Model pulse, week of 21 August 2026

This is a snapshot, not a buy list

This is the first dated Pulse on this blog. It describes what the published Zenith ruleset is built to surface on 21 August 2026. It is not a recommendation, not a performance report, and not a claim that we rescreened 50,000 names overnight.

Use it the way a desk uses a morning note: as a map of our model, then go read the live pages.

• Public screener
• Sector heatmap
• How we score a stock

---

Market backdrop (the week the tape is in)

Major US averages spent this week under pressure. Yields have been sitting near longer-term highs. Oil has been relatively calm. Commentators have been circling the next Nvidia print and whether AI-infrastructure leadership still carries the tape after a stretch of records earlier in August.

None of that is a Zenith forecast. It is context. The model does not ingest the 10-year or Brent as a live factor. If the tape is rotating, you will see it first in prices and in the heatmap, not in a secret macro score.

---

What the model is built to overweight

On the public universe, three overlays do most of the work:

1. AI-infrastructure beneficiaries. Names tagged AI Beneficiary (NVIDIA, TSMC, Broadcom, plus a short list of cloud/platform names) get a ranking boost on the Fund+TA shortlist and a +1 notch on the public composite when fundamentals are also strong. That is a published overlay, not a live news classifier.

2. The Microsoft Hold cap. Microsoft cannot print Buy or Strong Buy. High disruption risk on traditional software seat economics is a hard rule. If you want the long version, read Why the model will not Buy Microsoft.

3. Quality compounders that are not an AI trade. Visa, Costco, Eli Lilly sit in the liquid set with Low or mixed disruption tags. They will not look like GPU proxies on the heatmap. That is the point of a sector view.

---

What I would actually open this week

If you only have ten minutes:

1. Open the heatmap and look at Semiconductors vs Software vs Energy. The colors are model signal, not day P&L. Day % on that page is illustrative unless you click through to a live quote.
2. Open NVDA and MSFT on public analyze, side by side. Same three-leg composite. Different disruption tags. Different caps.
3. If you care about ranking rather than a single ticker, the homepage Fund+TA module shows two names live and locks the rest for Premium. The rank math is in this week's shortlist note.

Do not treat a Strong Buy tile as a trade. Treat it as "the ruleset currently leans this way — now argue with it."

---

What this pulse will never do

• Invent a weekly accuracy number.
• Quote a price we did not fetch in this article.
• Pretend Alpha Engine is a fourth secret model. It is the same published legs plus 30-day headline tone and sliders. Premium.
• Tell you to buy the dip in Mag 7.

Next pulse: same Friday slot. Same length. Same rule — if we cannot say it on the public pages, we will not say it here.

---

Educational only. Not financial advice. Not a recommendation to buy or sell any security. You can lose money.