Pulse Terminal · Multi-Market Decision Research

Step 1 · Market setup

Bitcoin · Kalshi 15-minute

Waiting for contract data

CHECKING

Time until contract close

Actual Kalshi close

--:--

YES · UP

—

Best bid —

NO · DOWN

—

Best bid —

Opening target / strike

—

Current reference spot

—

Checkpoint: One server-recorded assessment at 7:30 remaining. Waiting for data.

Coinbase spot or public index data is a proxy, not necessarily the Kalshi settlement reference.

Step 2 · Dual-engine analysis

ENGINE A · CURRENT BASELINE

WAIT

Score —

Kalshi quote + price momentum

ENGINE B · SHADOW MODEL

WAIT

Estimate —

Price-only bootstrap · uncalibrated

Model comparisonWAITING FOR DATA

Favorable conditions

— / 7

Model direction

—

Inspect seven confirmation checks

Collecting eligible market data…

Before halfway: live checks. After halfway: recorded checks, if available. Both engines are experimental, may share price signals and have not demonstrated predictive accuracy. Only Engine A controls the paper research decision. Expand checks for detail.

Step 3 · Research decision

WAITING FOR DATA

Read-only research. No orders are placed.

Signal strength · not historical confidence—

Halfway checkpointSCHEDULED

Engine agreement · comparison only—

Why this decision?

Waiting for a usable market and fresh reference data

⚠ A model score is not a calibrated win probability. Contract prices, fees and execution matter.

Decision frameworkFive or more favorable conditions, fresh data, compatible quote bias, and an 8-point experimental score-to-ask cushion. Otherwise PASS.

Live preview updates. A saved halfway decision stays locked for the round.

Price and contract context

REFERENCE FEED

Last reference price

—

Above / below strike

—

Reference spotKalshi strike

Model diagnostics

EXPERIMENTAL

Engine A · blended UP score

—

Price model UP

—

Kalshi YES midpoint

—

Engine B · price-only UP score

—

Shadow estimate · not validated probability

20s projection (crypto)

—

Candle volume proxy

—

Not actual order flow

Wick lean · 1m volatility

—

Recorded decisions and results

↓ Export CSV

Resolved halfway calls—

Directional hit rate—

Forecast Brier (matched)—

Kalshi quote Brier (matched)—

Only officially settled predictions count. Brier scores use the same matched rounds, lower is better; small samples are inconclusive. Export before redeploying without a persistent database.

Market	Recorded (local)	Time left	Engine A	Engine B	Agreement	Research decision	Check score	YES ask	NO ask	Winner	Result
No decisions saved yet

Model performance · identical settled halfway rounds

Matched settled rounds—Both engines present at halfway

Engine A · direction correct—Same rounds as B

Engine B · direction correct—Same rounds as A

Engine A · Brier—Lower is better · uncalibrated score

Engine B · Brier—Lower is better · uncalibrated estimate

Agreed on direction—Of matched settled forecasts

Awaiting settled side-by-side records. Legacy Engine A-only predictions remain in history, but do not affect head-to-head statistics.

Sources, rules and limitations

The market, target and best bids/asks come from Kalshi's public APIs. Crypto reference prices and candles come from Coinbase; index reference prices may come from Yahoo Finance and can be delayed. These reference prices are not the official settlement-index feed. Engine A blends quote and momentum using arbitrary weights. Engine B uses a price-only empirical return simulation based on the same reference history and is not independent data. Neither is statistically calibrated, and Engine B does not directly simulate Kalshi's official settlement-index feed. The seven checks describe alignment and data quality, not a trained or independently validated probability model. The quoted price is an ask to buy; fees and slippage may raise the true cost. The experimental cushion is a score difference, not expected value. A candidate is for paper research, not a recommendation to wager real money. A missing current contract, stale data or missed checkpoint prevents a valid locked call. The Python server makes one assessment halfway through an eligible 15-minute contract and checks official Kalshi outcomes when available; older journal entries remain. No account access, automatic trading, or money movement.

Database persistence depends on your Render disk configuration.