The same gates the live bot runs on, overridable for this run only โ
nothing here is written back to .env. Blank, or the first
option in a dropdown, means "use the configured value"; the
placeholders are what that currently is. Prices are in dollars a
share, so 0.03 is three points. The signal is a sharp Binance move:
at least the move bar inside a second, or, on a busy tape, the bar
multiple ร BTC's average 1-second move over the lookback, if that is
more (0 = a fixed bar). A move buy whose side is more than the max
sd behind on the settling average, at BTC $/s, is refused (0 = off).
Orders. Every sharp Binance move buys the side it went at the ask -- by mid, 10 shares at 0.50 rising to 30 at 0.40/0.60, 50 at 0.30/0.70, 75 at 0.20/0.80 and 100 at 0.10/0.90, read off the side being bought; fixed, the same shares every time -- and then waits before the next; a move made during the wait is dropped. The window's budget is across both sides together, and each order is checked against it at its limit before it goes out, exactly as the trader does.
A window's winner is the resolution Polymarket published, or โ when
asked for โ the last TWAP published inside it against
twap_open, which is what these markets resolve on. On
the TWAP a window without a usable average falls back to the
published resolution, and wherever Polymarket published one anyway
the two are compared: that check is the only thing standing between
this page and a plausible P&L scored off the wrong number.
No run yet.
The truth sits between these three, not on any one of them.
No run yet.
Realised P&L running through the windows in the order they traded. Both lines carry forward across windows the other one was alone in, so the gap between them is the whole comparison.
Each window the strategy entered, on its own. A window holding both sides is paid on its winning leg and loses the other, so what it made is that leg's shares against everything both legs cost.
One row per window, newest first: what the replay would have done against what really traded. Select a row to open that window on the live dashboard.
Binance is the first sharp move the replay bought on, in dollars as it crossed the bar โ positive a rise (bought Up), negative a drop (bought Down). Where the replay stood aside, the gate named is the one that refused it for the most seconds, not whichever one happened to have the last word.
No run yet.
Counted per evaluated snapshot after the entry delay, over every window in range โ including a move that cleared the gates and was let go because the window's budget was spent. Snapshots inside the delay are not counted; they would bury everything that actually decided something.
No run yet.
The source file, the range it covered, and the gates it ran with.
No run yet.