# Session VWAP as a reference — lesson

Gym heading: Situational awareness. Written before the code (Sep 7 2026); revised the same day after the first harness run (the distance and the slope are compared at the precision the reveal shows; the highest-recent-candle check is a cap on what that candle can tell, not an independence test). The teach screen shows this text; the answer key implements the rule in it and nothing else. Distances are in M everywhere. Harness, confidence, persistence, and reveal format follow stop-vs-noise; the bars come from point-at-the-bar's generator v2 with a VWAP anchor added: every bar carries a synthetic volume, and the VWAP is the volume-weighted average price of the bars since the session open, not a moving average.

Choices made where the order was silent:

- Price is the last close. The distance is |close − VWAP| in M; M is the median range of the last 20 bars, as everywhere in the gym.
- The slope is (VWAP now − VWAP 12 bars ago) / 12, in M per bar. Flat is under 0.05 M per bar in either direction. Trending needs 0.05 or more in the direction of price: up when price sits above VWAP, down when it sits below.
- The distance is read at two decimals of M and the slope at three decimals of M per bar, the precision the reveal shows, so the number on the screen is the number the rule compared.
- A VWAP sloping 0.05 M per bar or more against price (falling while price sits above it, or the mirror) is not a leg carrying price away from its reference. It scores as stretched. The generator never builds one on purpose; the harness counts how many turn up, and the target is zero.
- Between 0.25 and 0.5 M scores as AT FAIR, as ordered, and the reveal says why. The share of the at-fair reps built in that band is a config (drill 0.35, holdout 0.45), so every session carries a few.
- The three answers come up a third each (config). Some reps sit within a few hundredths of M of 0.25 or 0.5 on purpose, and some VWAP slopes land within a few hundredths of 0.05 M per bar; the reveal shows the numbers either way.
- The bars are 15 minutes. Bar 1 opens at 18:00 New York, the anchor; the chart shows every bar since, frozen on the current bar, which lands somewhere between the small hours and late morning New York. The bar count is a family range, the same draw for every case. The rule counts bars, not minutes, and reads the same on any interval.
- Volume per bar: a time-of-day profile from the anchor (thin overnight, heavier from 02:00, heaviest from 08:00 and the 09:30 cash open) times the bar's range, with noise. VWAP uses each bar's (high + low + close) / 3. The volume is not drawn: the read uses the VWAP line and the bands, not a histogram. It is in the session log.
- The bands at ±0.25 M and ±0.5 M are drawn along the whole VWAP line with the chart's M, so the fair band and the stretch line are visible at every bar, not only the last. The 12-bar slope window is marked.
- The session's opening (from the anchor to the tail) is generator v2's pre-line phase, rotation, leg or mixed at the family's shares, drawn the same way for every case. The tail sets the case, and every tail shape serves more than one case: a rotation is at fair in its middle and stretched at its edge; a leg is trending while VWAP chases it, stretched when it crawls, at fair once it has come back to the reference; a spike is stretched, and at fair after it has reverted; a leg that stalled is trending, stretched or at fair depending only on how long VWAP had to catch up. So "the highest recent candle" turns up in all three cases. The harness checks that the bar count, M, the side and the opening regime carry no case (within 3.5 standard errors of the pool), and that a last close at the highest or lowest close of the last 12 bars is common in every case and pins none: no case's share in that bucket, or outside it, passes 0.6. A leg ends at the extreme by nature and a reversion to VWAP does not, so that candle is allowed to hint; it is not allowed to decide.
- The next 20 bars are drawn with an outcome rate that is a config constant and is never shown; the reveal says what price did, in M, and nothing about how often.
- Soft clock 8 seconds: two checks (the distance in M, the slope of the VWAP), one comparison each.

Persistent header: **"Training your eye for the distance to fair, and whether fair is moving."**

## Before you fade the top

- **Session VWAP** — Session VWAP is the average price the session's volume has traded at, anchored at the Globex open (18:00 New York). It's a crowd reference in the same family as the POC: in balance it's a magnet, and distance from it is measured in M, never points.
- **At fair** — Within ±0.25 M the market is at fair — nothing is stretched, nothing to fade.
- **Stretched** — Beyond ±0.5 M with a flat VWAP slope, price is stretched and reversion toward VWAP is the read.
- **Trending** — Beyond ±0.5 M with VWAP sloping the same direction, price is trending away — that's not stretched, that's a leg, and fading it is the honeypot.
- **The slope** — VWAP now against VWAP 12 bars ago, divided by 12, in M per bar. Flat is under 0.05 M per bar in either direction. 0.05 or more in the direction of price is a leg.
- **Between the bands** — 0.25 to 0.5 M is ambiguous by design: not at fair, not stretched, and nothing to fade. It scores as AT FAIR, and the reveal says why.
- **Why it matters at the trade** — Decision: "at the top" means stretched from VWAP in M with a flat VWAP. It never means "the highest recent candle."
- **What it doesn't say** — Stretched is not a forecast. It says nothing about whether the reversion arrives, how far it runs (Runway), whether the stop survives the noise (Stop vs noise), or whether anyone is stuck at the extreme (Trapped or not trapped). It says the reference is sitting still and price is far from it, in M.
- **M** — The median high-to-low range of the last 20 bars. Every distance here is in M. The number for the current chart is on the screen.
- **Confidence is a number** — Low means 40%, Medium 65%, High 90% that your call is right. Scored Brier-style: sure and wrong costs 0.81, unsure and wrong costs 0.16, sure and right costs 0.01. Being certain is as expensive as being wrong.

## The read

Two numbers, both in M: how far the last close sits from VWAP, and how far VWAP itself has moved over the last 12 bars. Then the question: inside the fair band, stretched from a reference that's sitting still, or riding one that's moving?

## The rep

The session from the 18:00 anchor, VWAP drawn from it, bands at ±0.25 M and ±0.5 M, frozen on the current bar. M is stated. The question, in full: **"Session VWAP from the 18:00 anchor. Is price at fair, stretched, or trending?"** At fair, Stretched, or Trending, with a confidence.

## The reveal

The next 20 bars with VWAP and the bands carried through them, the slope window marked. The distance in M, the slope in M per bar, the verdict, and one takeaway line. Templates:

- At fair: **"Price is {d} M from VWAP. Nothing is stretched. Tools silent."**
- At fair, between the bands: **"Price is {d} M from VWAP: past 0.25 M, short of 0.5 M. Not at fair, not stretched, and nothing to fade. Scores as AT FAIR."**
- Stretched: **"{d} M {above|below} a flat VWAP. The magnet is {below|above}; the read is reversion, and the stop sits beyond the extreme."**
- Trending: **"{d} M {above|below} a VWAP that is {rising|falling} {k} M/bar. That's a leg, not a stretch. Fading it is {selling|buying} a trend."**

Every rep ends on one of these lines. Each is a decision: nothing to do, a reversion with the stop beyond the extreme, or leave the leg alone. Then what the next 20 bars did, in M. No win percentage is shown anywhere.

## Answer key

`d = |close − VWAP| / M`; `slope = (VWAP now − VWAP 12 bars ago) / 12 / M`, taken as positive when it points in price's direction.

- **At fair**: d ≤ 0.25.
- **At fair, ambiguous by design**: 0.25 < d ≤ 0.5. Scores as AT FAIR; the reveal says why.
- **Stretched**: d > 0.5 and the slope is under 0.05 M per bar in either direction. A slope of 0.05 or more against price also scores as stretched; it is never built on purpose.
- **Trending**: d > 0.5 and the slope is 0.05 M per bar or more in the direction of price.

The three answers come up a third each (config); the ambiguous band is a config share of the at-fair third. Nothing but those two numbers carries the case: not the bar count, not the side, not the opening's regime, not whether price sits at the session's high or low.