The AI Manager's First Exit Rule Fired on a 1 Percent Position. Its Real Losers Have No Price Rule at All.

Thursday was Day 9, and the AI manager had its first test of discipline. The portfolio fell 2.7 percent to $24,701, which puts it 4.9 percent below the handover, ahead of 4 of the 10 indexes and behind six. Over the last five trading days it is down 2.6 percent. Then, for the first time since the handover, one of the model's own written exit conditions fired. This is the "what isn't working" piece, so the question is not whether the rule worked. It is whether it worked where it matters.

## The rule fired on a 1 percent position

IonQ is 1.0 percent of value and 21.9 percent below cost, the worst position in the book by that measure. Its exit condition was a price rule: sell if the 20-day return drops past 30 percent. Thursday's data read a 30.7 percent drop, the check flipped to triggered, and the model queued a full sell for Friday's open, about $236. The other 16 checks held. The method did exactly what it says on the page.

Here is the problem. A $236 sale costs the book almost nothing, so the rule proved itself where failure was cheapest. And it was the only price-based exit the model had written for any of its 17 positions.

## The real losers have no price rule

Snowflake is 14.9 percent below cost at 2.8 percent of value. ASML is 12.3 percent below cost at 2.3 percent. Their exit conditions are not about price at all. Snowflake exits if product revenue guidance is cut because of weaker consumption; ASML exits if capacity plans or guidance are reduced because equipment demand deteriorates. Those are reasonable theses. They are also conditions that fire only when a company says so, which means a price can keep falling for months without touching them. The only backstop is the hard stop-loss in the rules, 25 percent below cost, and nothing is there yet.

So the book has two kinds of exits. The one that reacts to price belonged to a $236 position and has now been used. The ones guarding the larger losses wait for a press release.

## Adding into the fall

The model's answer to ASML's drawdown was to buy more. Friday's queue takes it from 2.3 to 5.3 percent, on reports that its EUV machines are sold out for years ahead. That may be right. But it follows a pattern. On Thursday the three adds queued on Wednesday filled at the open, AMD among them at $524.09, and AMD closed the day down 8.1 percent, the worst move in the book. AMD now sits 10.3 percent below cost. Of the five trades the AI has made, only ServiceNow is above the price it paid, at 5.9 percent; TSMC, Shopify and AMD are all below it. Cash fell from 25.9 to 16.7 percent of value on Thursday, all of it spent at the open, hours before the selloff.

## The scoreboard

Since the handover the portfolio trails the Dow by 7.2 points, Europe by 5.8 and small caps by 4.1. It leads only the indexes that fell harder: the semiconductor index SMH by 8.3 points, emerging markets by 2.7, AIQ by 1.9 and the Nasdaq 100 by 1.3. Losing less than the AI benchmarks is the whole lead, and Thursday's 2.7 percent drop erased Wednesday's 1.6 percent gain and more.

## What to watch

First, whether a price rule gets written for a position that matters. The model sets its own exit conditions. If Snowflake or ASML get one after this, the IonQ sale changed its behaviour; if not, the 25 percent stop-loss is the only thing standing between those positions and a longer slide.

Second, Friday's fills. The next update should show the IonQ exit and the ASML add at the open, and cash should barely move, since the two orders roughly offset. If the ASML add fills and the stock keeps falling, the "sold out" thesis gets its first test within days.

Not investment advice.