ADDENDUM: Actionable Edge Mapping + Prediction Market Ladder

Weather Lock-In Desk — MERGED FINAL

Base: Claude Opus 4.6 (complete, unmodified) + Panel Additions from Gemini Pro, GPT-4.1, Grok-4-Fast

Panel review: Opus (base), Gemini Pro, GPT-4.1, Grok-4-Fast, DeepSeek V3 (no unique additions)

This addendum supplements the original Weather Lock-In spec. It does NOT replace the existing spec — it adds two critical layers: (1) mapping every edge to a specific trade on a specific platform, and (2) a systematic methodology for scanning prediction market temperature ladders for mispriced contracts.

CRITICAL DATA SOURCE RULE: For all weather data, NWS API is the ONLY trusted source for temperatures and weather observations. NEVER trust web search for temps. All temperature anchors, intraday observations, and forecast updates must come from api.weather.gov or NWS-derived products (NBM, HRRR, GFS ensemble). Web search results for temperatures are unreliable and prohibited.

TOPIC 1: ACTIONABLE EDGE OUTPUT

1. Edge-to-Trade Mapping Table

2. Actionable Output Format

JSON output sent via Telegram and to other desks:

{
  "edge_id": "WEATHER-20260315-CHI-LOCKIN-001",
  "desk": "weather_lock_in",
  "timestamp": "2026-03-15T16:45:00Z",
  "edge_description": "Chicago high temperature locked in at 71F (4.2F drop from running high of 75.2F, post-peak). Kalshi and Polymarket contracts above 73F are overpriced.",
  "signal_type": "lock_in_reactive",
  "city": "Chicago",
  "date": "2026-03-15",
  "running_high_f": 75.2,
  "current_temp_f": 71.0,
  "drop_magnitude_f": 4.2,
  "peak_time_passed": true,
  "confidence_level": 93,
  "lock_in_probability": 0.97,
  "risk_adjusted_edge": "14c after Kelly sizing",
  "platforms": {
    "Kalshi": {
      "mispriced_contracts": [
        {
          "contract": "Chicago high above 76F",
          "current_price": 18,
          "fair_value": 2,
          "edge_cents": 16,
          "edge_after_fees": 14,
          "action": "Buy NO at 82c",
          "liquidity_usd": 280
        },
        {
          "contract": "Chicago high above 78F",
          "current_price": 8,
          "fair_value": 1,
          "edge_cents": 7,
          "edge_after_fees": 5,
          "action": "Buy NO at 92c",
          "liquidity_usd": 150
        }
      ]
    },
    "Polymarket": {
      "mispriced_contracts": [
        {
          "contract": "Chicago high above 76F",
          "current_price": 20,
          "fair_value": 2,
          "edge_cents": 18,
          "edge_after_fees": 16,
          "action": "Buy NO at 80c",
          "liquidity_usd": 120,
          "note": "Lower liquidity than Kalshi — use limit orders, monitor slippage"
        }
      ]
    }
  },
  "time_sensitivity": "MEDIUM — lock-in signal is confirmed, edge shrinks as market catches up. Estimated window: 30-90 minutes.",
  "sharp_reference": "NWS observed high at station KORD: 75.2F at 2:23 PM CT. Current temp 71.0F at 4:45 PM CT and falling.",
  "weather_context": "Cold front passage at 3:15 PM CT. Wind shifted NW at 15 mph. Clear skies but thermal recovery impossible — sun angle too low.",
  "risk_factors": ["Station discrepancy: Kalshi uses KORD, verify same station", "Rare late-day temperature spike possible if wind dies"]
}

Telegram summary format:

WEATHER LOCK-IN: Chicago, Mar 15

High LOCKED at 75.2F (dropped 4.2F, post-peak)

Kalshi Edges:
  Above 76F @ 18c | Fair 2c | Edge 16c | Buy NO @ 82c | $280 liq
  Above 78F @ 8c | Fair 1c | Edge 7c | Buy NO @ 92c | $150 liq

Polymarket Edges:
  Above 76F @ 20c | Fair 2c | Edge 18c | Buy NO @ 80c | $120 liq

Confidence: 93% | Lock-in prob: 97%
Catalyst: Cold front, NW wind 15mph, sun angle low
Window: 30-90 min

3. Platform Priority Ranking

Summary: The Weather Lock-In desk is primarily a Kalshi + Polymarket play. Both platforms have temperature and weather ladder structures with multiple contracts at different thresholds. Both must be scanned for every signal. All engineering effort should focus on API integration with both platforms, NWS real-time weather data, and fast signal-to-trade execution on both.

TOPIC 2: PREDICTION MARKET LADDER STRUCTURE

1. Desk Data Mapping to Prediction Market Contracts

Note: All 20 tracked cities map to Kalshi and Polymarket temperature ladders. Kalshi typically has 10-15 rungs per city per day at 2F intervals. Polymarket availability varies by event and season but should be checked for every city where Kalshi has an active contract. This means 200+ individual contracts to scan per day across all cities on Kalshi alone, plus equivalent Polymarket contracts where available.

2. Sharp Reference Identification

3. Probability Curve Construction Method

Distribution type: Empirical + Normal hybrid (with Poisson and Empirical Bootstrap for non-temperature markets)

Unlike sports where the outcome is generated by a stochastic game, temperature forecasts have a well-studied error distribution:

• NWS forecast error distribution: Approximately normal with mean near zero (slight warm bias in some seasons) and standard deviation that varies by city, season, and forecast horizon.
• Typical forecast error SD for day-of forecasts: 2.0-3.5F for most US cities. Desert cities (PHX, LAS) have lower SD (~1.8F) in summer. Coastal cities (SFO, SEA) have lower SD (~1.5F) year-round. Transition seasons (spring/fall) have higher SD (~3.5-4.0F).

Distribution type by market:

• Temperature (continuous): Normal distribution. Mean = NWS point forecast (from api.weather.gov — no other source). SD = historically calculated RMSE of NWS forecasts for that location + month.
• Precipitation (binary/ladder): Bernoulli for single binary markets (p = HRRR PoP directly). For precipitation ladder rungs (e.g., "over 0.5in"), use truncated normal if the distribution is symmetric, or empirical CDF bootstrapped from NOAA historicals if the distribution is skewed (common for low mean precipitation). [PANEL: Gemini Pro, GPT-4.1]
• Precipitation counts / wind gusts (discrete): Poisson distribution anchored to NWS expected total. For precipitation ladder: lambda = NWS expected precipitation total. P(total > threshold) = 1 - Poisson_CDF(threshold - 1, lambda). [PANEL: Gemini Pro, GPT-4.1]
• Extreme tails / hurricane strike probability: Empirical bootstrap from historical analogues. Fit a Beta distribution to NHC ensemble spread for hurricane strike probabilities. Use bootstrap from 50+ historical tropical systems for tail estimation. [PANEL: Grok-4-Fast, Gemini Pro]

Historical data feeding the model:

• NOAA ISD/LCD database: Hourly temperature observations going back decades. Free download. Use last 2-5 years of data for each city.
• NWS forecast verification data: Compare NWS forecasts to actual observations over the same period. Calculate the empirical error distribution per city, per season, per forecast hour.
• Open-Meteo historical API: Free access to historical GFS, ECMWF model output for backtesting model accuracy.
• NOAA daily normals: Free CSV via climate.gov, 30+ years of US station data for distribution calibration. [PANEL: GPT-4.1]

How the sharp reference anchors the curve:

1. Anchor: Take the NWS point forecast for the daily high from api.weather.gov (e.g., 76F for Chicago). This is the only trusted source.
2. Spread: Use the historical forecast error SD for this city + season (e.g., Chicago in March: SD = 3.2F).
3. Adjustment: If multiple models disagree, weight the ensemble mean but widen the SD by the model spread (e.g., if GFS says 78F and ECMWF says 73F, the spread is 5F — increase SD by ~1F to account for increased uncertainty).
4. Construct CDF: P(high > X) = 1 - Phi((X - forecast_high) / SD) where Phi is the standard normal CDF.
5. Intraday update: As the day progresses and observations come in, the distribution narrows dramatically:
   • Before any observations: use full forecast SD.
   • After morning observations: if current high is already 74F at noon, P(high > 74F) = 100%. Recenter distribution on current running high with reduced SD (only upward uncertainty remains).
   • After peak time: if temperature has been falling for 2+ hours and is now 4F+ below running high, the distribution collapses — P(high > running_high + 1F) approaches 0.

Recalculation frequency:

• Pre-dawn (5 AM): Full probability curve from overnight model runs.
• Every 2 hours pre-peak: Update with latest NWS forecast and observations from api.weather.gov.
• After observed peak: Recalculate every 15 minutes as temperature data comes in. Each observation narrows the distribution.
• On lock-in signal: Recalculate immediately — the distribution collapses to a near-certain outcome.
• Precipitation / precipitation ladder: Recalculate every time HRRR updates (hourly) or NBM updates (every 6h).

4. Ladder Scanning Methodology

Step-by-step process:

1. Pull all contracts for each city on both platforms: Call Kalshi API for all active temperature contracts for each of the 20 tracked cities. Call Polymarket API for equivalent contracts where available. Typically ~10-15 rungs per city per day on Kalshi, variable on Polymarket.

2. Determine current state:

   • What is the NWS forecast high? (api.weather.gov — no other source)
   • What is the current running observed high?
   • What is the current temperature?
   • Has peak time passed for this city? (Use solar noon + thermal lag model, NOT hardcoded times)
   • Has a lock-in signal fired? (4F+ drop post-peak)

3. For each contract threshold, calculate fair probability:

   • Pre-peak: Use the NWS forecast-anchored normal distribution. P(high > X) = 1 - Phi((X - forecast) / SD).
   • Post-peak, no drop yet: Narrow the distribution — the running high is the floor. P(high > X) = 1 if X < running_high; otherwise use narrowed distribution.
   • Post-peak, lock-in signal fired: P(high > running_high) = ~100%. P(high > running_high + 2F) = ~3% (rare late recovery). P(high > running_high + 4F) = ~0.5%. Contracts above the locked high are massively overpriced.

4. Calculate raw edge: fair_NO_price = 1 - fair_probability. If Kalshi/Polymarket YES price > fair_probability, buy NO (edge = market_YES_price - fair_probability). If YES price < fair_probability, buy YES (edge = fair_probability - market_YES_price).

5. Calculate net edge after fees:

   • Kalshi: subtract ~1.5c per contract.
   • Polymarket: subtract ~2c per contract (gas + platform fee).
   • net_edge = raw_edge - platform_fee

6. Filter:

   • Minimum 3-cent net edge.
   • Minimum $50 liquidity.
   • For lock-in signals: lower the minimum edge to 2c because confidence is so high (93%+).
   • For forecast-based signals: require minimum 5c edge because model uncertainty is higher.

7. Rank and flag:

   • Sort by net edge, largest first.
   • Flag all contracts where threshold is >4F above the locked high as "TAIL — high confidence."
   • Flag contracts where lock-in signal fired as "LOCK-IN — act immediately."
   • Flag contracts within 2F of the forecast high as "MAIN — moderate confidence, better liquidity."
   • Flag tail contracts (outside ±1.5 SD from forecast mean) explicitly for both platforms. [PANEL: Gemini Pro]

8. Track edge persistence: Log every detected edge and rescan every 15 minutes. Build a decay curve: how fast do weather edges shrink? Hypothesis: lock-in edges persist for 30-90 minutes because weather markets have fewer active traders than sports. Log the decay rate per contract type (e.g., ">78F edges drop ~2c/hour on Kalshi in the lock-in window"). [PANEL: Grok-4-Fast]

5. Edge Examples with Real Numbers

Example 1: Lock-In Signal — Chicago Post-Peak Drop

• City: Chicago, March 15, 2026
• NWS forecast high: 76F (source: api.weather.gov, station KORD — only trusted source)
• Observed running high: 75.2F at 2:23 PM CT (Kalshi station: KORD)
• Current temp: 71.0F at 4:45 PM CT (4.2F below running high)
• Peak time: Solar noon 12:07 PM + 3h thermal lag = ~3:07 PM CT. Peak has passed.
• Lock-in confirmed: Yes — 4.2F drop, post-peak, temperature falling steadily
• Kalshi ladder rung: "Chicago high above 78F" priced at 11 cents (implied 11%)
• Polymarket ladder rung: "Chicago high above 78F" priced at 14 cents (implied 14%)
• Fair probability: P(high > 78F) given high is locked at 75.2F = ~1% (would require a 2.8F recovery post-peak with falling temps and a cold front — virtually impossible)
• Kalshi raw edge: 11% - 1% = 10 cents | Net after fee: 8.5 cents
• Polymarket raw edge: 14% - 1% = 13 cents | Net after fee: 11 cents
• Kalshi trade: Buy NO on "Chicago high above 78F" at 89c. Expected value: +8.5c/contract.
• Polymarket trade: Buy NO on equivalent contract at 86c. Expected value: +11c/contract.

Example 2: Forecast-Based Predictive — Model Disagreement in Phoenix

• City: Phoenix, March 15, 2026
• NWS forecast high: 88F (api.weather.gov)
• GFS model: 89F
• ECMWF model: 85F
• HRRR model: 86F
• Ensemble mean: 87F (lower than NWS by 1F, with 4F model spread indicating uncertainty)
• Adjusted SD: Normal SD for PHX March = 2.0F. Add 0.5F for model disagreement = 2.5F.
• Kalshi ladder rung: "Phoenix high above 92F" priced at 20 cents (implied 20%)
• Polymarket ladder rung: "Phoenix high above 92F" priced at 23 cents (implied 23%)
• Fair probability: P(Normal(87, 2.5) > 92) = P(Z > (92-87)/2.5) = P(Z > 2.0) = 2.3%
• Kalshi raw edge: 20% - 2.3% = 17.7 cents | Net after fee: 16.2 cents
• Polymarket raw edge: 23% - 2.3% = 20.7 cents | Net after fee: 18.7 cents
• Trades: Buy NO on both platforms. Polymarket has a larger raw edge but thinner liquidity — use limit orders.

Example 3: Multi-City Correlation — Cold Front Sweep

• Situation: Cold front sweeping through the Midwest on March 15, 2026. Affects CHI, DET, CLE, PIT.
• Chicago: Lock-in signal already fired (Example 1 above). Edge captured on Kalshi and Polymarket.
• Detroit: Running high 73.8F at 3:30 PM ET. Current temp 70.1F and falling (3.7F drop). Peak time passed. Not quite at 4F threshold yet, but the front is arriving.
• Kalshi ladder rung: "Detroit high above 76F" priced at 25 cents (implied 25%)
• Polymarket ladder rung: "Detroit high above 76F" priced at 28 cents (implied 28%)
• Fair probability: Based on the pattern from Chicago (which is 2 hours ahead on the front's track), Detroit will lock in at or below 74F with ~90% probability. P(high > 76F) = ~5%.
• Kalshi net edge: 25% - 5% - 1.5% fee = 18.5 cents
• Polymarket net edge: 28% - 5% - 2% fee = 21 cents
• Cleveland and Pittsburgh: Same front arrives 1-2 hours after Detroit. Pre-position in CLE and PIT "above X" contracts on both Kalshi and Polymarket before the lock-in signal fires there.

Example 4: Precipitation Edge — High-Confidence Rain Signal

• Market: Kalshi + Polymarket, "Will Philadelphia receive 0.01in+ of rain between 1pm-4pm on [date]?"
• Sharp Reference: 12Z HRRR model run shows 90% PoP for PHL grid box during that 3-hour window (from api.weather.gov HRRR output)
• Kalshi price: 74 cents (implied 74%). Polymarket price: 70 cents (implied 70%).
• Fair probability: 90% (HRRR PoP — the sharpest available short-term precipitation model)
• Kalshi edge: 90% - 74% - 1.5% = 14.5 cents. Buy YES.
• Polymarket edge: 90% - 70% - 2% = 18 cents. Buy YES — larger edge due to market inefficiency.
• Key rule: NWS/HRRR is the ONLY source for this probability. Web search rain estimates are unreliable and prohibited. [PANEL: Gemini Pro]

IMPORTANT NOTES

• The Weather Lock-In desk is unique among all desks because the lock-in signal has an extremely high win rate (92-99% depending on drop threshold). This justifies larger position sizes than other desks.
• Temperature resolution is ~1.8F (1C) in many data sources. Use NWS API which reports in 1F increments natively. NWS API (api.weather.gov) is the ONLY trusted source. Never use web search for temperatures.
• Verify which weather station Kalshi AND Polymarket each use for each city. A 2-3F discrepancy between stations will create phantom edges or cause real edges to settle wrong. Platforms may reference different stations for the same city — confirm both.
• The biggest edges are on contracts 4F+ ABOVE the locked-in high. These are the tail contracts. Prioritize them.
• Weather patterns are correlated across nearby cities. When one city locks in, immediately scan neighboring cities on both Kalshi and Polymarket.
• Seasonal variation matters: the lock-in model needs different SDs for summer (lower uncertainty, more reliable signals) vs spring/fall (higher uncertainty, more false signals).
• Both Kalshi and Polymarket have temperature ladder structures with multiple contracts at different thresholds. Neither platform is secondary for this desk. Scan both every cycle. Polymarket may show larger raw edges due to lower market efficiency, while Kalshi typically has deeper liquidity on US city temperatures.
• For precipitation markets: HRRR is the primary sharp reference for short-range (<18h) probability. NBM for longer range. Never use web search PoP estimates.
• Track edge decay rates per platform separately. Kalshi edges may persist longer than Polymarket on the same event if one platform's market makers are more active.