Order book & liquidity
bids — asks —
Delta & absorption
Who was aggressive, and whether it moved the price. Every candle's DELTA — market buys
minus market sells, from the exchange's own taker split — and the CVD against the price; ABSORPTION, heavy
aggression the passive side soaked up without the price giving way; INITIATIVE, aggression that carried the price
through its range. Every read in the loaded history is graded by the close 3 / 5 / 10 bars later next to the
baseline of all bars; the live bar's footprint comes from this session's tape.
Delta is a candle's taker-buy volume minus its taker-sell volume, priced at the bar's
average price; the CVD is its running sum. EFFORT is measured against THIS market's own recent bars: the bar's delta as
a robust z-score (its distance from the median delta of the previous 50 bars, in units of their median absolute
deviation × 1.4826) — perpetuals often lean to the sell side bar after bar, so a bar is judged against that
lean, not against zero, and its own delta must still point the way it is called. ABSORPTION: |z| ≥ 2 on at least the
median volume of the previous 20 bars, and the close made less than a tenth of an ATR in the aggressor's favour, or
went against it (marked “at the 20-bar low / high” when the aggressor drove into that extreme); the textbook
reads it the other way — selling absorbed → up. INITIATIVE: |z| ≥ 2 on 1.2× volume, a close at least
half an ATR the aggressor's way, in the bar's outer third and beyond the prior 20-bar high or low — continuation.
THE DELTA TREND: over the last 20 closed bars, the price change in ATR beside the net delta share in excess of the
market's own drift (the 100 bars before), in units of what 20 ordinary bars wander — a rise on net buying or a
fall on net selling CONFIRMS, a rise on net selling or a fall on net buying DIVERGES (the textbook reads the other
way), heavy one-sided delta with the price held within half an ATR is ABSORBING. The FOOTPRINT (this session's tape,
back-filled to the bar's start where the exchange's trade history allows): bought and sold at each price row, the
POC row, a DIAGONAL IMBALANCE where the buying at a row is at least 3× the selling one row below (or the
mirror), and three in a row STACKED. Measured (backtest/orderflow_hunt.py, Oct 2026: these exact reads, thresholds
fixed by how often they fire before any outcome was looked at, graded by the close five bars later against the side they
call, 16 markets, 2022 to date, 1h and 4h): none of 20 pre-registered cells clears the bar . Absorption is
zero — +0.003 ATR five bars on at 1h over 9,921 reads, +0.004 at 4h over 2,431, and no better at the 20-bar
extreme than in mid-range (drawn stronger there only because the textbook says so); the textbook delta divergence
loses — a rise on net selling kept rising (−0.17 ATR at 4h, t −2.9); initiative buying came closest
(+0.16 ATR at 1h, +0.29 at 4h, t ≈ 3 in sample but 1.7 out of sample, before the 0.1–0.2 ATR a trade pays in
costs). On GOLD (XAUUSDT since its December 2025 listing, 7,240 hourly bars) none of 20 either: absorption fires far
more often there (about 6% of hourly bars against 1–2% on crypto) and went slightly the wrong way at 1h
(−0.16 ATR, 47% its way). All of it is description, graded so it can be judged; nothing here is a signal of the
trading system.
Book watch 🔔 notify
The order that has not traded yet is the only thing on the screen still ahead of price: a large resting
order sits in the book until it is hit, pulled or walked through. This card watches that — before the candle prints —
and grades every event by where the mid went 1 / 5 / 15 minutes later.
A wall is a single resting order that is an outlier for THIS book: at least 8× the median level and 4% of
its side within the band, and at least $150K. It must sit for 3 seconds to count (PLACED); it is PULLED when withdrawn before
price reaches it (the spoof signature), EATEN when price trades through it, HOLDING when it absorbs at the touch for 15 seconds,
and the book FLIPS when resting orders within ±0.5% lean one way for 10 seconds. Each event carries the direction the
textbook expects (a sell wall placed → down, pulled → up) and is scored against the mid afterwards. Measured on this
basket (Sep 2026, 16 coins, 2022–2026): resting-book pressure is not a trading edge after costs at 1h / 4h. This is a
timing read for your own decisions, graded live in this session only; nothing here is a signal of the trading system.
Order flow & institutional prints
The tape, as it prints: every aggressive trade from the exchange's public trade stream and every
forced order from its liquidation stream. Institutional-sized BLOCK prints, SWEEPS through several levels, aggression
that was ABSORBED without moving price, ICEBERG levels that kept refilling, and LIQUIDATION cascades — each marked
on the chart with the direction the textbook assigns and graded by the price 1 / 5 / 15 minutes later.
A block is one aggressive print at or above the block threshold: the larger of 0.01% of the 24h volume,
the 99th percentile of the last 2,000 prints, 8× the median print and $20K — scale-free, so it means the same
on BTC and on a small cap. A sweep is same-side aggression inside one second across three or more price levels worth two
blocks. Absorption is a minute of one-sided aggression worth three blocks that moved price less than a tenth of a normal
minute’s range: the passive side took it all (the accumulation / distribution signature). An iceberg is a price
level that filled at least three times what the book displayed there in twenty seconds and still shows size. A liquidation
is the exchange’s own forced order; three on one side inside thirty seconds are a cascade. Size is the only tell the
public feed gives: institutional-SIZED, never a known institution. Measured on this basket (16 markets, 2022–2026):
taker / book / open-interest pressure carries no standalone trading edge after costs at 1h and 4h (Sep 2026), and the
trading system’s own closed-candle prints (big buy / sell, absorption, climax) cleared no pre-registered cell (Sep 2026).
This card grades the live version in your session only; nothing here is a signal of the trading system.
Support & resistance
Volume profile (the teal band): where the bars on screen did their business — the POC is the
busiest price, the value area holds 70% of the volume. Informational, not a setup: measured on this basket (Sep 2026, 192
pre-registered cells), entering on a retest of the POC after a box breakout loses net of costs both ways at 1h and 4h, and
waiting for the retest does worse than the breakout itself. Nothing on this card is graded from it.
Trend · every timeframe
A trend is read as a STATE, not a word: direction (EMA20 against EMA50 with the EMA20’s slope),
strength (ADX 14: strong ≥ 30, moderate ≥ 20, weak), phase (price riding the EMA20, extended beyond 1.5 ATR, pulling back
to it, or through the EMA50 — failing), the age of the trend (bars since the EMAs crossed), the retracement of the
last leg in the fib bands traders quote, and the volatility regime (ATR against its own last 100 bars). Every closed bar of
the loaded history is classified the same causal way and followed 6 and 24 bars on, so “continued” is a count
on THIS chart in THIS state next to the baseline of all bars — a description of what happened, not a forecast. The
higher timeframes name the line of least resistance; a lower-timeframe move against it is counter-trend.
Patterns · graded
Candlestick and chart patterns, found causally and named on the chart — and graded: every
pattern that completed in the loaded history is followed 3 / 5 / 10 bars on, next to the baseline of all bars, so each
name carries what it has been worth on this chart. Measured on this basket (16 markets, 2022–2026) with exactly this
grading (backtest/pattern_hunt.py, Oct 2026): 0 of 68 pattern cells clear the bar — pooled, the move 5 bars after a
chart pattern is +0.03 ATR at 1h (t 1.8) and +0.05 ATR at 4h (t 1.5), after a candle pattern +0.01 and +0.00 ATR, with
46–55% going the pattern’s way against a 50% baseline; earlier, candle patterns showed no edge over a blind
entry even at 4h and four textbook breakout patterns cleared none of 80 pre-registered cells. A pattern is a description
of what price did, not a reason to trade.
Scanning the loaded history…
Candle patterns are read on closed bars sized against the ATR (engulfing, hammer / shooting star, morning /
evening star, three soldiers / crows, outside bars, tweezers, doji). Chart patterns are read off the CONFIRMED swing pivots
the structure layer finds (a pivot counts only once it is confirmed, five bars later) and complete on the close that breaks
their neckline or line: double top / bottom, head & shoulders and its inverse, ascending / descending / symmetrical
triangles, rising / falling wedges, channels, flags and range boxes. A forming shape is drawn with the level that completes
it; a completed one is graded by the close 3, 5 and 10 bars later in ATR units, signed by its textbook direction (a triangle
or box takes the direction it broke). The delta divergence (price higher high, cumulative delta lower high, or the mirror)
is read the same way over the last 30 bars. Nothing here is graded as a trade: no stop, no target, no cost — it is the
forward move after the name, which is the most a pattern can claim.
Chart-call record
Textbook chart technique, honestly graded — replayed over its full history it loses
after costs (−0.10R per call at 1h, −0.05R at 4h across the basket); only a trailing exit with no flipping
brings it to zero. The sweet-spot sweep (192 pre-registered cells, Oct 2026: side × stop × exit ×
daily-trend gate × 1h to 1d) found none: the quickest exits lose the most at every timeframe, the best
cell is +0.01R at 4h and +0.03R at 1d (t ≤ 1.3 — zero), and the side that “pays” is the
market’s own drift. We publish it anyway: judge it yourself. The money side of THIS chart is read below,
next to blind entries on the same bars.
The same analysis replayed at every candle close over the loaded history, each call
resolved by the candles that followed, one position at a time: stop checked before target when both print
in one bar; every call entered at market on the close of its signal bar (no limit fills assumed); half off at TP1
with the stop moved to breakeven, the rest to the target; an opposite call flips
the running one at market; unresolved calls expire at mark-to-market after 48 bars; every call pays a 0.12%
round-trip cost allowance. Win rate is the share of closed calls that ended above zero. Active-at-close calls
only. Measured this way over the full history (16 markets, 2022 to date, 190,770 calls at 1h and 46,098 at 4h),
textbook chart technique loses after costs: −0.10R per call at 1h and −0.05R at 4h under this policy,
negative in every year, two thirds of the calls ending as flips. A 1.75-ATR trailing exit with no flipping brings
the 4h record to zero (+0.004R, t 0.3); no policy beats this one by the pre-registered 0.05R margin, so this one
stays. Four textbook patterns not drawn here (Donchian breakout, failed breakout, inside-bar breakout, EMA pullback)
were measured the same way, both directions: none clears the bar either. The ORDER-BLOCK retest drawn on the chart
was measured the same way when it was added (Oct 2026, 16 markets, 2022 to date): −0.06R per call on both
sides at 4h (about 3,900 calls, t −3) and −0.10R at 1h (about 17,000 calls, t −11), in line with the
other setups and not an edge. That is the measurement, not a
flaw in the page: the trading system's edge does not come from chart patterns, and this card grades the chart
technique, not the system. The sweet spot (backtest/sweet_spot.py, Oct 2026, 16 markets, 2022 to date): both
sides / longs only / shorts only × the setup’s level stop or a 1.75-ATR stop × four exits (this 2R
scale-out policy, the chandelier held 48 or 250 bars, the trail with no flipping) × with or without an
own-market daily-trend gate × 1h, 2h, 4h, 1d — 192 pre-registered cells, each judged next to a blind entry
every 20th bar under the same policy. None clears the bar (false-discovery control, out of sample t ≥ 2, most
years positive). The “quick” exits are the worst cell at every timeframe (4h: this policy −0.053R
per call at an 8-hour median hold; the no-flip trail 0.000R at 24 hours); the best cells anywhere are +0.012R at
4h (t 0.9) and +0.034R at 1d (t 0.6; out of sample +0.147R but two of four years negative). The daily gate adds
+0.02 to +0.08R to the trail cells at 4h and 1d and nothing at 1h or 2h; the ATR stop lifts 1h and 2h by about
0.02R and flips no sign. The side that pays is the basket’s own drift — 2h / 4h shorts and 1d longs beat
the other side blind as well, and at 4h the chart’s short setups earned LESS than blind shorts under the same
trail. The read is real but small: at 1h the setups beat blind entries by +0.04R (t 10.7) under a −0.14R fee
floor. The money-side block at the top of this card is that same comparison for the chart you are looking at.
Self-training · call policy
The record's own knobs — the flip rule (always / only a losing call / never), the
take-profit mode (scale out at TP1 or take it all at TP1) and the contact minimum a level needs (1 or 2) —
times the SIDE (both, longs only, shorts only) are 54 policies (the take-profit mode includes a 1.75-ATR chandelier
trail with no target — the exit the trading system itself runs); one more step drops the setup kind that scored worst. Every policy is replayed on the FIRST
60% of the loaded bars, the winner is then judged on the LAST 40% it never saw, and it is adopted only if it
beats the default by at least 0.05R per call in both halves with at least 10 holdout calls. When adopted, new
calls are recorded under it here; the record card above always keeps the pre-registered policy, so its numbers
are never in-sample. Re-learned on every load and every timeframe; nothing is stored.