Congressional stock disclosures are a popular signal and a badly measured one. The usual treatment counts winners. This counts alpha: every disclosed BUY in the desk's mirror, measured from the member's own transaction date to the 26 August 2026 close, minus SPY's return over the identical window.

The headline result is not the one the internet expects.

The scoreboard

Pelosi's disclosed equity buys have not beaten the index. A 75% win rate with −1.4% median alpha is what buying mega-cap technology in a rising market produces. She owns GOOGL, AMZN, NVDA, AAPL, AVGO, INTC — the index's largest constituents. Winning three times out of four while matching SPY is beta, not edge.

Read the important caveat before concluding anything: this mirror records equity-equivalent positions only. Pelosi's disclosures are frequently long-dated call options, and leverage is precisely what an equity mirror cannot represent. The correct statement is that her directional stock selection shows no measurable edge — not that her portfolio didn't make money. Those are different claims and only the first is supported here.

The member with an actual edge is Michael McCaul, who nobody memes about. Twenty-four buys, 75% win rate, +14.2% median alpha — the largest sample in the set and the only one where the number is both large and repeated. Mastercard three times (+19.7% to +21.7%), S&P Global four times, Danaher twice (+23.2%, +30.0%), Arista +18.8%.

Wasserman Schultz's +364% is a lottery ticket, not a signal. Two buys. Viasat at $10.23, now $71.51. One trade decides the entire number, and n=2 supports no conclusion at all.

The part that matters if you were thinking of copying anyone

Disclosure is not instantaneous. The STOCK Act allows 30 days from notice and 45 days from the trade, so the public learns after the fact. Re-running the same measurement from 45 days after each transaction — roughly the earliest a follower could act — gives the return that was actually available:

McCaul's edge survives the delay but shrinks by roughly 60%, from +14.2% to +5.7%. Everyone else lands within a couple of points of the index in both columns. The copyable version of this signal is much weaker than the raw version, which is the normal fate of a lagged signal.

What they bought, and what it says about the AI trade

The two most active books took opposite sides of the buildout, and that contrast is more useful than either book alone.

Gottheimer bought the supply chain. His 2026 purchases include SNDK, MU, LITE, FN, AMD and ADI — memory and optical, which turned out to be the layer that got paid: SNDK +3066.6% 1y, MU +706.5% 1y, LITE +653.5% 1y. He sold FN on 13 July 2026; FN now sits 41.5% below its high while LITE, the same layer, is 13.5% off. Despite owning the right layer, his median alpha is +1.7% — he bought late in the move.

Pelosi bought the platforms and the electricity. GOOGL, AMZN, NVDA, AAPL, AVGO, INTC, TEM, UBER — plus VST (January 2025, again January 2026) and BE, purchased twice in July 2026 and the most recent new position in the book. There is no memory, optical or packaging exposure at all. VST is −28.2% 1y; BE is +336.9% 1y.

That second point connects to the desk's current work: the electricity layer is the one NVIDIA named as its binding constraint in the 26 August 10-Q, and it is the layer that has not re-rated. Pelosi's two most recent Bloom Energy purchases are a bet on it. Whether that is foresight or coincidence, the disclosure is a data point on a thesis the desk arrived at independently from the filings.

McCaul bought neither. Mastercard, S&P Global, Danaher, Arista, Coherent, EQT, JPMorgan — payments, data, life-science tools, one networking name. The best risk-adjusted book in the set has almost no AI-buildout concentration.

Method and its limits

• Source: the desk's congressional trade mirrors, built from House Periodic Transaction Report filings. Entry prices are the close on the transaction date recorded in the filing, not the disclosure date.
• Alpha is per-trade return to the 26 August 2026 close minus SPY's return from the same start date. Median, not mean, so a single outlier cannot carry a member.
• BUYs only. Sells are excluded, so nothing here measures exit skill or realised P&L.
• No position sizing. Disclosures give a range, not an amount. A $1M–5M purchase counts the same as a $1K–15K one.
• No options. The single largest limitation, and it bears most on Pelosi.
• Samples are small. Only McCaul (24), Khanna (23) and Pelosi (20) have enough trades to say anything, and 20 is still thin.
• Household trades may be advisor-directed rather than member-directed.

What would change the read

• McCaul's alpha surviving another 20 disclosed buys — that would move him from interesting to tracked.
• Any member's book showing concentration in the electricity layer before it re-rates.
• An options-aware mirror, which would materially change the Pelosi number in either direction.

Cadence

This is a repeatable format. The underlying pull already runs as a tracked loop; what did not exist was a reader surface. Whether it becomes a standing weekly row is a call for the desk, not something to schedule automatically.

Sources

• paper{pelosi,gottheimer,khanna,mccaul,wasserman-schultz,crenshaw}-trades/LEDGER.json`
• congress — House PTR page captures
• congress-import — transaction-date assignment and price sourcing
• Prices and benchmark: * and data/stocks/SPY/ohlc/, 26 August 2026 close
• Related: the 26 August 2026 roll-up on where AI buildout money is actually flowing