Research

validate split · 5-session horizon · 3% target · next_open entry

ADR 112 — the first kill criterion fired

0 of 0 scored cells survive Benjamini-Hochberg at q ≤ 0.05. The smallest q-value in the grid is —. 16 cells are suppressed for insufficient effective sample.

These events do not predict five-session returns better than each ticker’s own base rate, at this sample. Every hit rate below is a historical frequency conditional on a past state — not a forecast, and not an edge.

Cell grid

Every cell, suppressed included. A suppressed cell shows the reason it was suppressed and never a rate — v_stats nulls the number in SQL, so no code path here can render one.

Validate

Signal	Bucket	n	n_eff	Hit
bb lower	0-10	1058	15.0	n_eff 14.5 below min_n_eff 30
bb lower	10-20	531	6.0	n_eff 6.0 below min_n_eff 30
bull close	0-10	21	2.0	n_eff 1.5 below min_n_eff 30
bull close	10-20	36	2.0	n_eff 1.9 below min_n_eff 30
confluence low	0-10	175	2.0	n_eff 1.5 below min_n_eff 30
confluence low	10-20	349	3.0	n_eff 2.7 below min_n_eff 30
oversold	0-10	748	7.0	n_eff 6.8 below min_n_eff 30
oversold	10-20	680	6.0	n_eff 5.8 below min_n_eff 30
bb upper	0-10	1010	18.0	n_eff 17.8 below min_n_eff 30
bb upper	10-20	199	3.0	n_eff 3.0 below min_n_eff 30
bear close	0-10	39	3.0	n_eff 3.0 below min_n_eff 30
bear close	10-20	9	1.0	n_eff 0.8 below min_n_eff 30
confluence high	0-10	648	7.0	n_eff 6.8 below min_n_eff 30
confluence high	10-20	80	1.0	n_eff 0.7 below min_n_eff 30
overbought	0-10	1792	15.0	n_eff 14.7 below min_n_eff 30
overbought	10-20	497	4.0	n_eff 3.7 below min_n_eff 30

Train

Signal	Bucket	n	n_eff	Hit	Base	q
bb lower	0-10	6019	88.0	18.8%	17.0%	0.642
bb lower	10-20	1470	19.0	n_eff 19.1 below min_n_eff 30
bull close	0-10	202	10.0	n_eff 9.5 below min_n_eff 30
bull close	10-20	84	2.0	n_eff 2.5 below min_n_eff 30
confluence low	0-10	892	14.0	n_eff 14.5 below min_n_eff 30
confluence low	10-20	806	9.0	n_eff 9.1 below min_n_eff 30
oversold	0-10	4709	70.0	16.4%	17.1%

Era breakdown

Descriptive only — no q-values, because ADR 103 makes era a reporting split rather than a tested family. Era 2024+ is absent: it is the holdout window.

Signal	Bucket	2010-2014train/validate	2015-2019validate/train	2020-2023train/validate
bb lower	0-10	—	—	—
bb lower	10-20	—	—	—
bull close	0-10	—	—	—
bull close	10-20	—	—	—
confluence low	0-10	—	—	—
confluence low	10-20	—	—	—
oversold	0-10	—

Arms

The signal arm carries two rows, and the difference is the finding: ADR 099’s breadth split disagrees in sign on validate.

Arm	Breadth	Return	Sharpe	Max DD	Deployed	Trades
buy_hold	breadth_high	-2.2%	-0.087	20.0%	100%	547
signal	breadth_high	+12.4%	0.314	17.0%	61%	1483
buy_hold	pooled	-2.2%	-0.087	20.0%	100%	547
dca_fixed	pooled	+9.4%	—	11.7%	100%	12
dca_hybrid	pooled	+7.2%	—	13.5%	100%	7
dca_lump	pooled	-2.2%	—	20.0%	100%	1
dca_signal	pooled	+11.7%	—	9.2%	100%	418
signal	pooled	+17.6%	0.437	22.6%	100%	4448
trim	pooled	+1.7%	-0.011	15.3%	100%	1786

Random-entry null

Drawdown slice

ADR 015 calls this the project’s central claim: signals in deeper drawdowns resolve up more often. Session 14 measured every interval crossing zero.

0 of 0 hit-rate intervals contain their own baseline.

0-10% drawdown

Signal	Bucket	n	n_eff	Hit
bb lower	0-10	1058	15.0	n_eff 14.5 below min_n_eff 30
bull close	0-10	21	2.0	n_eff 1.5 below min_n_eff 30
confluence low	0-10	175	2.0	n_eff 1.5 below min_n_eff 30
oversold	0-10	748	7.0	n_eff 6.8 below min_n_eff 30
bb upper	0-10	1010	18.0	n_eff 17.8 below min_n_eff 30
bear close	0-10	39	3.0	n_eff 3.0 below min_n_eff 30
confluence high	0-10

Holdout is not on this page and cannot be requested from it. It is evaluated once, at the end, and published whatever it says.