Overview
Waiting for the first cycle…
Equity
Action center
Crypto spot
Kalshi movers 15 min
Kalshi tape
Needs attention
P&L by strategy realized + open
Recent activity
Live feeds
Exchange prices, Kalshi trade tape and cross-venue gaps, updated every second.
Exchange board composite = median of live exchanges; CF index when Kalshi streams it
Kalshi trade tape
Top movers 15 min
Kalshi ↔ Polymarket gap = Polymarket − Kalshi
Positions
Action center positions worth a manual look · Sam never acts on these flags itself
Opened by Sam
Your Kalshi account
Everything the account holds, including positions you opened yourself. Read only: trade these on Kalshi.
| Loading… |
Resting orders
Activity
Trade log, daily results, alerts and engine log.
Daily P&L
Results by strategy
Trade log
Alerts & events
Engine log
Data flow
Where Sam's information comes from, and how it becomes an order.
From sources to orders
Sources
Health
Data feeds, connections and how fast the agent reacts.
Connections
Instant reaction
Milliseconds. Every strategy reprices a market the moment its Kalshi quote, trade tape, Coinbase price, Polymarket pair or a news story changes. "Event to order" runs from the first event received to the order being sent. Feed lag includes any clock difference with the exchange.
Crypto settlement check
Sam's index (the official CF Benchmarks index when Kalshi streams it, the exchange median otherwise) vs Kalshi's settlement value, and how well each forecast predicted the result (Brier score: lower is better). Builds up as crypto markets settle.
15-minute crypto
Closed trades of both crypto strategies in the 15-minute up/down series named above, and how accurate each forecast was at each point before the close (Brier score: lower is better). If Sam's blend doesn't beat Kalshi at a horizon, the model has no edge there. Exits use the expected-value rule: hold to settlement unless the market pays more than Sam's estimate.
Learning · 15-minute scalper
Each settled window is scored before either model learns from it (log loss: lower is better). Trading uses the active model; the candidate keeps learning. Promote it when it beats both the active and the base model over enough windows (the default judgement needs 200); roll back if it disappoints. Weights add to the base model's log-odds: + favours YES when the signal is positive.
Releases & nowcasts
Upcoming releases from Kalshi's own release markets (and FRED, with a key). The release guard holds back new entries around each one. Priors are the nowcasts the macro strategy blends into its ladder fit.
Signals
Every trade idea, and what the risk manager decided.
Strategies
Turn strategies on or off. Every one is sized by fractional Kelly.
Markets
The most active Kalshi markets the agent is watching.
Polymarket US pairs
Add a pair only after checking both contracts resolve on the same event, date and source.
Backtest
Runs the same strategy, sizing and risk code as live trading.
A backtest normally uses your live settings (min edge, limits, Kelly). If a strategy shows no trades, tick this to see whether your settings are what's blocking it.
Equity
Attribution by strategy
Why signals weren't traded
Your settings that differ from the defaults
Code review
Reports from Sam's code reviewer (coding and trading-math review, read-only).
Reports
Report
Settings
Changes apply live and are saved on the server.