TL;DR
Our March allocation overweighted Semiconductors (SOXX) at 35% going into Q1 earnings season. The call returned -2.1% while SPY returned +1.4%. We were wrong, and here's why.
What we believed
Q1 earnings would confirm the AI-driven semiconductor super-cycle. Our 5-round debate consensus weighted memory and GPU exposure heavily, expecting NVDA and AMD to drag the sector higher.
What actually happened
NVDA reported in-line and guided conservatively. Memory pricing softened earlier than our model predicted. Semis underperformed cyclically by 350bp in 3 weeks.
The hidden assumption
We assumed Q1 guidance would mirror Q4's bullishness. Round 4 (Devil's Advocate) raised this objection: "What if guidance normalizes?" — and we dismissed it as low-probability. In hindsight, the objection deserved 30-40% probability weight, not 5%.
What we changed
For April debates, Devil's Advocate now has veto power on any single-thesis sector exposure exceeding 30%. This rule has held for 3 months.
Why we publish this
Because "the AI got it wrong" is a feature, not a bug.
Not investment advice.