Centre = each market at its own band midpoint
For each market over the selected window (1m ≈ 30 days · 1q · 1y …), the centre is where every
leg sits at the middle of its own window high–low range.
The band, in numbers
H = window high · L = window low · mid = (H+L)/2 ·
pos = (P − mid) / ((H−L)/2) (−1 at the low … +1 at the high) ·
band% = (pos+1)/2.
The dot = mid-deviations mapped on the triangle
dot = centre + ½·Σ pos·(vertex − centre), clamped to the triangle. Mids anchor the
centre on every window, and deviations get the full triangle: one market sweeping its
whole band (others at mid) moves the dot the entire half-median each way.
All three at their mids
→ the geometric centre. Nothing is stretched relative to anything else.
One high, two low
One at its window HIGH while the other two sit at their LOWS → exactly that
corner.
Two rich, one cheap
Two rich vs one deeply cheap saturates on the rich pair’s edge (e.g. gold+oil high in their
bands, NAS100 low in its own) → dot on the rich pair’s edge, far from the cheap corner.
Own band first
Richness is measured in each market’s own band first; the three compare only in the
mix.
H/L scan the full series
Highs and lows come from the whole window series (closes at the structure res), not the
decimated trail samples.
Closed markets hold their last close
Markets whose session is shut (Seoul overnight, EWY off-hours) sit flat in their band — a
closed leg never blanks the triangle.
High-water (HW)
Dotted red circle; radius = max distance of the dot from centre over the selected window
(session of peak stretch marked).
Extreme callouts
Per ticker: when it last printed its window high or low (whichever is more recent;
diamond on its ray, with date/time).
Perimeter
Band balance of that pair in percentage points; “=” means both sit at the same position in
their own bands.
Radial ray
That market’s band: centre end = window low · mid tick · vertex = window high; the marker is
now. Hover the trail for time + per-leg band% / per-pair Δband. Asset colours are fixed.
USD
Equal-weight EUR·GBP·JPY·CHF·CNY·KRW — not DXY.
The index legs are FUTURES
SPX · NAS100 · Dow 30 · Russell 2000 are the continuous front-month futures
ES · NQ · YM · RTY, not the cash indices: cash prints only 6.5 hours a day and would leave a
leg frozen for 17.5 hours against three live ones (and our realtime entitlement is futures,
not cash indices). All four run ~90 seconds behind the tape.
Mag 7 = the seven stocks, 1/7 each
A synthetic equal-weight basket of
AAPL · MSFT · NVDA · AMZN · GOOGL · META · TSLA — 1/7 (14.286%) each, geometric,
= 100 on 2026-05-11. Every one of the seven legs is also a series in its own right in the same
store, so the basket never hides its parts.
Mag 7 is NOT the MAGS ETF
Measured pre-market on 2026-08-17, MAGS printed 4 one-minute bars to its components’ 9–12,
with its newest trade 1315s old against their 955s. The wrapper is thinner and later than the
things it wraps, and it hides its holdings.
Why geometric, not an average
Geometric = equal weight in returns. An average of share prices would silently be
price-weighted — MSFT at ~$412 would outweigh NVDA at ~$219 purely on share count.
Mag 7 bucket rule
A bar is published only when all seven priced it, within a bounded and counted 900s
carry-forward. A partial basket is never published as the basket.
SKHY US listed 2026-07-10
SK Hynix’s Nasdaq listing (≈1/7 of a Seoul share) is only weeks old, so on the 1q and 1y
windows its two triangles start there — each panel’s header reports the
trail’s true date range, so read it rather than assuming a full year.
SKHY KR has no realtime feed
No vendor sells us Korean equities, so the Seoul line is Yahoo on a ~20-minute delay and holds
flat outside 00:00–06:30 UTC. It appears in the EWY · SKHY US · SKHY KR
panel.