BOOK STATUS

YELLOW

FULL SIZE — a regime signal is red, but REGIME IS DISPROVEN AS A SIZE SIGNAL (2026 replay: 38 regime de-risk signals, 39% accurate, −$24,358 if followed — it cuts winners). Regime is CONTEXT ONLY. Your performance is within norms → trade full size.

raw regime: RED · performance: within norms (not enough live data yet)

updated 2026-08-20

How to read this. Two kinds of signal, two speeds — and one calibration lesson. LEADING = regime (continuation, ER, autocorrelation) turn first but — proven against 2026 trade P&L — are weak predictors of whether the duo actually loses. So they mean WATCH / size cautiously, not stand-down. LAGGING = performance (drawdown, expectancy, drought vs the backtest envelope) are the real signal → they tell you to stand down / rotate. fine watch, size cautiously performance breached — act.

Two failures, two responses: your specific names slow down → check the bench for a replacement (but it's thin). The whole momentum regime flips and performance breaches → stand down, no ticker swap helps.

Size ladder what each state means & exactly what to do — your current state is highlighted above

State	Trigger	What to do
FULL	regime + performance within backtest norms	Trade full size.
WATCH (context only)	regime red, performance fine — regime is DISPROVEN as a size signal (2026: 39% accurate, −$24,358 if followed)	Trade full size. Regime is background context, not a reason to cut.
SCALE DOWN	live performance worse than backtested — DD past ~4.5%, before the kill-line	Cut size ~half, proportional to how far past normal.
STAND DOWN	performance breached the kill-line — DD >~10%, or expectancy <0, or drought beyond backtest	Stop / go to the bench.

Key: your SIZE is driven by performance, not regime. Regime red = watch only (calibrated on your 2026 trades — the duo made money across all regimes). Only your live equity drifting past its backtest envelope scales you down or out.

Proven findings signals tested against real P&L — what to trust, what to ignore

Signal	Tested	Verdict
Regime → cut size ER / continuation flips red	38 signals replayed vs 2026 P&L 39% accurate · −$24,358 if followed	DISPROVEN — ignore cuts winners; choppy days are profitable. Alerts disabled; kept as context only.
Performance vs envelope live DD / drought / expectancy past backtest norms	not a prediction — a measurement	USE THIS the only thing that changes your size. Near-tautological: worse than anything backtested = something changed.
Bench leaderboard your strategy backtested on 35 candidates (QC, nightly)	real backtests	USE THIS rotation reference — but the bench is thin (only MRVL/MU strong).

Live performance-tier signals logged so far: 0 matured, 0 pending — these fire only a few times a year, so this stays near zero by design.

Why this card exists: every intuitive signal gets tested against actual P&L before it's allowed to change your position size — and most fail. Regime sizing felt obviously right and would have cost you $24k in 2026. Nothing gets automated until it proves it makes money.

Watcher · your book's health

RSI live performance vs backtest envelope

Live trades

Net

Drawdown

$0 / env max ~6.0%

Expectancy

— / env ~0.9

Drought

0 / env max 18

Win rate

—% / env 55%

Live data still accumulating — judged against the backtest envelope until enough real trades build up.

Live drawdown vs backtest. How far below its peak your live equity is, vs the worst the backtest showed (~4.5%). Yellow 1.5× (~7%), Red 2.5× (~10%). A live drawdown bigger than anything the backtest produced = the edge may be broken.

Rolling expectancy (per trade). Average $ won per trade recently, normalized. Backtest ~1.0+. Yellow below 0.30, Red below 0 (losing per trade). Lags the regime signals but confirms them.

Drought (trades since equity high). Trades since your last equity peak, vs the worst dry spell in the backtest (~14). Going BEYOND the backtest's worst drought means the distribution changed — regime, confirmed by silence.

MRVL live performance vs backtest envelope

Live trades

Net

$-204

Drawdown

$204 / env max ~4.5%

Expectancy

-1.0 / env ~1.0

Drought

1 / env max 14

Win rate

0% / env 42%

Live data still accumulating — judged against the backtest envelope until enough real trades build up.

Rolling expectancy (per trade). Average $ won per trade recently, normalized. Backtest ~1.0+. Yellow below 0.30, Red below 0 (losing per trade). Lags the regime signals but confirms them.

MU live performance vs backtest envelope

Live trades

Net

$-266

Drawdown

$266 / env max ~4.5%

Expectancy

-1.0 / env ~1.0

Drought

1 / env max 14

Win rate

0% / env 42%

Live data still accumulating — judged against the backtest envelope until enough real trades build up.

Rolling expectancy (per trade). Average $ won per trade recently, normalized. Backtest ~1.0+. Yellow below 0.30, Red below 0 (losing per trade). Lags the regime signals but confirms them.

Watcher · regime on your traded names

MRVL $237.27 · regime

Intraday continuation

-41bp recent / +25 base

Efficiency ratio

0.13 36th pct

Autocorrelation

-0.02

Realized vol

96%

ATR% (range)

7.3%

Intraday continuation (the DIRECT regime measure). The signal your edge actually depends on: over recent days, when the morning move (9:30→10:00) is up, does the rest of the day (10:00→close) continue up (your momentum ride wins) or reverse (you get chopped)? Shown as average aligned move in bps — positive = moves continue (good), negative = moves reverse (your edge is actively fighting the tape). This beats the daily efficiency ratio because it measures intraday behavior directly. 'recent' = last 20 days, 'base' = ~3-month window; recent falling below base = the regime is turning against you.

Efficiency Ratio (Kaufman, 20-day) — judged vs the name's OWN history. How trending vs choppy the stock is — the net 20-day move divided by the total path it wandered. 1.0 = clean straight trend, near 0 = chop. Your momentum ride NEEDS trend. Shown as a percentile of the name's own 3-year history, because absolute levels mislead (MRVL/MU normally sit ~0.20). Green = trending more than its norm (>60th pct); Red = unusually choppy for IT (<15th pct).

Daily-return autocorrelation (30-day). Do up-days follow up-days? Positive = momentum regime (moves persist, edge works); negative = mean-reversion regime (moves reverse, edge gets chopped). A flip negative is the clearest 'this kind of strategy is dying' signal. Green >0.05, Red <-0.05.

MU $937.11 · regime

Intraday continuation

-195bp recent / -53 base

Efficiency ratio

0.03 8th pct

Autocorrelation

-0.16

Realized vol

100%

ATR% (range)

6.8%

RSI universe · per-name regime which of your 21 names are trending vs slowing

Ticker	Eff.ratio	Autocorr	Vol	ATR%	Gap%
ADBE	0.34 (69p)	+0.10	55%	4.1%	28%
AMZN	0.18 (41p)	+0.36	63%	3.6%	12%
APLD	0.07 (19p)	-0.16	107%	8.2%	50%
ARM	0.16 (36p)	-0.03	92%	8.3%	63%
BE	0.05 (15p)	-0.14	134%	10.8%	67%
BKNG	0.42 (83p)	+0.21	45%	3.4%	13%
CLSK	0.26 (62p)	-0.21	108%	9.2%	57%
COIN	0.06 (14p)	-0.29	68%	5.5%	37%
CRM	0.38 (78p)	+0.07	50%	4.0%	30%
CRWV	0.08 (19p)	-0.01	148%	10.1%	53%
CVNA	0.15 (33p)	+0.04	70%	5.5%	18%
HIMS	0.02 (4p)	-0.24	114%	6.3%	33%
HOOD	0.17 (37p)	+0.03	52%	4.8%	33%
MSTR	0.06 (14p)	-0.44	74%	5.5%	48%
NBIS	0.02 (5p)	+0.01	195%	11.8%	62%
NET	0.14 (41p)	-0.07	53%	6.2%	18%
OKLO	0.03 (9p)	-0.40	101%	7.7%	38%
ORCL	0.21 (38p)	-0.02	62%	4.9%	28%
PLTR	0.46 (89p)	-0.08	115%	5.2%	33%
TMUS	0.11 (26p)	-0.38	53%	2.6%	13%
VRT	0.20 (41p)	+0.33	86%	6.1%	48%

Bench · replacements ready to rotate in

How the bench works. Every candidate is scored two ways: fingerprint fit (how closely it matches your proven names — lower is better) and backtested edge (your actual strategy run on it nightly via QC — populating soon). When the Watcher above flags your book, take the top 1–2 here, confirm the nightly backtest, paper/small-live, then rotate in. A good fingerprint is a shortlist, not proof — the backtest is the gate.

Duo bench fit = closeness to MRVL/MU · edge = duo strategy backtested on it (QC)

#	Ticker	Fit ↓	ATR%	Vol	Gap%	Backtest edge
1	LRCX Semis	1.45	6.1%	91%	67%	-4% 28.0%w · 248t
2	AMD Semis	1.57	5.7%	79%	60%	-1% 31.0%w · 246t
3	KLAC Semis	2.30	5.5%	67%	62%	+1% 30.0%w · 230t
4	ARM High-beta movers	2.31	8.3%	92%	63%	-2% 28.0%w · 232t
5	DELL AI / data-center	2.43	7.6%	81%	42%	+1% 28.0%w · 222t
6	SMCI Semis	2.52	6.8%	98%	47%	-3% 23.0%w · 194t
7	MSTR High-beta movers	2.57	5.5%	74%	48%	-5% 19.0%w · 236t
8	VRT AI / data-center	2.58	6.1%	86%	48%	-10% 24.0%w · 226t
9	CRWV AI / data-center	2.78	10.1%	148%	53%	-7% 20.0%w · 232t
10	AVGO Semis	2.89	4.6%	44%	43%	-2% 34.0%w · 244t
11	ANET High-beta software	3.03	5.7%	61%	40%	-6% 24.0%w · 252t
12	GEV AI / data-center	3.08	4.6%	54%	40%	-4% 25.0%w · 228t
13	COIN High-beta movers	3.21	5.5%	68%	37%	-9% 18.0%w · 228t
14	TSM Semis	3.24	2.9%	40%	45%	-4% 36.0%w · 240t
15	APLD AI / data-center	3.25	8.2%	107%	50%	+1% 24.0%w · 212t
16	QCOM Semis	3.29	3.9%	48%	45%	-7% 28.0%w · 240t
17	DDOG High-beta software	3.42	7.7%	91%	25%	-7% 25.0%w · 240t
18	HOOD High-beta movers	3.50	4.8%	52%	33%	-10% 23.0%w · 218t
19	RDDT High-beta movers	3.52	8.4%	108%	23%	-9% 23.0%w · 226t
20	NOW High-beta software	3.53	5.1%	60%	50%	-7% 25.0%w · 234t
21	APP High-beta movers	3.61	6.9%	82%	32%	-6% 23.0%w · 222t
22	ON Semis	3.68	5.2%	53%	53%	-7% 24.0%w · 232t
23	OKLO AI / data-center	3.76	7.7%	101%	38%	-6% 20.0%w · 232t
24	PLTR High-beta software	3.82	5.2%	115%	33%	-0% 31.0%w · 230t
25	SOFI High-beta movers	3.90	4.0%	64%	27%	-4% 21.0%w · 164t
26	NBIS AI / data-center	3.92	11.8%	195%	62%	+2% 23.0%w · 240t
27	NET High-beta software	4.05	6.2%	53%	18%	-12% 22.0%w · 218t
28	NVDA Semis	4.18	3.0%	38%	17%	+7% 48.0%w · 238t
29	CVNA High-beta movers	4.22	5.5%	70%	18%	-9% 16.0%w · 196t
30	TSLA High-beta movers	4.29	3.2%	62%	7%	+2% 41.0%w · 224t
31	CRWD High-beta software	4.36	4.6%	47%	13%	-10% 24.0%w · 222t
32	VST AI / data-center	4.86	4.2%	51%	18%	-7% 26.0%w · 206t
33	CEG AI / data-center	4.98	3.9%	35%	12%	-4% 33.0%w · 216t
34	AFRM High-beta movers	5.11	4.9%	55%	18%	-8% 21.0%w · 216t
35	SNOW High-beta software	5.49	4.0%	36%	18%	-8% 28.0%w · 232t

Duo-fit — fingerprint distance to MRVL/MU. Your duo edge tested as MRVL/MU-specific, so the best duo replacements are names whose fingerprint (vol, range, trend, gaps, liquidity) most closely MATCHES MRVL/MU. Lower = more like MRVL/MU = better duo candidate. A SHORTLIST — a close match still MUST be backtested on the strategy before you trade it.

RSI-leg bench fit = closeness to RSI universe · edge = RSI strategy backtested on it (QC)

#	Ticker	Fit ↓	ATR%	Vol	Gap%	Backtest edge
1	VRT in universe AI / data-center	0.64	6.1%	86%	48%	+1% 25.0%w · 8t
2	SMCI Semis	0.73	6.8%	98%	47%	-0% 33.0%w · 12t
3	ANET High-beta software	1.07	5.7%	61%	40%	-0% 50.0%w · 4t
4	COIN in universe High-beta movers	1.14	5.5%	68%	37%	pending nightly QC
5	MSTR in universe High-beta movers	1.20	5.5%	74%	48%	pending nightly QC
6	DDOG High-beta software	1.25	7.7%	91%	25%	+0% 0.0%w · 0t
7	HOOD in universe High-beta movers	1.38	4.8%	52%	33%	pending nightly QC
8	DELL AI / data-center	1.43	7.6%	81%	42%	-1% 40.0%w · 10t
9	RDDT High-beta movers	1.43	8.4%	108%	23%	pending nightly QC
10	CVNA in universe High-beta movers	1.44	5.5%	70%	18%	pending nightly QC
11	SOFI High-beta movers	1.53	4.0%	64%	27%	pending nightly QC
12	NET in universe High-beta software	1.55	6.2%	53%	18%	-0% 50.0%w · 4t
13	KLAC Semis	1.58	5.5%	67%	62%	-0% 50.0%w · 4t
14	ARM in universe High-beta movers	1.70	8.3%	92%	63%	pending nightly QC
15	APLD in universe AI / data-center	1.75	8.2%	107%	50%	+0% 33.0%w · 18t
16	APP High-beta movers	1.76	6.9%	82%	32%	pending nightly QC
17	QCOM Semis	1.76	3.9%	48%	45%	-0% 67.0%w · 6t
18	OKLO in universe AI / data-center	1.77	7.7%	101%	38%	+0% 30.0%w · 20t
19	GEV AI / data-center	1.87	4.6%	54%	40%	-0% 25.0%w · 16t
20	ON Semis	1.95	5.2%	53%	53%	-0% 50.0%w · 4t
21	LRCX Semis	1.96	6.1%	91%	67%	-0% 50.0%w · 8t
22	CRWD High-beta software	2.03	4.6%	47%	13%	+0% 44.0%w · 18t
23	AVGO Semis	2.22	4.6%	44%	43%	+0% 50.0%w · 8t
24	VST AI / data-center	2.29	4.2%	51%	18%	+0% 50.0%w · 16t
25	NOW High-beta software	2.30	5.1%	60%	50%	pending nightly QC
26	AMD Semis	2.39	5.7%	79%	60%	-1% 17.0%w · 12t
27	AFRM High-beta movers	2.52	4.9%	55%	18%	pending nightly QC
28	TSM Semis	2.62	2.9%	40%	45%	+0% 33.0%w · 6t
29	PLTR in universe High-beta software	2.81	5.2%	115%	33%	+0% 57.0%w · 14t
30	CEG AI / data-center	2.83	3.9%	35%	12%	-0% 33.0%w · 18t
31	CRWV in universe AI / data-center	2.88	10.1%	148%	53%	+2% 80.0%w · 10t
32	TSLA High-beta movers	3.08	3.2%	62%	7%	pending nightly QC
33	SNOW High-beta software	3.48	4.0%	36%	18%	-1% 40.0%w · 10t
34	NVDA Semis	3.64	3.0%	38%	17%	+0% 50.0%w · 28t
35	NBIS in universe AI / data-center	4.72	11.8%	195%	62%	+2% 60.0%w · 10t

RSI-fit — fingerprint distance to the RSI universe. Same idea for the RSI leg: how closely a name matches the profile of your current 21-name RSI universe. Lower = better RSI-leg candidate. Names already IN the universe are marked; the rest are rotation-in candidates. Still a shortlist — backtest before trading.