Live alerts
Arm a confirmed search once. EdgeDepth checks the same condition on every closed market minute and records the moment it turns true, so you do not have to watch every chart.
Armed means the exact confirmed search was saved and enabled. The evaluator polls saved changes every 30 seconds, watches the selected markets, and fires only when the complete condition changes from false to true. The same definition cannot fire again on the same market inside its one-hour episode dedupe window. Missing warmup data means no fire, never a made-up zero.
Each row is one saved search, checked on every closed market minute.
Each row is one recorded market moment, frozen when the condition completed.
A fire records what the market did, not a simulated trade. It has no assumed side, entry fill, fees, stop, size, or exit, so calling it a trade return would be false precision. Forward market moves can be added here after each horizon completes.
The in-app bell always has every fire. Telegram and Discord delivery add a copy of each one, with durable retries if a channel fails.
Your exact question will be waiting. Sign in to interpret it and run it against the recorded market.
FEATURE STORE feature_defs.v1 · BINANCE USDT-M · UTC · LIVE ENGINE
EdgeDepth Research searches recorded market microstructure: 847 Binance USDT-M perpetual markets captured tick by tick, including perps that track Nvidia, QQQ, gold, silver and oil. Those are USDT-settled perps, not direct stock or futures-exchange feeds.
Your phrase becomes a structured query over a closed library of 58 versioned point-in-time readings, spanning VPIN, order-book imbalance, spread, funding, open-interest velocity, positioning, liquidation intensity, cascade clustering, returns, realized volatility, VWAP distance, taker flow, whale-sized prints, time to funding and completed-candle formations. The search returns every minute the condition held, and the number of eligible market-minutes it scanned rides every count.
Nothing here is a backtest. Prose is input only: the model proposes exact values, and nothing runs until you confirm them. Outcomes are measured forward from each match and can never be used as filters. When a reading has not warmed up at some minute, that minute leaves the pool the count is measured against and the exclusion is reported, so the result set never shrinks silently. Missing is reported as missing; it is never counted as zero. A verified zero is a finding, not an empty page.
Every result carries a reproducibility key: the same key reruns to the same bytes, and every occurrence deep-links into tick-by-tick order-book replay at its anchor. Read how research works, see the feature overview, or browse published research reports.