Anatomy of a signal:
NVDA 65 → 87
On 5 January 2026 the engine flagged NVDA at score 65 — top decile, BUY-tier conviction. Six weeks later that score read 87 and the price had moved 34%. This page walks through what the signal looked like the day it fired, what happened next, and what a user would have seen in real time. No math wall — story first.
5 January 2026 — the day the engine flagged it
Markets opened on a Monday after the Christmas drift. SPY was sideways. The engine ran its 06:00 UTC universe sweep across 374 tickers. NVDA came out at 65 — top decile of the universe that day, with a verdict of BUY and a 90% conformal interval of [56, 78].
Five factors out of twelve carried the score. Three made up most of the conviction. Here's what the user would have seen on /stocks/NVDA that morning:
None of these are individual genius-takes. Each one is a published academic factor — Novy-Marx 2013 for quality, Jegadeesh-Titman 1993 for momentum, Bernard-Thomas 1989 for post-earnings drift. The engine combines them; the conviction comes from the joint setup, not any single number.
Why the engine cared about THIS day
The interesting bit isn't that any single factor was elevated. Quality of 78 in isolation isn't newsworthy — top-quartile but not extreme. What flagged this signal was the joint configuration:
- Quality 78 + Momentum 71: Asness-Frazzini-Pedersen 2014 — quality and momentum on the same name historically pairs into the strongest cross-sectional bull setup. Either alone is fine; together is structural.
- PEAD 62: NVDA was 18 trading days into a post-earnings window. Bernard-Thomas' drift hadn't completed.
- Risk-on regime: BOCPD posterior placed q(risk_on) at 0.71 with stability 0.84 — mature regime, not a fresh transition. Momentum amplifier was at full strength.
- Tail-dependence λ_U = 0.34: Quality and Momentum were genuinely co-moving in the tails for this regime, not just on average. The pattern wasn't a coincidence.
The Confluence Engine fired the QUALITY COMPOUNDER pattern: quality + value + accruals on the same name in a risk-on regime, with measured tail-alignment. That added a +5 boost to the raw weighted composite — a small δ, but it's the boost that academic literature backs (Novy-Marx's gross-profitability premium specifically pairs with value).
What the user would have done
On the ticker page they'd have seen the signal card above plus a Kelly-sizing block. With score 65, regime risk-on, and a 90% interval of [56, 78], the engine's recommended position size was 2.4% of portfolio — half-Kelly, capped. Not a max-aggressive bet. The interval width (78−56=22) means the engine wasn't maximally certain; sizing reflected that.
A retail user following the signal would buy NVDA on 5 January at the open and hold. The dashboard would surface two things over the next six weeks:
- Daily score updates — visible drift from 65 → 71 → 76 → 82 → 87 across the period
- Confluence pattern persistence — QUALITY COMPOUNDER stayed lit through 14 February before transitioning to PRICED FOR PERFECTION (warning sign — hold, but don't add)
Six weeks later
By 16 February 2026 the price had moved from $87 to $116 — roughly a 34% gain on the position. The score read 87. The interval had narrowed to [80, 94]. Confidence: high.
The engine's next signal: PRICED FOR PERFECTION (Shiller 2000 asymmetric-downside pattern). Quality factor still elevated, but value had collapsed (P/E expansion outpacing earnings revisions). The engine quietly downgraded the verdict from BUY-tier to MIXED-tier and widened the interval. No alarm bell — just a soft signal to stop adding.
A user who saw all this in real time would have:
- Bought 2.4% on 5 January at score 65
- Held through the drift, watching score climb
- Stopped adding when QUALITY COMPOUNDER transitioned out
- Optionally trimmed when PRICED FOR PERFECTION fired
What this story is and isn't
Is: a faithful walkthrough of how an APEX signal reads — the factor breakdown, the conformal interval, the Kelly sizing, the confluence patterns. The structure is exactly what a real signal looks like on /stocks/NVDA.
Isn't: a backtested track record. The 5 January 2026 signal date and the specific factor numbers are illustrative — we don't publish backtest numbers until the calibration window matures (16 May 2026). After that the engine's real signals get tracked on /track-record with measured forward returns.
We chose to tell the story this way — explicitly illustrative — because the alternative is a marketing page with cherry-picked real winners. That's less honest. Pre-track-record, you should see how the math reads, not which winners we picked.
Try it on a real ticker
Every ticker on the universe has the same five-section signal page that the story walks through. Pick something and read it.
Or read /methodology for the academic stack behind every signal, or /learn for plain-English definitions of every term used in this story.