AI-Infra Portfolio Monitor

RULES — follows the written plan to the letter. When a name falls into a staged buy zone AND is still above its 200-day trend, it buys one tranche. When a name breaks below that trend, it sells a third of what it holds.

ALL ARMED — buys the same buy zones but ignores the trend rule completely, and never sells. It exists to answer one question: is the trend filter helping, or just costing opportunities?

SCAN — ignores the buy list entirely. Every time a new name appears in the global bargain scan's top ten, it puts CHF 5,000 into it, up to 25 names, and never sells.

SCAN TM — an experiment, not a recommendation. It watches the same new scan names but waits for one to steady — ten sessions without setting a new closing low — before putting CHF 5,000 in, and it sells each holding about a year later whatever has happened since. When it runs out of cash it sells its worst holding to pay for the next one. The backtest behind these rules was not validated, which is exactly why it is being run live.

SCAN WIDE — an experiment too, and by design a far busier one. It drops the top-ten cap altogether: every name that newly falls 25% below its 52-week high gets CHF 1,000 at the next session's close, whether or not it is among the deepest ten. It sells each lot about a year later whatever has happened since, and only sells early when the book is full and a new name needs funding. That signal fires roughly 1,236 times a year, so this book runs on CHF 1,250,000 rather than CHF 100,000 — enough room to hold about 1,250 positions at once, which is the point: holding for the year, not churning. Like SCAN TM it is not validated; it is the live test.

The mirror — every franc a book spends is also spent, on the same day, on the Core World ETF in a shadow book. That is the only fair comparison: the same money, the same days, doing the boring thing instead.