apps2 · rsiobv — do momentum and volume agree?

The five rules — frozen first, backtested after

Definitions were written down BEFORE the backtest ran (the numbers attach to the rules, never the reverse). Stats: each asset's own tops/bottoms (21-bar extrema) within 10 sessions of a fire, ~25y daily × 20 assets, pooled fires with an independent-episode count, era-split. Full grid in the research artifact. In-sample throughout — a lift is history, not a promise.

loading rules…

Active combined signals

Every (asset × candle) where a rule is live RIGHT NOW, ranked — the Candle column says which candle size printed the signal (30m · 45m · 1h · 4h · 1d · 1w). Click a column to sort; click a row to open its three-pane chart. Divergence rules stay live while the confirming pivot is ≤ 60 candles old; extreme rules while the condition still holds. Alerts: three labelled streams under [RSIOBV] — RSI divergences, OBV divergences, combined signals — on 30m · 45m · 1h · 4h · 1d candles (combined also 1w) · each line names candle size, divergence width in candles, candles since the confirming pivot · 5-min scan.

Loading…

loading the covered-instrument list…

Notes

Percentile first, level second Extremes are triggered by where a reading sits vs ITS OWN trailing history (≥90th / ≤10th percentile over ~2 years of bars), not by fixed levels. The classic RSI 70/30 lines are drawn for reference only. OBV has no fixed levels at all — it is an unbounded running total, so its extreme is the "stretch": distance from its own 50-bar average, in units of its own typical wobble, ranked against its own history.

Why some rules ship with "no edge" The backtest attaches numbers to rules that were frozen first. Where a rule scored at or below its base rate (both-divergences-bearish, hollow-rally), the table says exactly that — the rule still renders on the board as a factual description of what RSI and OBV are doing, it just carries no historical edge claim.

Where the stats come from ~25 years of daily bars per asset (paged from the house store; younger listings carry what they have), each asset scored against its OWN 21-bar tops/bottoms within 10 sessions, fires pooled across assets WITH an independent-episode count (same rule, same day, many assets = one episode). Intraday windows show the rules live but contribute no stats. All in-sample.

Fire timing honesty A divergence confirming at a pivot needs 5 more bars to be knowable (the swing must hold). Every fire — in the backtest AND in the live scanner — is clocked at that later bar, never at the pivot itself. Extreme episodes fire at the first bar the condition became true.

Windows & resolution Rules and alerts compute on the candle size, not the history window. Candles: 30m · 45m · 1h · 4h · 1d · 1w (alert-grade). The window only decides how much history is loaded. Auto maps a window onto one candle: 1d→30m · 1w→45m · 1m→1h · 1q→4h · 1y→daily · 5y→weekly. The candle selector (auto/30m/45m/1h/4h/1d/1w) overrides per window where feasible (45m/4h aggregated in-browser on epoch-UTC buckets; 30m/1w native). Windows end at the latest session with data.

Telegram alerts Three labelled streams under [RSIOBV] — RSI divergences, OBV divergences and combined signals — on 30m · 45m · 1h · 4h · 1d candles (combined also 1w). Each line names the candle size, the divergence width in candles and the candles closed since the confirming pivot. 5-min scan. Conflict combined-signals stay board-only by default. Same Telegram room as the /obv and /rsidiv alerts.

All apps2 · data via basics → obv → rsiobv (single-writer chain) · engines: shared signals.js + obv/engine.js (classic OBV, parity-tested) + frozen rules (rsiobv/engine.js) · backtest: unlock-clusters methodology, node port · obv · rsidiv