Every chart shows what price did. Ours answers why — and whether it matters.

A context strip under every chart shows where the visible window sits inside everything loaded. Drag the body to pan, the handles to resize, the left edge to fetch deeper history — without ever losing your place.

Pre-market and after-hours shaded, open and close ticked, the previous close as a dashed reference, a market-status chip with the exchange clock. The calendar knows holidays and half-days.

All series share a union time grid; every line crosses the baseline together at the first common bar; a series that starts later joins at a marked point, its absent history never faked; incomplete trailing bars are dropped. Enforced in unit tests, not conventions.

Resolution follows the window, not the preset: the finest timeframe that keeps the chart within a points-per-pixel budget, with pan headroom prefetched. No more 56 points stretched across 900 pixels.

The rebase anchor has a physical marker. If the baseline pans out of frame, an edge cue says where it went — "+4%" always has something to be +4% of.

Adjacent series are separable for colorblind users — validated by a simulator in continuous integration, not by eye, with direct labels on every series.

Two synced panels: the legs above, and below them the β-hedged spread as a z-score — the mean-reversion instrument itself — with ±1σ/±2σ bands and divergence shading. One x-axis, one crosshair, one viewport.

Three representations of the same data are one chart in three states. Toggle to "spread" and two lines visibly collapse into one — you never lose the thread of what you're looking at.

Relationships aren't constants. A rolling strip under every comparison answers, at a glance, whether the spread you're watching still exists — a β break is the story.

Chart NVDA against Semis-ex-NVDA — the composite recomputed without the member, so the name is never compared against itself. The default drill-down from every basket row.

A five-day chart becomes five sessions shoulder-to-shoulder with subtle seams — every pixel is tradable time, instead of ⅔ dead air.

Earnings, a Fed day, the day a name joined a basket: align any set of series on the event and compare the reactions directly.

Scrub any historical window with the live layers on — priors bands populating, spreads diverging, the decomposition updating bar by bar. For post-mortems and honest "would I have seen it?" review. Only mpriors can replay with explanations, because the explanation layers run on the same pipeline.

Reversion tendency ("historically pulled back toward the group" vs "drifts — no pull detected"), half-life ("stretches like today's typically halved in ~9 sessions"), relationship kind (return co-movement vs price-level equilibrium), and sample honesty. Never a grade, never a recommendation.

Every watched relationship, sorted by how stretched it is versus how it has historically behaved. "What stretched today?" answered in under five seconds.

Mean reversion breaks rarely and violently — usually at maximum confidence. Mpriors monitors the assumptions: β stability, cohesion decay, half-life drift, record stretch. When the ground moves, you get a notice that names the assumption, not an action.

Watching six correlated spreads is not six bets. The book view computes your effective independent bets from the correlation of your divergences — six spreads at ρ̄ 0.5 are about 1.7 — and shows which one is redundant.

Scan for candidate relationships and every result carries its trials count: "1,240 pairs examined; ~62 would look this good by chance; this one held for 3 distinct regimes." The only scanner a statistician wouldn't wince at.