Verdict: Bottleneck Beneficiaries (AI build-out) (M-024) is rated “NOT VALIDATED — pending” in its validation review dated 8 Oct 2026.
Key result: Universe vs all liquid US names, 12m EW: +11.0 pp/yr [7.9, 14.4], 8 of 10 yrs.
Key result: Q5 precision, top-10 (bar 0.37): best 0.30: NOT MET.
How it was tested: training data not stated, test data 2016→2025; out-of-sample design (how it was checked on data it never saw while being built): dated filings rule; no fit.
Horizon: 12m. Coverage: US filers whose own filings use core AI-theme language (point-in-time).
Forward record: 1 published list is being tracked against SPY; none has reached its 252-session horizon yet (the first matures on 7 Oct 2027).
Stage: Shadow. Version v1; the hub record is as of 6 Nov 2026.
The verdict
NOT VALIDATED — pending. Pending the independent review; until it reports the list is SHADOW and nothing is sized from it.
The validation record also states the rating as “SOUND WITH CONDITIONS (decision review; outputs NOT VALIDATED as a rank — SHADOW)”.
Development summary: Universe selection is the measured edge (+11.0 pp/yr [7.9, 14.4] vs all liquid US names, 8 of 10 years); the order inside it is M-001's price score; constraint, deals and backlog are JUDGMENT, never ranked.
What the review allows
Read the universe and the thesis strings as research input
Shadow-track the liquid universe and the top-10 by M-001 score on the Hub
What the review rules out
Size or trade from the list before the validation reports
Read any JUDGMENT column as a rank, a score or a validated signal
What it predicts
Item
Detail
Question it answers
Which US companies are positioned to be paid by the AI build-out's bottlenecks over the next 6-12 months?
Method
Filings-rule universe + M-001
How it works
Rule: dated filings-text theme universe (SEC full text, point-in-time), ranked inside by M-001's validated price score; JUDGMENT columns shown apart, never ranked
Inputs
Filings: data centre + AI/GPU language, Filings: power, memory, optics, cooling, M-001 price score inside the universe, Constraint status (JUDGMENT), Verified deals ÷ mcap (JUDGMENT)
Output
Monthly universe list in M-001 order + thesis per name; JUDGMENT apart
Horizon
12m
Coverage
US filers whose own filings use core AI-theme language (point-in-time)
Markets
US
Model type
Model · Rule
Stage
Shadow
Version
v1, first recorded 8 Oct 2026
Validation evidence
Training data: not stated. Test data: 2016→2025. Out-of-sample design: dated filings rule; no fit.
Headline results from the model's development record, as the hub publishes them.
Measure
Result
Compared with
Note
Universe vs all liquid US names, 12m EW, 2016-25
+11.0 pp/yr [7.9, 14.4]
market EW
8 of 10 years > 0; 2025 +60 pp; survivor-only, hindsight keywords
M-001 top-10, broad US (the borrowed order)
+26.5 pp [13.1, 45.5]
US floored EW
Price sorts inside the active PIT universe
+1 to +3 pp
universe EW
the order adds little inside the theme
Q5 precision, top-10, 2019-25
best 0.30
bar 0.37 (2x base)
NOT MET
Deal announcements, days 0..+5
+14.9% [+9, +22]
theme
no drift after day 5: judgment only
The hub's standard metrics for this model class (only those the record measured).
Metric
Sample / window
Compared with
Value
Range (CI)
Excess return at horizon (headline)
12m · universe EW · cohorts 2016-25 (8 of 10 years > 0)
all liquid US names, equal-weight
+11.00 pp
[7.9, 14.4] pp
Precision at k
Q5-12m · top-10 per month · strictly active PIT universe 2019-25 · best configuration tested
bar 37% (2x base)
30.00%
n/a
Forward record: 1 published list is being tracked against SPY; none has reached its 252-session horizon yet (the first matures on 7 Oct 2027).
Limitations
Survivor-only history: every absolute figure is an upper bound.
Keyword sets chosen in hindsight: the universe rule is mechanical and dated, but its design knew the AI theme won.
US only until a non-US point-in-time universe exists (exchange filings: TDnet, DART, MOPS).
2025 dominates every fundamental result; ex-2025 they average about 0.
Precision bar NOT met (best Q5 precision 0.30 vs 0.37); no fundamental or news feature ranks.
Judgment inputs are thin: the R1 map and the deals database were assembled in 2026 (not point-in-time), and 54 contract events are too few for a ranking signal.
Questions
Is the Bottleneck Beneficiaries (AI build-out) validated?
Its validation review dated 8 Oct 2026 rated it NOT VALIDATED — pending. Pending the independent review; until it reports the list is SHADOW and nothing is sized from it.
What does the Bottleneck Beneficiaries (AI build-out) predict?
It answers: Which US companies are positioned to be paid by the AI build-out's bottlenecks over the next 6-12 months? Output: Monthly universe list in M-001 order + thesis per name; JUDGMENT apart. Horizon: 12m. Coverage: US filers whose own filings use core AI-theme language (point-in-time).
How was the Bottleneck Beneficiaries (AI build-out) tested?
Training data not stated, test data 2016→2025, out-of-sample design: dated filings rule; no fit. Universe vs all liquid US names, 12m EW: +11.0 pp/yr [7.9, 14.4], 8 of 10 yrs; Q5 precision, top-10 (bar 0.37): best 0.30: NOT MET; Price sorts inside the active universe: +1 to +3 pp. Forward record: 1 published list is being tracked against SPY; none has reached its 252-session horizon yet (the first matures on 7 Oct 2027).