Latency percentile fans
p50 = typical · p95 = slow 1-in-20 · p99 = worst 1-in-100Each fan runs p50 → p95 → p99 (arrowhead), tail p95→p99 shaded darker, tick = max. A short fan = fast & consistent; a long red tail = occasional stalls. All values in ms on a shared scale.
Latency Distribution Lab
SecurityType × OrderType matrix
click a cell to cross-filter the whole pageBy security type
Percentile fan of the active metric per asset class, with fill rate and n. Sorted fastest p50 first.
By order type
Percentile fan of the active metric per order type, with fill rate and n. Sorted fastest p50 first.
Tail Spotlight — worst order
click any outlier dot or row to load it hereThe slowest order in the current filter, decomposed place → Submitted → Filled/terminal. The faint ghost bar underneath is the current p50 fill latency for scale. Segments show whether the delay lived in broker accept (submit) or broker-side execution.
Stage attribution — where does the time go?
Per slice: submit latency (integration/API accept) vs broker-side execution (market-exec for Market orders; resolution for resting orders), at p50 and p95. Segment medians are attributed independently, so they need not sum exactly to the fill median.
Latency over the run & serialized throughput
SERIALIZED — one order at a time, ≤10 s WS ack per order, no RateGate; NOT broker parallel capacityEvery order, x = monotonic elapsed ms (Stopwatch offset — not wall-clock), y = active metric. ▲ buy / ▼ sell; hollow = not filled; color = SLO band; rolling p50/p95 lines. Drag-select (brush) to filter every panel and the grid below.
Fill reliability
Fill rate = filled ÷ all placed (combo legs deduped to one parent order). Stacked final_status per asset class; resolution fan applies to resting order types only.
resolution_ms — resting orders (Limit / Stop)
Execution quality — slippage
Signed, side-adjusted: positive = worse than the quote at order time, negative = price improvement, zero = exactly at quote (common, kept visible). Lead unit is $ / unit; %-of-spread is the stable relative view. bps is inflated on low-premium options — never the lead number
Time-of-day (session) comparison
session comes from the CSV column only — never recomputed from timestampsET windows: open 09:30–10:00 · midday 12:00–13:00 · close 15:30–16:00 · other = everything else. Submit & fill fans plus median bid-ask spread per bucket.