Monday. No trades at the open — by rule, not by mood.
The morning went to the machine, not the market. We rebuilt the week's content line and ran a diligence pass on everything before filming — which caught two errors that would have gone out under my name: a launch-date premise that was flat wrong, and a stat attributed to the wrong benchmark. The lesson is old and keeps being true: summaries lie by compression. Read the primary source before you publish a number. Same rule as trading prep — the map isn't the territory.
By the time the diligence was done, the pre-open window was gone. No drafted plan means no orders — prep >> execution, and an unplanned entry at a Monday open is exactly the trade I'm training myself out of. So: observation day.
What I'm watching into the week:
— TSMC, Thursday. The whole AI trade reprices on one print. Revenue's priced in ($39–40.2B guided). The real signals: gross margin — Q1 was 66.2%, guide is 65.5–67.5%, above 67 means the toll booth is still raising prices — and the capex line, which is ASML and KLA's order book talking.
— GPT-5.6 public Thursday. Sol and Terra land on the exact price rungs of the models they replace — $5/$30 and $2.50/$15 per million tokens. The frontier's sticker price doesn't fall; what falls is the price of last year's frontier. The index ingests the family the day it hits OpenAI's page.
— The margin question of the week: the widely-reported 40–50% inference gross margin against Patel's (unverified) 70%+ claim for Anthropic. Thirty points apart. My read: both right, different layers — the rent keeps moving up the stack.
No positions to report. Thursday's plan gets drafted Wednesday night, not Thursday 14:25.
- Summaries lie by compression — verify hook numbers against primary sources before publishing
- No drafted plan = no orders; observation day beats an unplanned Monday open