments.json; dashboard note no longer calls the band a dry/wet scenario range; a test file's entry point moved below all tests.