Normalized overlay + stretch tapeTop: all assets on one scale — eyeball where two lines that usually track each other pull apart. Below, flush: the stretch tape — the SAME universe, each row vs its OWN history (percentile heatmap), where relative rich/cheap divergences pop that overlapping lines hide.
Multi-scale divergence — RSI price↔momentum across 1h · 4h · 1d
A divergence is when price and its momentum (RSI) disagree — price grinds to a new
high while momentum quietly rolls over (a top warning), or a new low while momentum turns up (a bottom warning).
It is a computed read you cannot see by eyeballing the line, so it is the one class of marker we label
(R-UI-23). Each instrument gets 9 cells = 3 timeframes (1h / 4h / 1d) × 3 swing lookbacks
(short 5 / normal 13 / wide 34). Cell color = type (bear ▼,
bull ▲, hidden-bear ▿,
hidden-bull ▵); size = strength; it fades as it ages; blank = none.
The Confluence column nets bull vs bear across all 9 — a divergence confirmed across many scales is
far higher-conviction than one lone cell. Pick a view mode for a different lens (each is explained
on-screen). Click any row to load it in the price+RSI chart, which draws the exact swing pair the read is
computed from. Slow macro levels & yields-as-levels are excluded — divergence is a price-action tool.
Retail P/C ref
loading…
loading divergence grid…
Excluded from this grid on purpose (divergence is a price-action tool and these have no real
intraday cadence — charting them here would fabricate a signal): 10Y yield, 10Y real yield (yields-as-levels),
M2 (weekly), US M2, Inflation (CPI), Fed balance sheet, Reverse repo (RRP),
Treasury cash (TGA). Those macro/monetary series live in the radar & correlation panels below.
Gold decoupling spotlight — the crystal-ball trigger (gold vs real yield + volume)
Decoupling alerts — premise · when it broke · live timer
Each card is a curated pair that normally moves a known way (its baseline premise, printed in plain language — that is the expectation the signal is measured against). We track the rolling correlation of the pair through time; when it deviates materially from that normal relationship — weakening past its baseline, or crossing to the wrong side (an inverse pair moving together / a positive pair moving opposite) — that is a decoupling. This is the same corr-vs-baseline read the top banner & correlation matrix already use (governed by the Corr break control), so the alerts here stay consistent with the rest of the page. Each card shows the premise, the date it deviated and the trigger correlation, a live timer of how long ago it decoupled (named to the pair), and a chart of the correlation (actual) vs its normal band + baseline (expected) with the onset pinned. Pairs still holding their normal link show a calm "holding" card — never blank. Correlations use the Corr window control; the panel follows the page View/zoom/pan and the synced dateline.
Snowball projection + edge checkIf the trigger fires, the chain reaction it historically sets off (from graph.json) — direction firm, magnitude hazy. The edge check below tests each declared causal edge against the DATA: declared lag vs the empirically strongest lead-lag, its correlation, and n — weak/insignificant edges (within the noise band) are greyed.
Decoupling radarPer asset: how extreme, how fast, and how far off what its peers predict. Flashing row = a break newly fired since your last visit.
Multi-scale tensorSame decoupling read at three lookbacks (micro · meso · macro). A break that climbs from micro upward is the earliest tell.
Correlation matrix (now) — click any cell for its lead-lagEach cell = correlation right now vs its baseline. Flashing = decorrelated; red = inverse flip (moved together, now oppose). Click a cell to see WHICH asset leads — the full lag profile (corr of A today vs B shifted ±N bars) plus how unusual today's lagged correlation is vs its own history.
Pairwise lead-lag matrix — who leads whom
Each cell tests every shift of one asset against another and keeps the shift with the strongest correlation. +N = the row asset leads the column asset by N bars (row today looks like column N bars later); −N = it lags. Cell color = the correlation AT that best lag (blue = move together, red = move opposite); greyed = within the noise band (no reliable lead-lag). Switch timescale to see if the lead survives across horizons. Click a cell for the full profile + the two z-lines.
timescale:
loading lead-lag matrix…
What's firing — ranked by heatEvery active break across all panels, hottest first.
Status & roadmap — what's live, what's coming
Data-first wireframeThis is a data-first wireframe. Look & feel comes after the signals are right.
The goalcatch trend changes early by watching cross-asset relationships break (extremes, decouplings, decorrelations) before price confirms — then project the chain reaction.