Continue recording candles after a position closes to see what happened next. 0 = disabled.

0 = built-in default (30/min, the vendor's documented keyless ceiling). Applied by the endpoint pool within ~5 minutes, no restart; clamped to 1–300. The knee differs per exit IP, so this is an operational fact rather than a constant — measure before raising: python tools/measure_gecko_knee.py.

Credits the funding-trace gateway may spend per UTC day. Spend is persisted, so it survives a restart — the budget is a day's budget, not a process's. When the day's ledger cannot be read the gateway treats itself as exhausted rather than spending blind.

Comma-separated endpoint names (as they appear in /api/copy-trade/gecko-stats). A rested exit keeps every other duty — RPC, WSS relay, health probe — and simply stops being handed GeckoTerminal calls. Empty rests nothing, which is the default.
Read rate_limited_pct on the stats endpoint, not the raw refusal count: an exit that made almost no calls shows few refusals and looks clean. Resting is ignored for any call where every un-rested exit is unavailable, so it cannot strand the lane.

0 = off, one shared budget (today's behaviour). Nonzero holds back this many of each endpoint's slots for the candidate-tracker pass so the wallet-scanner grading pass cannot take the whole window — it measured 83% of all limiter waits. Scheduling only: the total rate above is unchanged, so this buys no extra calls. The reserve lapses if the tracker is idle for a full minute, so an unused slot is never wasted. Keep it below the rate above; applied within ~5 minutes, no restart.

0 = every pass, which is the default and the point of the lane.

Most wall-clock one wallet's reprice phase may take before the pass moves on (#2243). 0 = no bound. Default 600. A cut wallet keeps every position it already repriced, is not snapshotted, and resumes next pass — it is never marked covered.

The same bound for the ingest phase (#1734), which the budget above deliberately did not cover. Sampling found the three longest holds in a 69-minute window all in ingest — 1099s, 618s and 492s — while every healthy wallet finished inside 210s. 0 = no bound. Default 300, which sits in that gap. A wallet cut here keeps the positions it opened and keeps its checkpoint, so the window is re-read next pass rather than skipped (#2056).