How much of the raw trend the model trusts, based on proximity to sunrise/peak-heat hour and current cloud/wind. Low near those turning points, since the trend is about to reverse.

Widens when local pressure is falling (unsettled weather approaching), and independently when the marine push or offshore flow signals below are elevated - separate from the trend confidence above.

Widened: local pressure falling; marine push signal rising.

The same station readings the model itself runs on - what the trend fit and the cross-station gradient network are built from. History only; the dashed projection above is temperature-specific.

Measured from finalized days whose marine-push, offshore-flow and pressure-gradient state at prediction time matched today's, not one average across every day regardless of conditions (calibration_log.py). Falls back to the all-days figure until a specific combination has 5 matching days behind it - the sample count above always says which one you are looking at.

final NWS climate report (CLI) - the value Kalshi settled on

From the NWS hourly forecast (HRRR model) - real atmospheric dynamics, not this tool's own trend/persistence guess. Reference confidence is measured from the last 19 finalized next-day predictions using this same source (calibration_log.py) - not a placeholder.

From the NWS hourly forecast (HRRR model) - real atmospheric dynamics, not this tool's own trend/persistence guess. Reference confidence is measured from the last 19 finalized next-day predictions using this same source (calibration_log.py) - not a placeholder.

Highlighted bracket is the one this tool's own estimate falls into - not a probability call, just a quick eyeball check against the market's pricing.

Bid/ask are the current highest buy offer and lowest sell offer for that bracket (last-price is the most recent trade, which can lag). A wide gap between them usually means a thin, less reliable market.

Trades stay manual for now - this is a side-by-side view for a human to review, not an automated signal.

Highlighted bracket is the one this tool's tomorrow-high estimate falls into.

Highlighted bracket is the one this tool's tomorrow-low estimate falls into.

Trend confidence is how much the model trusts its own short-term extrapolation right now - not confidence in the readings above. It's deliberately low near sunrise or the day's peak-heat hour, since the trend is about to reverse there regardless of how strong it looks right before. Marine push (coastal/strait pressure gradient) and offshore flow (interior gap-wind gradient) are two distinguishable patterns behind Seattle temperature swings - the latter is the mechanism behind most real heat events, not just solar heating. Both are still first-pass signals, not yet calibrated against real outcomes (see calibration log).

A separate, closer-to-home comparison point - not part of the KSEA model above, and not blended into it. Sea-Tac Airport runs warmer than Fremont in summer (further from Lake Union/marine moderation), so treat these as two different microclimates. Current conditions are real observations from KBFI (Boeing Field, nearest full ASOS station - still ~5-6mi from Fremont). The high/low is NWS's own HRRR-based forecast for Fremont's actual coordinates, not this tool's in-house trend/gradient-network model - no backtesting or paper trading behind these numbers yet.