Cross-asset regime research · live

Market Plumbing

An ex-hedge-fund trader's framework for a "real market edge" — One-Time Framing regime detection, and Dalton's rule that price reclaiming a value area rotates 80% of the way to the opposite side — tested mechanically against real OHLCV data across crypto, metals, equity indices, and FX. 8 of 21 target instruments processed so far; this page updates as the rest land.

Source video: "$50M+ Ex-Hedge Fund Manager Reveals How to Find & Trade a REAL Edge" (IQCapital) · Data: Dukascopy 1-min BID candles (2015–present), Bybit USDT-perp klines

Pipeline status

Bybit crypto universe (50 perps, market-cap stratified) finished first. Dukascopy runs sequentially — the feed throttles under concurrency — so it lands one instrument at a time.

50 / 50Bybit crypto perps

7 / 19Dukascopy instruments

37,858H5 events measured so far

Gold114m

Silver133m

S&P 500168m

Nasdaq 100245m

EUR/USD162m

GBP/USD158m

USD/JPY~46%

AUD/USD

USD/CAD

Apple

NVIDIA

Tesla

Microsoft

Alphabet C

Alphabet A

Amazon

US T-Bond

Euro Bund

TLT (20Y+)

BTC + 49 perpsdone

02 — H5

The 80% rule doesn't replicate

Dalton's rule, as commonly cited: once price reclaims a value area from outside and holds, there's an ~80% chance it fully rotates to the opposite extreme. Redefined here with an explicit hold-confirmation buffer (2 consecutive closes back inside) and measured as a completion percentage — not a binary hit/miss — across 8 assets and four asset classes.

Median across 8 assets, 49,158 qualifying reclaim events

~26% not 80%

Full-rotation rate ranges 23.7%–28.3% across crypto, metals, equity indices, and FX — a tight band regardless of asset class, session structure, or whether the market has a real closing auction at all. That consistency is itself the finding: this doesn't look like noise in one market's data, it looks like the redefined rule's real base rate.

Asset	Events	Median completion	Reaches 100%
BTC/USDcrypto	6,325	21.5%	24.4%
ETH/USDcrypto	5,416	21.6%	23.7%
Goldmetal	6,061	23.1%	26.2%
Silvermetal	6,775	21.2%	24.3%
S&P 500index	5,882	24.4%	28.3%
Nasdaq 100index	5,238	24.1%	27.0%
EUR/USDfx	6,488	20.1%	24.7%
GBP/USDfx	6,373	22.6%	25.0%

Bottom quartile is a flat 0% in every single asset. At least 25% of "reclaim + hold" events get rejected right back out with zero progress toward the opposite side — the hold filter isn't as clean a front-run buffer as hoped. But the distribution isn't uniformly weak either: P75 sits at 92–118%, so a real subset of events do travel most or all of the way. This reads as closer to bimodal (mostly-fails or mostly-completes) than a smooth "usually gets most of the way there."

Caveat on the table above: it uses yesterday's closed value area checked against up to 5 days of forward price action — a cross-session test. Dalton's rule is classically described as same-session: does price reclaim the developing value area and rotate before today's own session ends. Worth testing that more faithful version before concluding anything — see below.

Retest: same-session version

Rebuilt as Dalton actually describes it: a 1-hour Initial Balance profile at the start of each session, then reclaim-and-hold tracked only through the rest of that same day — no cross-day lookforward. Closer to the source claim, and it does close some of the gap — but not to 80%.

Asset	Events	Cross-session (yesterday → next 5d)	Same-session (this session only)
Nasdaq 100	4,179	27.0%	37.1%
Gold	5,145	26.2%	34.9%
BTC/USD	5,970	24.4%	32.0%

1-hour Initial Balance, 30-min hold confirmation. Stable across parameter sweeps (2-hour IB, 1-bar hold all land within ±3 points of these numbers) — this isn't a fragile result.

So: definition mattered, but not enough. Scoping to the correct same-session mechanic lifts the full-rotation rate by 6–10 points across all three assets tested — a real, consistent effect, not noise. But it plateaus in the low-to- mid 30s, nowhere close to 80%. That's reasonably strong evidence the commonly-cited figure is inflated relative to how it plays out even when tested faithfully — though the UTC-day-as-session approximation (still not each market's real exchange session) remains the one unresolved gap between this test and Dalton's original methodology.

03 — H1

One-Time Framing: crypto reverses, Gold continues

Strict OTF (higher-high + higher-low + close above the prior candle's high) tagged as up/down/balance, then measured against the forward 5-bar return. If OTF is a real continuation signal, the "up" tag should out-earn "down" and the unconditional baseline. On crypto, it's the opposite.

BTC/USD, daily

	Mean fwd. 5-bar	Win rate
Unconditional	+0.536%	53.7%
OTF up	+0.869%	51.2%
OTF down	+0.443%	57.3%

"Down" tags win more often than "up" tags over the next 5 bars — short-horizon mean-reversion, not continuation.

Gold, daily

	Mean fwd. 5-bar	Win rate
Unconditional	+0.190%	54.3%
OTF up	+0.214%	54.0%
OTF down	+0.028%	50.5%

Here "up" beats "down" as expected — the continuation reading the framework claims.

Immediate-failure rate favors the strict definition either way: across every symbol tested, strict OTF's tag flips within 3 bars 28–45% of the time, versus 63–75% for Dalton's original looser definition (just "no lower low"). Whatever OTF's forward-return edge turns out to be, the strict filter is meaningfully less whipsaw-prone.

This isn't a clean refutation of OTF — a 5-bar-forward horizon tests something narrower than the framework's actual claim, which is about multi-timeframe alignment producing a multi-day/week bias, not a single-timeframe pop 5 candles later. Testing longer horizons and the full 3-timeframe alignment (monthly/weekly/daily or compressed equivalents) is queued once more instruments land.

04 — H6

Profile shape: no signal yet, honestly

Does the shape of a day's volume profile (top-heavy / bottom-heavy / balanced / thin-trend), conditioned on the prior week's confirmed regime, predict which way the next day breaks — beyond what the regime alone predicts? Flagged in the original framework as the most exploratory, lowest-confidence component. First pass agrees.

Asset	Weekly regime	Shape	n	Mean fwd. 1d	Win rate
BTC/USD	up	top-heavy	71	+0.002%	56.3%
	up	bottom-heavy	64	+0.686%	57.8%
	up	balanced (baseline)	421	+0.146%	49.4%
Gold	up	top-heavy	186	+0.070%	58.6%
	up	bottom-heavy	128	−0.081%	43.0%
	up	balanced (baseline)	653	+0.042%	51.6%

Sample sizes per cell are 45–372 events — an order of magnitude below the 1,000+ bar this project holds everything else to. Top-heavy days show flat-to-weak continuation on BTC but a real edge on Gold; bottom-heavy is the mirror image. Reads as genuinely inconclusive on one-asset first passes, not as a signal to trade. The thin_trend shape also essentially never fires on crypto's 24/7 profile (0.0%) and is rare even on Gold (0.2%) — worth recalibrating the threshold once more session-structured assets land.

Methodology

otf_strict

Higher high, higher low, and close above the previous candle's high → tagged "up" (mirror for "down"); anything else → "balance." This is the video's own tightened version of Dalton's original One-Time-Framing concept.

value_area

Built per UTC calendar day from 15-min bars: volume distributed uniformly across each bar's [low, high] range into 50 price bins, then expanded outward from the POC bin until 70% of volume is captured. An approximation from bar data, not true tick-by-price volume.

reclaim + hold

Price closes back inside a value area after being outside it, then 2 consecutive closes stay inside (the front-run buffer) before the event counts. Completion is tracked as % of the value-area width traveled toward the opposite extreme, with a 5% stop-out buffer on reversal.

history

Dukascopy pulls are capped at 2015-01-01 (rather than each instrument's full available depth back to 2000–2003) — a deliberate tradeoff to keep the full 19-instrument pull to hours rather than days, while still giving 1h/4h/1d and 4h/1d/1w triads strong sample sizes.

significance

1,000+ qualifying events is the bar for treating a result as more than exploratory (the standard the source video itself sets for backtest validity). H5's 5,000–6,800 events per asset clears it comfortably; H6's 45–372 per cell does not.