THE TRADFI HEAT MAP

WHERE THE STATE WORDS COME FROM

Each word is cut from this instrument’s own recent distribution, not from a number anyone picked. Expansion vs pace is the session’s range so far divided by the median range at the same minute of its recent sessions, so a session at 10:30 is compared with other 10:30s and never with a whole day; relative volume is the same ratio for cumulative volume. The first fifteen minutes are left out of the pools, because a ratio against a near-zero denominator is arithmetic noise rather than a reading.

QUIET is pace below p30 with volume below its own p50. COILING is the same compressed range with volume at or above p50 — the volume is arriving and the range is not going anywhere. EXPANDING is the broad middle, pace between p30 and p85, which is where most sessions sit most of the time. RUNNING is pace at or above p85. EXHAUST-candidate is RUNNING that has stopped extending, and the stall cut in the table above is worth reading literally: it comes out at 0.00 on all four instruments, so the test is that the session has made no new high and no new low at all in the last thirty minutes. That is not a rare event and the table says how often it happens — on roughly half of the minutes when an instrument is already running. Read the word as “this move has stopped going anywhere”, which is what it measures, and not as a call that the move is over. It is a bar-grade proxy for exhaustion and it is not a tape reading of one.

WHAT THIS PAGE CANNOT SEE

It cannot see order flow, delta, absorption, or who is on which side of a print, because the programme holds no tape for these four instruments. It reads bars. Distances are signed in TR14 units, so a reading of −0.40 TR14 against VWAP means price sits four tenths of a fourteen-session true range below it. The day-table tag is labelled T+1 because it describes a session that has already closed; it is never a claim about the session in front of you.

HOW FAR BACK THE BASELINE LOOKS

Each instrument is measured against its own most recent 120 complete sessions, roughly six months, and each tile prints the exact dates that window covers. The window matters more than it sounds like it should. Both of the things this page normalises against have moved a very long way over the life of the corpora: gold’s median session range was 9.10 points in 2019 and 60.90 in 2026, and the Nasdaq micro’s session volume has multiplied by 2.4 since 2021 while gold’s has halved. A median taken over all of that is not a measure of normal; it is an average of several different markets, stale in a different direction for each instrument — which would quietly wreck the one thing a grid like this is for, which is reading four instruments against each other.

Two completeness tests decide which sessions are allowed in. A session must have printed at least 90% of its minutes — measured as minutes that actually traded, not as the distance to the last print, because a day that traded at the open, traded at the close and barely printed in between scores a perfect span and is not a session. And its volume must be at least 30% of the recent median, which removes contract-roll days, where the continuous series keeps quoting an expiring month whose volume has already left, and the tail of a data download, where prices arrive without their full trade aggregation. Both counts are kept and both are printed above. The same standard decides which session is allowed to be the dated tile: a day not good enough to shape the baseline is not good enough to be displayed as one.