I bet on Brown, Sechan, Belsky. The data says Steve Weiss.

I bet on Josh Brown, Rob Sechan, and Brian Belski. The data says Steve Weiss. For the last four months, a local pipeline at Evenrock has been swallowing every CNBC Halftime Report and Fast Money clip published between January 2 and April 30, 2026 — 365 clips, 907 named tickers, 738 scored commentary rows — and grading every pick against forward returns versus SPY at 30‑day and 90‑day horizons. I started the project with a working hypothesis about who actually knows what they're talking about. The leaderboard disagrees. This isn't a story about Brown being bad — he is, by my own rubric, the best communicator on the show. It's a story about the gap between commentary I want to listen to and picks I would actually want to own.

The leaderboard

Sorted by conviction‑weighted 90‑day alpha vs. SPY. Minimum 10 picks. Commentary score is a 0–10 rubric (clarity, evidence, falsifiability, originality) averaged across that pundit's clips.

And the three I came in betting on:

What the data says

Joe Terranova is the leader, and it isn't close on volume‑adjusted basis. Fifty picks, an 18% 90‑day hit rate against SPY, and a conviction‑weighted alpha of +2.79%. He is not the best communicator on the show — the rubric has him at 5.8, mid‑pack — but he sizes his calls correctly. When Joe says he's adding, the trade tends to work. When he hedges, he hedges. The signal-to-noise on his book is the cleanest in the dataset.

Tim Seymour is the only top‑five name with a hit rate above 30%. Fifty‑two picks, 30.8% beat rate, +2.48% conv‑weighted. Seymour talks his book in ways that read as repetitive on TV — the same names come back week after week — but the names keep working. Repetition turns out to be a feature when the underlying thesis is right.

Steve Weiss is the headline. He has 18 picks, the smallest book of any regular in the top ten, and his 90‑day window hasn't fully matured — most of his calls land in February and March, and the 90‑day forward is still rolling. But on the 30‑day horizon he is the standout of the period: a high beat rate on a small, concentrated book, with conviction stamps on names that moved fast. He grades a 4.8 on commentary — he's terse, sometimes cranky, doesn't explain himself — and yet here we are. The picks work.

Rob Sechan, who I came in expecting to crown, lands in the middle of the pack. Twelve picks, zero conviction‑weighted alpha, 5.0 commentary. Not bad. Not interesting. The hypothesis got falsified the cleanest way possible — by an unremarkable result.

Josh Brown is the most interesting row in the entire table. A 6.4 commentary score puts him at the top of the named pundits in this dataset on the rubric I care about: he frames trades clearly, names what would falsify him, references real flow. He is the person I would want to learn markets from. He also has 75 picks and a conviction‑weighted 90‑day alpha of −0.74%. That's not a small sample. The communicator I rate highest is, on this window, the worst‑performing high‑volume pundit on the show. I have to sit with that.

Methodology

The pipeline runs locally in Python. A scraper pulls Halftime and Fast Money clips off CNBC's site, Whisper transcribes audio, Claude extracts named tickers with direction (long/short), conviction (1–3), and a raw quote, and yfinance scores each pick at 30/90/180/365 days versus SPY from the clip publish date. A separate prompt grades each clip's commentary on a 0–10 rubric. The two axes are deliberately separate. Commentary quality is about how well you explain a view; pick performance is about whether the view paid. They turn out, here, to be uncorrelated.

The leaderboard above is the corrected version. The first run mis‑heard "Sechan" as "Seachen" and dropped half his picks under a phantom commentator named "Rob." A human‑review pass on transcripts caught the gap, the name was canonicalized, and his full 12‑pick book showed up. There are still rows in the raw data labeled simply "Josh," "Jim," "Joe," "Weiss," "Malcolm," and "Karen" — first‑name fragments where Whisper didn't catch the surname and the on‑screen chyron wasn't OCR'd. Those rows are excluded from the headline leaderboard but kept in the raw set. The "Unknown" bucket — 64 picks, anonymous — is a reminder that ~7% of named picks still lack attribution, and is treated as a control rather than a competitor.

Best three calls of the period

Worst three calls of the period

The best book has a technician, a value investor, and an industrial picker. The worst book has two "they need to do AI better" longs, which is the 2026 version of the 2010s "they need to do mobile better" thesis — vibes dressed as catalyst.

What to watch in Q2

The April picks haven't fully matured. Anything after February 1 still has 90‑day forward returns rolling in live, and the leaderboard will move — possibly meaningfully, given how concentrated the top is in two or three calls per pundit. Steve Weiss's standing in particular depends on how his March/April book ages. Joe Terranova's lead is the most robust to time, because his book is the largest among the leaders. Josh Brown's deficit, similarly, is hard to walk back at 75 picks. The data is signal in progress, not a final verdict — but it's already enough to retire the hypothesis I came in with. Brown stays in my podcast queue. Terranova, Seymour, and Weiss go in the trade journal.