loading

Open positions

Arms

Closed trades

Arms side by side

Tracked wallets c8 clusters, one vote each

Live survivors

Open positions

Closed trades

Live deployments

Open position the venue's, with the stop the supervisor holds

Controls (header menu) act on the live book only here: Pause stops new live quotes and deployments, Cancel pulls the live quotes, Liquidate closes the live positions through the supervisor. The paper book keeps trading.

Funding

Hummingbot

Paper vs real arms eligible for live (PROBATION or SCALED)

The point of the live desk: the paper fill model measured against the venue. Fill rate is maker fills per hour quoted; markout is the mark 60 s after a maker fill from our side (negative is adverse selection). A gap in red means the paper model is optimistic.

Round trips per book win rate, expectancy, average gross win and loss, fees, markout

Pre-fill drift: the price move over the 5 s before each maker fill and the 5 s trade-flow imbalance, from the venue trade tape, signed from our side (negative: the price was running through the quote). Paper fills are measured at the first tape trade that reached their price.

Scouts

Maker round trips

c1 and c5 live on these numbers: the share of closes that exit as a maker, the spread captured per round trip, and the markout: the mark 60 s after each maker fill from our side (negative is adverse selection), for touch and trade-through fills.

Judge calls

Model calls per day on this desk and their estimated cost (DESK_COST_PER_JUDGE_CALL_USD each). Reused: answered without a call (the same coin, arm version and bar; the same dossier; or one call shared by several arms). Stub: the rules judge, no model.

Calibration

Lessons

Priors

Hypotheses

Sunday table

missing