Futures Desk — Final Ruling: Data Sources & Advanced Metrics

Panel: Opus, Sonnet, Grok 3, Gemini 2.5 Pro Judge: Opus (via synthesis) Date: 2026-03-25 Grade: A- (strong consensus on foundation, Gemini truncated but aligned)

Consensus Points

1. Futures Prices Are THE #1 Missing Piece (UNANIMOUS — all 4)

Without current prices, you can't calculate term structure, crack spreads, metal ratios, fair value models, or any cross-asset signal. Yahoo Finance 15-min delayed via /v8/finance/chart/ endpoint is fine for daily signals.

Tickers: CL=F, NG=F, GC=F, SI=F, HG=F, ZC=F, ZS=F, ZW=F, ES=F, NQ=F, ZN=F, ZB=F Also add: RB=F (RBOB gasoline), HO=F (heating oil) — needed for crack spreads Frequency: Every 15 min during market hours, 4x/day otherwise

2. NOAA Degree Day / Weather Data (STRONG — Opus, Sonnet, Grok, Gemini)

All 4 panelists flagged this as critical for natural gas binary pricing. Population-weighted HDD/CDD forecasts directly predict weekly gas storage draws/injections.

CPC 6-10 day and 8-14 day temperature outlooks
Population-weighted HDDs/CDDs
Endpoint: https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/cdus/degree_days/
1 HDD above normal ≈ 1.1-1.3 Bcf additional gas demand per day

3. Treasury Auction Results (STRONG — Opus, Sonnet, Gemini)

Bid-to-cover ratio, indirect bidder %, tail (high yield vs when-issued). Single best leading indicator for yield direction over 1-2 weeks.

Endpoint: https://api.fiscaldata.treasury.gov/services/api/fiscal_service/v1/accounting/od/auctions_query (free JSON, no key)
Event-driven collection, within 30 min of auction close

4. Term Structure = Free Math Once You Have Prices (UNANIMOUS)

Pull multiple contract months from Yahoo (CLF, CLG, CLH for crude; NGF, NGG for nat gas). Calculate contango/backwardation curves. This is the #2 priority behind prices themselves.

5. Crack Spreads + Metal Ratios = Free Math (UNANIMOUS)

All calculated from price data, zero additional API calls:

3-2-1 crack spread: (2 x RBOB + 1 x heating oil) / 3 - WTI
Gold/silver ratio: GC/SI (mean ~70, >80 = recession fear, <60 = growth)
Copper/gold ratio: HG/GC (leading indicator for 10Y yields)
Store: absolute value, z-score vs 1/3/5yr, percentile rank

6. USDA WASDE Is High-Value (STRONG — Opus, Sonnet, Grok)

Monthly crop report. Ag binary markets have less sophisticated participants than energy = more edge.

USDA PSD Online API: https://apps.fas.usda.gov/PSDOnlineV2/api/ (free, no key)
Also: NASS QuickStats for Crop Progress
Opus moonshot: WASDE revision momentum — when USDA revises ending stocks in same direction for 3+ months, next revision continues ~70% of the time

7. Drop SHFE-LME Copper Arb (UNANIMOUS — all 4)

Requires expensive Shanghai data, not tradeable on Kalshi/Polymarket, timezone complexity. Skip.

8. Drop Central Bank Gold Purchases (STRONG — Opus, Sonnet, Grok)

1-3 month data lag. Fully priced by the time you see it. Read the quarterly WGC report manually, don't build a scraper.

9. Deprioritize OPEC MOMR (STRONG — Opus, Sonnet, Grok)

Monthly PDF, brittle to parse, key numbers leak via Reuters 1-2 days before release. EIA covers US supply better. Build only if trading Brent-specific contracts.

New Sources NOT on Original List (Panel Additions)

NY Fed Treasury Term Premium — INCLUDE (Sonnet)

Decomposes 10Y yield into expected rate path + term premium. Term premium spikes predict bond selloffs before yield levels move. Sharpest free signal for Kalshi yield binary markets.

CSV export from NY Fed site, daily

TGA + RRP Liquidity (Gemini, Opus)

Treasury General Account and Reverse Repo balances are massive daily liquidity drivers.

FRED series: WTREGEN (TGA), RRPONTSYD (RRP)
Already have FRED scraper — just add these two series

SOFR Futures Curve (Opus)

More liquid than Fed Funds futures at longer tenors. Calculate implied forward rates to price rate decision contracts.

CME settlements API, product SR3, daily

US Drought Monitor (Grok)

Weekly drought conditions directly affect crop yields. Leading indicator for WASDE revisions.

https://droughtmonitor.unl.edu/Data.aspx (JSON/CSV, weekly Thursdays)

FOMC Minutes Semantic Analysis (Opus moonshot)

Bag-of-words hawkish/dovish word count across consecutive minutes. When hawk_velocity > 0 but FedWatch prices cuts > 50%, one side is wrong.

Free text from federalreserve.gov, no LLM needed

ICE Brent COT (Sonnet, Grok)

CFTC COT only covers NYMEX/COMEX. ICE has its own COT for Brent crude.

https://www.theice.com/marketdata/reports/122 (weekly CSV)
Brent-WTI spread + positioning divergence is a signal

Calculated Metrics (Synthesized from all 4)

Tier 1: Build Immediately (from price data)

Inventory Surprise Score (Opus + Sonnet + Grok)
- surprise = (actual_draw - expected_draw) / std_52wk
- Build own expected draw from: rig count trend + seasonal pattern + degree days
- Sharper than API consensus because it incorporates real-time weather
COT Positioning Extreme + Momentum Divergence (Opus + Sonnet + Grok)
- net_spec_pctile = percentile_rank(managed_money_net, 156_weeks)
- When >90th pctile long + price falling = reversal setup
- Extend existing forex COT signal to commodities
CPI Surprise Probability Model (Opus)
- Inputs: ISM Prices Paid, AAA gas change, Cleveland Nowcast, 5Y breakeven, Michigan inflation, FAO food
- cpi_surprise_score = weighted_z_scores → logistic → P(CPI above consensus)
- When score > 1.0, CPI came in above consensus ~72% historically
Futures Curve Signal (Opus + Sonnet + Grok)
- Annualized roll yield, contango/backwardation z-score
- Deep backwardation = tight supply = price support = "above $X" underpriced
Auction Demand Score → Yield Direction (Opus + Sonnet)
- score = 0.4*z(bid_to_cover) + 0.3*z(indirect%) + 0.2*-z(tail) + 0.1*z(direct%)
- Positive = strong demand = yields fall, negative = weak = yields rise
Cross-Market Dislocation Score (Sonnet)
- dislocation = kalshi_implied_prob - fair_value_from_futures
- The master signal. Every other metric feeds INTO this.

Tier 2: Build After Phase 1

NFP Surprise Model (Opus)
- Inputs: initial claims, continuing claims, ADP surprise, ISM employment, SAHM
- Same logistic conversion as CPI model
Gold Fair Value Model (Opus)
- Inputs: real yield, dollar strength, VIX, COT positioning
- signal = weighted_z_scores → when high but gold hasn't moved → "gold above $X" underpriced
Supply Elasticity Score (Sonnet)
- Combines rig count delta + inventory buffer → single directional signal for energy
Event Risk Premium (Sonnet)
- premium = pre_event_kalshi_spread / normal_spread
- When > 2.0, Kalshi overcharging for uncertainty → sell the spread

Moonshot: Correlation Regime Breaks (Opus)

When normally-correlated assets simultaneously break correlation, prediction markets are still pricing the old regime. Track rolling 20d vs 60d correlations across crude/equities, gold/yields, copper/China, dollar/rates. When break_score z > 2.0, all binary contracts priced on historical patterns are mispriced.

What to DROP/SKIP

Item	Verdict	Reason
#13 SHFE-LME copper arb	DROP	Expensive Shanghai data, not tradeable on Kalshi
#10 Central bank gold	DROP	1-3 month lag, fully priced
#5 OPEC MOMR	DEFER	Monthly PDF, key numbers leak early, EIA covers US
#7 EIA DPR	DEFER	Monthly, too slow for most binary expirations
#15 Seasonal scoring	DROP	Well-known patterns already priced. Deviations captured by inventory surprise.
#9 COMEX/LME warehouse	DEFER	Only if actively trading metals binaries

KEEP: #1 (prices), #2 (WASDE), #3 (Crop Progress), #4 (Export Sales), #6 (STEO), #8 (term structure), #11 (crack spreads), #12 (metal ratios), #14 (ag spreads)

Build Order

Phase 1: Foundation (Week 1)

Yahoo Finance futures prices — 14 symbols (CL, NG, GC, SI, HG, ZC, ZS, ZW, ES, NQ, ZN, ZB, RB, HO), every 15min during RTH, 4x/day otherwise
Term structure — calculated from multi-month contracts, same frequency
Crack spreads + metal ratios — calculated from prices, daily
Add TGA + RRP to FRED pull — 2 extra series IDs

Phase 2: Sharp Anchors (Week 2)

Treasury auction results — Fiscal Data API, event-driven
NOAA degree days — daily forecast, weekly actuals
Kalshi macro market filtering — add CRUDE-, NATGAS-, CPI-, FOMC-, GDP-, NFP-, TREASURY- series to collector
SOFR futures curve — CME settlements, daily

Phase 3: USDA (Week 3)

USDA WASDE — monthly, PSD Online API
USDA Crop Progress — weekly Apr-Nov, NASS QuickStats API
USDA Export Sales — weekly Thursdays
US Drought Monitor — weekly Thursdays

Phase 4: Calculated Metrics (Week 4)

All 10 metrics from above — pure code, no new data collection
Cross-Market Dislocation Score (master signal)
Correlation Regime Break detector

Phase 5: Nice to Have

EIA STEO (monthly forecast)
ICE Brent COT
FOMC minutes semantic analysis
Ag spreads (crush, corn/wheat)
NY Fed term premium

Cron Schedule (All Times ET)

Entry	Schedule	What
futures-prices	Every 15min 9:30AM-4:30PM weekdays	Yahoo Finance 14 symbols
futures-prices-offpeak	6AM, 8AM, 6PM, 10PM weekdays	Off-hours snapshot
futures-term-structure	4:30PM weekdays (after prices)	Calculate contango/backwardation
futures-spreads	4:35PM weekdays	Crack spreads, metal ratios, ag spreads
futures-metrics	4:40PM weekdays	All calculated metrics
treasury-auctions	Event-driven (see schedule)	Within 30min of auction close
noaa-degree-days	8AM daily	HDD/CDD forecasts
usda-wasde	Monthly ~10th, 12:15PM	Pull immediately on release
usda-crop-progress	4:15PM Mondays Apr-Nov	Crop conditions
usda-export-sales	8:45AM Thursdays	Export commitments
drought-monitor	9AM Thursdays	Drought conditions

Key Architectural Point (Opus)

"Your most valuable pattern is: government data release → surprise calculation → probability model → compare to Kalshi price → trade the gap. Every data source should feed into this pattern. Don't collect data for its own sake. If you can't convert it into a probability estimate that maps to a specific Kalshi/Polymarket contract, it's not worth building."

Source: ~/edgeclaw/results/panel-results/futures-data-final-ruling.md