The AI Manager Is Beating 5 of 10 Indexes After Three Days. It Has Not Made a Single Trade.
Day 3 of the index race: the AI-run paper portfolio is down 2.3 percent since inception and still ahead of QQQ, IWM, EEM, SMH and AIQ, because the chip index fell 6.4 percent and this book is not all chips. Here is what is actually working (two positions whose companies shipped products this week), what is not (a quantum stock down 21.2 percent against cost), and why the model's Day 3 decision was made on a single news article.
Three trading days into the index race, the AI manager has made zero trades and is beating 5 of the 10 indexes it is measured against. Both facts are on the scoreboard. Only one of them is the result of anything it did.
## The scoreboard on Day 3
As of October 14 the portfolio is worth $25,368, down 2.3 percent since inception and down 2.0 percent on the day. Against the panel it is ahead of SMH (down 6.4 percent, a spread of 4.1 points in the portfolio's favour), AIQ (down 4.0 percent, spread 1.6), EEM (down 3.6 percent, spread 1.3), QQQ (down 3.0 percent, spread 0.6) and IWM (down 2.7 percent, spread 0.4). It trails SPY (down 1.1 percent, 1.2 points against), ACWI (down 1.3 percent, 1.0 against), URTH (down 0.9 percent, 1.4 against), and DIA and VGK (each down 0.1 percent, 2.3 against).
The pattern is plain. The portfolio beats every index that is more concentrated in AI and chips than it is, and trails every broad one. On three days when semiconductors sold off, a book of AI stocks that is not all chips, sitting on the cash the human left behind, lost less than the chip index. That is the whole explanation for "5 of 10". The model chose neither the cash, nor the positions, nor the timing.
## What is working: the two leaders shipped
Only three of the seventeen positions are above their cost. Meta is up 2.6 percent against cost at 2.8 percent of value, Apple is up 1.9 percent at 1.1 percent of value and Shopify is up 1.0 percent at 0.7 percent of value.
The two leaders shipped product this week. Meta launched its Muse Wonder models on October 12, frontier-class systems built for long-horizon personal and professional agents. Apple introduced HomeView on October 13, a 7-inch smart display built around its new Siri AI assistant, with preorders open immediately and no EU sale because Siri AI is not available there yet.
Neither story reached the model. Both records are in our news database, and neither is among the articles any of the three daily decision runs received. The Apple record carries an uncertain publication date, which excludes it by design. The Meta record does not.
## What is not working: the smallest position is the biggest loser
IonQ is down 21.2 percent against cost at 0.9 percent of value. Snowflake is down 9.2 percent at 2.9 percent of value and Datadog is down 8.2 percent at 1.5 percent of value. The stop-loss sits at 25 percent against average cost under the new rules, so IonQ is still inside the limit.
None of the three losers is visible to the model either. Of the 82 names in universe v2, 24 are mapped to a company record in our news database. The other 58, including six current holdings (ServiceNow, Datadog, Shopify, Monday.com, IonQ and Qualcomm), have no mapping, so no article about them can reach the decision prompt.
## The decision that stopped being forced
The rules changed on October 14, v1 to v2: the maximum position weight went from 15 to 25 percent, the minimum order from $1,000 to $500, and a trim of an over-cap position back to the cap is now exempt from the minimum. On Days 1 and 2 the model's only proposal was to trim Palantir, the largest position, to the old 15 percent cap, and the rules rejected it both times: an order of $627 against a $1,000 minimum on Day 1, $617 on Day 2. Under v2 that position is inside the cap, and on Day 3 the model proposed nothing.
It also had almost nothing to go on. The Day 1 run received seven articles, Day 2 four, Day 3 one: Microsoft's Xbox Labs announcement, a memo about AI-assisted game ports. A manager asked to beat ten indexes and shown one article about retro games will, correctly, hold.
## What to watch
First, the SMH spread. The lead over the chip index is the largest spread on the panel and the one most likely to reverse: it exists because this book holds less of what fell. If semiconductors bounce, "5 of 10" becomes a smaller number before the model does anything. The count is on the homepage every trading day.
Second, the first trade. With the Palantir trim no longer forced and the universe widened to 82 names, the next proposal will be the first the model makes for a reason of its own. The trade log shows the reasoning and the articles cited. If the cited list keeps coming up short, the fix is in the news mapping, not in the model.
Hypothetical paper portfolio. Not investment advice. No real money is traded; decisions are made by an automated AI system.