Setup: Attention-spike fade

The idea and the mechanism

Retail attention-buying is price-insensitive: people buy the name they just heard about, not the one that's cheap. That crowding tends to reverse over a short horizon once attention moves on. Better still for us, the attention inflates the stock's option prices — so the trade becomes selling that inflated premium, which rhymes with what the book already does well. The strongest cases live in messy, small names the big players can't size into.

Thematically close to the premium-selling book but reached through a totally different door (crowd behaviour, not capitulation). Highest data cost on the list because it carries two blockers at once.

The frozen gate

Will be considered to have passed Stage 1 if:

• Mean per-trade EV ≥ +0.50% of notional [suggested, to freeze]
• Welch p<0.05 AND mean>0 [suggested, to freeze]
• Win rate ≥ 60% [suggested, to freeze]
• Sample size ≥ 300 attention-spike events [suggested, to freeze]
• Effect present in both sample halves [suggested, to freeze]
• Worst single trade ≤ −5× mean win (tail check — selling premium on attention names is exactly where short-vol blows up when the crowd is right) [suggested, to freeze]
• Premium-conditional check: events where the option premium is materially elevated (IV percentile ≥ 70 of the name's own history) [suggested, to freeze] must outperform events at median IV — confirms the "inflated premium" half of the rationale is doing real work

What we expect to find

Effect probably exists in some form but is the noisiest setup on the list because (a) the attention-measure is easy to game, (b) the crowd is sometimes right and the move continues, and (c) the option- premium signal is often already faded by the time the attention is visible at the daily level. Probability of clearing the gate is low-moderate (~25-35%). Most likely failure modes: tail-check fires on a single meme-stock blowup, or premium-conditional check shows no incremental edge over a simpler vol-rank filter.