AI & Humanoid Screen

1. Is it on sale?%B < 0.20

%B says where price sits inside its normal trading channel. Below 0.20 means it has pulled back to the bottom of that channel — on sale. Above 1.00 is extended: too stretched to buy, and it returned 2.65 pts below simply holding.

2. Above or below the 200-day?a trade-off, not a gate

Two studies disagree, so this is not a filter the verdict column applies. 46 large caps, 15y, fwd 60d: dips above the 200-day hit 71% vs 68% below — better odds, but below-200d averaged slightly more (+7.16% vs +6.96%). The AI complex, 2023+: dips below the 200-day were the best bucket of all, +7.49 pts vs the null. The Meta trade in the box below was 11.8% below its 200-day. Above = steadier; below = bigger and lumpier. Pick per your tolerance.

3. Does it survive bad timing? 2y pos ≥ 80% AND 2y p10 > -10%

Both halves are gates, and 2y pos is often the binding one. p10 is the 10th-percentile outcome across past 2-year holds — your unlucky-but-realistic case; a −50% p10 means one in ten 2-year holds halved. These are the exact thresholds the verdict column applies, not a rule of thumb.

The exit is a rule, not a guess — but be clear which rule. Everything in these tables is a fixed hold: the +3.47 / −2.65 / +7.49 pt figures are 60 trading days, entry at the next open, and the 2y pos / 2y p10 columns are 504-day holds. Nothing on this page measures a middle-band exit.
Selling on the first close above the 20-day middle band is the rule used by the separate SOXX/SOXL/SOXS band strategy in the morning brief, and it is a reasonable discipline — it just is not what produced these numbers. Reproduce the tabulated edge by holding the horizon, not by exiting at the midband.

Why there are two signals. Meta launched Muse on 2026-09-08 and ran +20.3% before the first analyst upgrade on 09-21 — following that upgrade captured 2% of the move. Replaying these rules over those sessions:

%B < 0.20 dip rule	2026-08-21 @ 549.47	+35.4%	18 days before the launch, and META was 11.8% below its 200-day
move + volume spike	2026-09-09 @ 653.17	+13.9%	the day after the launch
follow the analyst	2026-09-21 @ 741.25	+0.4%	the move was over

The dip rule caught it by accident — it was buying a drawdown, not predicting a product. The spike rule reacts to the catalyst, late but not uselessly. They are labelled separately so a reaction is never mistaken for a setup.

Semis — SOXX / SOXL / SOXS

The 3x pair moves with SOXX, so read SOXX's %B and size with SOXL/SOXS's ATR. SOXS is inverse: its %B means the opposite.

Ticker	Close	1d	%B	vs 20d	vs 200d	range	ATR	$vol/d	2y pos	2y p10	msgs	Verdict
SOXX semis_etf	565.72	-1.23%	1.01	+8.4%	+27.0%	77%	3.2%	4.3B	95% 4w	+7.3%	—	extended
SOXL semis_etfleveraged	146.25	-3.75%	0.99	+23.3%	+28.5%	43%	8.8%	8.1B	76% 4w	-52.2%	—	leveraged · not rated
SOXS semis_etfleveragedinverse	33.66	+3.82%	-0.00	-25.4%	-81.8%	0%	13.5%	2.6B	0% 4w	-98.7%	30	leveraged · not rated

What is moving SOXX

The ten holdings behind SOXX, their fund weight, and how much of the day each one contributed (fund weight × its move, so together they explain about 61% of the fund).

Weight above its own 20-day87%

Weight that rose today8%

Top-10 contribution today-0.81%

Largest moverMU -0.20% · 23% of gross movement

Tiles are the nightly snapshot, as of 2026-09-23. Refresh live re-prices the rows; the tiles and the chart only update when all ten come back for the same session.

9 of 10 holdings have a real 1-day return for 2026-09-23 (the tiles cover 93% of the top-10 fund weight); excluded: KLAC (no bar for 2026-09-22 (its 1d would span a gap)).

Weighted daily return of the top ten (renormalised to 100%), last 20 sessions · most recent on the right

Last 5 sessions — return: +3.6%, +3.0%, +4.8%, +1.9%, -1.4% · weight rising: 100%, 92%, 100%, 100%, 8%

Ticker	Fund wt	1d	Contrib	5d	20d	vs 20d
NVDA	9.43%	-1.47%	-0.138%	+5.4%	+6.0%	above 20d
MU	8.91%	-2.22%	-0.197%	+15.7%	+14.9%	above 20d
AMD	8.23%	-1.47%	-0.121%	+19.9%	+28.3%	above 20d
AVGO	7.48%	-2.62%	-0.196%	+4.7%	-0.3%	BELOW 20d
INTC	5.09%	-1.02%	-0.052%	+21.3%	+40.1%	above 20d
MRVL	4.66%	-0.56%	-0.026%	+13.6%	+8.5%	above 20d
TSM	4.65%	-1.20%	-0.056%	+6.9%	+7.3%	above 20d
AMAT	4.59%	+0.41%	+0.019%	+14.2%	-1.2%	above 20d
LRCX	4.27%	-1.08%	-0.046%	+14.3%	-2.2%	above 20d
KLAC no 1d	4.12%	—	—	+12.2%	+2.5%	n/a

How to use this for SOXL vs SOXS. SOXX %B 1.01 is above 1.00 — the stall zone: −0.93 pts vs base at 5 days (n=189, 10 years). That is a stall, not a short — nothing measured here favours SOXS.

Breadth is context, not a trigger. Reproduce every number here with scripts/soxx_breadth_study.py (8 years of SOXX, entry at the next open). Broad strength has no edge: top-10 weight above its 20-day > 80% was +0.04 pts vs the null at 20 days (n=727; +0.00 at 5d, −0.03 at 1d). Weakness in the holdings (breadth < 20%, or their composite %B < 0.20) pointed the same way as SOXX's own dip signal but smaller (about +0.7 to +0.9 pts vs +1.3) and rests on a few hundred overlapping observations, so treat it as noise-level: it adds nothing you cannot read off SOXX's own %B. Use this table to see why SOXX moved, broad or one name, and take the direction call from SOXX's %B. SOXL and SOXS are daily-reset products and are not rated on this page.

Mega caps

META, GOOGL, MSFT, AAPL, AMZN, NFLX — tracked every day whether or not they signal.

Ticker	Close	1d	%B	vs 20d	vs 200d	range	ATR	$vol/d	2y pos	2y p10	msgs	Verdict
META ndx100ai_platformmegacap	744.10	+1.02%	0.96	+16.0%	+19.2%	94%	3.5%	10.5B	79% 4w	-28.1%	—	neutral
GOOGL ndx100ai_platformmegacap	337.83	-3.80%	0.34	-1.2%	+0.1%	61%	2.7%	8.8B	90% 4w	-0.1%	—	neutral
MSFT ndx100ai_platformmegacap	500.59	+0.52%	0.55	+0.2%	+16.3%	80%	2.2%	11.5B	96% 4w	+10.0%	—	neutral
AAPL ndx100megacap	337.02	-0.80%	0.79	+3.1%	+17.5%	97%	2.1%	13.6B	99% 4w	+16.2%	—	neutral
AMZN ndx100megacap	249.27	-2.24%	0.18	-2.3%	+3.6%	59%	2.4%	8.9B	79% 4w	-25.5%	—	DIP · FAILS HOLD
NFLX ndx100megacap	71.36	-1.11%	0.06	-8.1%	-16.0%	7%	3.2%	2.5B	77% 4w	-39.2%	—	DIP · FAILS HOLD

Buy zone

%B below 0.2, liquid, and passes the 1–3 year hold gate (≥80% of rolling 2-year windows positive with a 10th percentile above -10%).

Ticker	Close	1d	%B	vs 20d	vs 200d	range	ATR	$vol/d	2y pos	2y p10	msgs	Verdict
CTAS ndx100	191.97	-2.49%	-0.17	-4.1%	+2.8%	55%	2.1%	380M	99% 4w	+23.3%	3	BUY ZONE
PAYX ndx100	104.49	-9.15%	-0.17	-12.8%	+1.9%	48%	3.2%	312M	88% 4w	-2.1%	1	BUY ZONE SPIKE -9.1% · 3.1x
AEP ndx100	118.40	-1.36%	-0.08	-3.3%	-5.6%	43%	1.7%	470M	86% 4w	-2.5%	1	BUY ZONE
PCAR ndx100	112.09	-2.30%	-0.07	-8.2%	-6.7%	45%	2.6%	341M	98% 4w	+8.1%	0	BUY ZONE
XEL ndx100	70.66	-1.66%	-0.06	-5.2%	-8.3%	4%	1.9%	393M	82% 4w	-6.9%	0	BUY ZONE
CSCO ndx100	106.43	-0.01%	0.02	-2.9%	+10.6%	64%	2.8%	2.0B	82% 4w	-3.2%	30	BUY ZONE
TMUS ndx100	165.66	+0.24%	0.03	-6.9%	-13.2%	1%	3.3%	797M	95% 4w	+8.9%	—	BUY ZONE
BKNG ndx100	155.90	-5.07%	0.07	-14.4%	-15.1%	4%	4.4%	1.1B	84% 4w	-5.0%	—	BUY ZONE
ADP ndx100	263.68	-2.32%	0.09	-4.1%	+12.4%	78%	2.1%	580M	93% 4w	+4.2%	—	BUY ZONE
CCEP ndx100	99.69	-0.84%	0.10	-5.1%	+1.9%	57%	2.2%	183M	94% 4w	+5.4%	—	BUY ZONE
VRSK ndx100	171.80	-0.13%	0.12	-6.0%	-10.1%	17%	3.6%	261M	83% 4w	-9.1%	—	BUY ZONE
CSX ndx100	47.19	+2.61%	0.19	-3.8%	+8.0%	71%	2.2%	507M	90% 4w	-0.2%	—	BUY ZONE

Catalyst spikes today

A move of 5%+ on 2×+ the 20-day average volume. This is a reaction, not a setup — something happened, and the rule does not know what or which direction it resolves.

Ticker	Close	1d	%B	vs 20d	vs 200d	range	ATR	$vol/d	2y pos	2y p10	msgs	Verdict
PAYX ndx100	104.49	-9.15%	-0.17	-12.8%	+1.9%	48%	3.2%	312M	88% 4w	-2.1%	1	BUY ZONE SPIKE -9.1% · 3.1x
ABNB ndx100	149.58	-10.35%	-0.12	-14.7%	+5.2%	48%	3.8%	624M	—	—	3	dip · unmeasured SPIKE -10.3% · 3.0x

Near the band / neutral

Dips that fail the hold gate, plus names with no measured edge either way.

Ticker	Close	1d	%B	vs 20d	vs 200d	range	ATR	$vol/d	2y pos	2y p10	msgs	Verdict
FANG ndx100	185.65	-1.91%	-0.11	-7.1%	+2.1%	67%	3.5%	382M	68% 4w	-49.1%	1	DIP · FAILS HOLD
KHC ndx100	23.79	-2.38%	-0.04	-3.9%	+2.3%	49%	2.8%	357M	37% 4w	-58.1%	2	DIP · FAILS HOLD
INTU ndx100	286.80	-5.70%	-0.03	-13.0%	-29.5%	7%	4.9%	1.2B	84% 4w	-10.2%	30	DIP · FAILS HOLD
TTD ndx100	12.68	-3.79%	-0.02	-9.5%	-46.1%	0%	5.4%	268M	70% 4w	-46.8%	30	DIP · FAILS HOLD
CSGP ndx100	28.26	-2.69%	0.00	-8.2%	-32.1%	2%	4.4%	184M	62% 4w	-22.7%	0	DIP · FAILS HOLD
PEP ndx100	130.18	+0.46%	0.03	-4.6%	-10.3%	2%	1.7%	1.1B	78% 4w	-11.1%	24	DIP · FAILS HOLD
NFLX ndx100megacap	71.36	-1.11%	0.06	-8.1%	-16.0%	7%	3.2%	2.5B	77% 4w	-39.2%	—	DIP · FAILS HOLD
CPRT ndx100	28.87	-0.14%	0.06	-8.7%	-14.9%	9%	3.7%	302M	87% 4w	-15.9%	—	DIP · FAILS HOLD
SBUX ndx100	94.14	-0.80%	0.11	-7.4%	-3.4%	55%	2.6%	657M	75% 4w	-8.6%	—	DIP · FAILS HOLD
MELI ndx100	1,799.24	-1.17%	0.11	-5.6%	-2.7%	26%	3.3%	744M	78% 4w	-24.0%	—	DIP · FAILS HOLD
ROP ndx100	366.36	+0.26%	0.11	-8.0%	-1.4%	26%	2.8%	285M	85% 4w	-15.1%	—	DIP · FAILS HOLD
CMCSA ndx100	22.55	+0.58%	0.12	-10.1%	-14.6%	6%	3.3%	667M	57% 4w	-26.8%	—	DIP · FAILS HOLD
ADBE ndx100	240.69	+1.02%	0.16	-8.9%	-8.7%	28%	4.1%	1.2B	64% 4w	-43.1%	—	DIP · FAILS HOLD
AMZN ndx100megacap	249.27	-2.24%	0.18	-2.3%	+3.6%	59%	2.4%	8.9B	79% 4w	-25.5%	—	DIP · FAILS HOLD
TTWO ndx100	206.32	-1.71%	0.19	-5.2%	-8.8%	23%	3.7%	505M	76% 4w	-25.0%	—	DIP · FAILS HOLD
ALB humanoid_materials	113.47	+0.49%	0.22	-8.8%	-26.9%	25%	4.6%	269M	49% 4w	-62.4%	—	neutral
AXON ndx100	450.61	-0.32%	0.23	-12.0%	-10.3%	24%	5.9%	449M	98% 4w	+21.9%	—	neutral
ORCL ai_platform	144.56	-3.11%	0.28	-3.5%	-12.2%	15%	4.9%	4.3B	98% 4w	+14.7%	—	neutral
RRX humanoid_disclosed	152.42	+1.12%	0.28	-3.1%	-18.9%	22%	4.2%	203M	80% 4w	-7.0%	—	neutral
BKR ndx100	58.03	+0.35%	0.28	-4.3%	-1.1%	56%	3.1%	446M	65% 4w	-44.0%	—	neutral
ODFL ndx100	177.72	+1.35%	0.28	-4.2%	-10.4%	43%	3.3%	311M	88% 4w	-5.7%	—	neutral
VRTX ndx100	516.34	+1.10%	0.29	-2.8%	+9.7%	77%	2.4%	582M	92% 4w	+1.7%	—	neutral
MDLZ ndx100	61.44	+2.11%	0.30	-0.9%	+4.5%	75%	2.1%	479M	79% 4w	-7.6%	—	neutral
TRI ndx100	98.45	+3.16%	0.31	-3.5%	+0.6%	28%	4.4%	152M	90% 4w	-2.3%	—	neutral
ADSK ndx100	217.15	-0.78%	0.31	-6.2%	-10.5%	21%	4.1%	450M	74% 4w	-26.7%	—	neutral

Extended

%B above 1.00 — the stall zone, −2.65 pts vs the null at 60 days. Not a short signal; a not-today signal.

Ticker	Close	1d	%B	vs 20d	vs 200d	range	ATR	$vol/d	2y pos	2y p10	msgs	Verdict
ETN ai_power	438.76	+0.76%	1.01	+7.0%	+14.4%	86%	3.4%	774M	99% 4w	+14.0%	—	extended
SOXX semis_etf	565.72	-1.23%	1.01	+8.4%	+27.0%	77%	3.2%	4.3B	95% 4w	+7.3%	—	extended
TXN ndx100	272.62	+0.65%	1.03	+4.2%	+12.1%	68%	3.2%	1.8B	88% 4w	-1.8%	—	extended
AME humanoid_association	246.45	+2.16%	1.03	+3.9%	+8.3%	85%	2.2%	276M	99% 4w	+15.3%	—	extended
ZS ndx100	214.45	+3.53%	1.04	+17.6%	+27.7%	44%	4.6%	411M	67% 4w	-36.5%	—	extended
MPWR ndx100	1,355.47	+6.09%	1.04	+10.1%	+7.2%	60%	4.1%	965M	97% 4w	+22.7%	—	extended
ADI ndx100	385.23	+0.58%	1.04	+5.2%	+9.1%	73%	3.0%	1.3B	98% 4w	+15.1%	—	extended
ENTG ai_equipment	150.93	+5.44%	1.06	+9.8%	+16.6%	71%	4.7%	323M	73% 4w	-20.8%	—	extended
STX ndx100ai_memory	923.86	+5.30%	1.06	+10.3%	+47.8%	81%	5.6%	3.9B	80% 4w	-13.2%	—	extended
NXPI ndx100humanoid_association	235.44	+1.30%	1.06	+4.3%	-3.2%	37%	3.3%	852M	71% 4w	-11.7%	—	extended
BIIB ndx100	227.53	+3.86%	1.09	+4.2%	+17.7%	100%	2.7%	185M	33% 4w	-45.3%	—	extended
TEL humanoid_association	214.39	+2.09%	1.11	+4.2%	+0.3%	39%	2.8%	398M	83% 4w	-7.9%	—	extended
WBD ndx100	30.76	-0.13%	1.19	+7.7%	+11.6%	100%	1.7%	513M	40% 4w	-61.0%	—	extended
ALNT humanoid_association	113.60	+3.89%	1.22	+18.8%	+49.7%	99%	5.0%	22M	70% 4w	-31.0%	—	extended

Does the dip rule hold up as a quarterly rotation?

Buy the dip names at each quarter-end, hold to the next, versus the same number of names drawn at random from the same list (2000 draws), after 0.20% costs. Thresholds were written before each run. Neither variant cleared them, so nothing here changes the verdicts above. The universe is today's AI leaders, which flatters dip entries; the mean is carried by one quarter (2026-03-31) in v1.

Test	signal qtrs	mean net excess	qtrs positive	ex best qtr	p vs random	result
v1 · dip (%B < 0.20)	23	+4.2 pts	52%	+1.0 pts	0.071	not promoted (failed: P1, P4)
v2 · dip + above 200-day	18	+3.3 pts	39%	-4.8 pts	0.259	not promoted (failed: P1, P3, P4, P5, P6)

The 1–3 year durability gate cannot be tested point-in-time on 10 years of data (needs 2016 prior sessions); it is judged forward only. Method and thresholds: docs/experiments/2026-09-24-preregistration-qr-s1-v{1,2}.md.