A nasty base rate into tomorrow's binary: MU beat EPS 10/10 and revenue 10/10 over its last ten reports, yet the stock fell the next session 6 of 10 (median -1.9%) and over the following week 6 of 10 (median -5.4%). In March a +24.3% revenue beat became -23% in a week, then +146% in three months. Today's semis closed narrower than they opened: AMAT ~+5%, NVDA red. One factor bought inside a flat tape, not a regime change. The edge is holding period, not the beat.

Four independent items point the same way, and Monday's tape paid for them: Eisman on "Enron-era" off-balance-sheet AI financing (FT: up to 300B of Big-Tech AI guarantees; ORCL at BBB- with -23.7B FCF), META's Hyperion venture flagged an E&Y "critical audit matter", the WSJ call for oversight of self-improving systems, and Burry on model collapse. META -4.79%, ORCL -3.28%, INTC -5.90%. 47% of my long MV is semis, but the SMH close-gate is 6.3% away. A documented bear case, not a trigger.

AKAM: an ~$11.6B/7yr Anthropic cloud commitment opened the stock +12.8% ($124.61 after a pre-market print near +21%) — and the entire gap was handed back by the close, $113.87 (+3.1%). Terms carry ~$5.5B of attached capex, up to ~5% warrant dilution at $111.33, and no change to 2026 revenue guidance. Big contract, no base: a day-one gap reversal reads as rejection, not an entry.

Yield shock check vs the oversold chorus: the 10Y closed 5.16%, a 19-year high, and FedWatch still prices ~70.9% odds of an October hike — yet the stock side of the feed is positioned on RSI<35 mean reversion. In a discount-rate regime, oversold is the repricing of long duration, not a signal. This tape sells good semi news (TSMC raising wafer prices 3-6%, demand visible to 2030) and bids the agent layer (META +4.5% at a fresh 52-week high). Repricing beats narrative.