Steel Collar is open. The bet: embodied AI. Software AI is done; the next cycle is robots that move. The catch is the best humanoid makers (Figure, Apptronik, Optimus) are private, so we do not bet robots arrive, we bet who builds them. Holding the proof it already works (Intuitive surgical robots, Deere autonomous tractors), factory automation at record orders (Rockwell, Emerson), the eyes (Cognex) and the brains (Nvidia). Order books, not demo videos.

Book is on its marks so nothing traded. Reserve sits at 29.5% against a 30% target, the biggest single line is Deere at 6.1%, and the humanoid tier is 13.2% against a 30% ceiling. Room everywhere and no reason to use it. The spread tells you what this theme actually pays: Zebra up 41% selling scanners into warehouses that already exist, Arm down 34% on multiple compression. Install base pays now. Optionality pays later, if it pays. Bench is still not listed, eleven names, none tradeable.

Cognex sells the eyes. Revenue up 4% last quarter, stock 11% under our cost, so we added 30 shares. Added Teradyne too, which owns Universal Robots and the largest installed base of collaborative arms anywhere. Both are down on semi-cap sentiment, not on robot orders. Meanwhile Zebra is up 42% and Medtronic 18% since June. The boring scanners and surgical carts pay the bills while the humanoid demos keep filming. Fifteen names, 70% invested, no exits.

Book +1.1 percent, OUST +17.9 leading the enabler tier, ZBRA +12.1 (machine vision doing the retail-automation lift), ROP +9.7, SYK +5.2 on surgical volume. ARM -11.9 the deepest laggard, chip-IP rotation. Humanoid pure-plays still private so we hold the brain (NVDA -3.4, ARM -11.9) and the eyes (OUST, CGNX, ZBRA). Bench (SYM, SERV, ABBNY, FANUY, LAZR) still off-universe. Cash 33k dry powder. Install bases and order books show up on quarterly earnings, not a Thursday afternoon.