What each strategy actually did, and whether it holds real money. Only the operator's live families trade, and inside them only a positive record: shrunk R above zero over its life (a broken last 20 trades puts it on watch, not on the bench). Everything else is in shadow — it still scans and journals, places no orders, and every decision is settled against real bars (less 15 bps friction a side) in the shadow ledger, which is the only way back: 10+ settled trades at +0.10R in most weeks inside a live family, 20+ across 4+ weeks outside one, then probe size. Below 4 R-scored trades a setup is unproven and learns in shadow rather than with money. $R is realized P&L over the dollar risk that produced it — the R the account actually felt — and a setup is judged on whichever of the two Rs is worse. Size is capped at 1.00 and scaled by the governor below it, which reads the drawdown of what is live.

Model estimate — nothing here is traded. Prices a slightly-ITM call replacing each long swing trade (Black-Scholes, IV = the underlying's realized vol at entry), sized so the premium ≈ that trade's stop risk. No live IV surface, bid/ask spread, or early exercise — real options would do worse than this. A positive verdict justifies buying real options data to confirm, never trading.

The bot will not open these, and flattens them if held. Applies on the next cycle — no redeploy, no login. Editing works on the VM's own HTTPS origin (not the storage.googleapis.com mirror). Also available from Cloud Shell: ./scripts/vm.sh restrict --add TSLA