A share price is two numbers multiplied: what the company earns, and what the market pays for each dollar of those earnings — the tag. So when nearly everything we cover was marked down this summer, every fall had only two possible suspects. This page names the suspect for every company we cover, and refreshes the verdict with every data pull. The finding so far is lopsided: almost everywhere, the earnings kept growing. The tag did the falling.
Prices to September 6, 2026 (giants: September 1, 2026) · Our standard data pull; tags vs each company’s own five-year usual
Parts one & two · The story — the markup, then the markdown
Did the business fall, or did the crowd?
Between June and July of 2026, most of the shelf we cover peaked together and pulled back together. A falling share price feels like one event, but it is always two numbers multiplied: price = the earnings × the tag (the price-earnings ratio — what the market pays for each dollar a company earns). When a stock drops 40%, either the company started earning less, or the crowd cut what it will pay per dollar earned, or both. Which suspect it was changes everything about what the markdown means — a business that broke is one story; a crowd that cooled on a business still growing is a very different one.
And before the markdown, remember, there was the markup — the same two suspects ran the prices UP. The three companies below show the whole range of how: one run built by the crowd, one built by both engines, and one built entirely by the business while the crowd hung back. Read these three and you can read any stock’s history.
First, the markup — three companies, three very different ways up
Price, earnings and the tag, before and at the peak. The earnings change times the tag change is exactly the price change — every time, for every stock.
The mirror image: earnings grew twenty-eight-fold while the tag actually FELL. The crowd never kept up with the business.
Three worked examples of one identity: price = earnings × the tag. Earnings shown per share (price ÷ the trailing tag, from the same pull); tags are trailing price-earnings ratios — in early 2023 several of these businesses were at a cycle bottom, which is why some starting tags look odd: a low price over even lower earnings. The map and table below run the same split for all 35 names — and for the road back down.
The map — who did the falling: the earnings, or the tag?
Since each stock’s own peak: across, what the tag (P/E) did; up and down, what the earnings did. The two multiply into the markdown, so a dot’s distance from the gold corner (“still at the peak”) is NOT the size of its fall — the faint dotted curves are: every position on one curve multiplies into the same price change. Hover a dot for the arithmetic; click it to open the report.
Because price = earnings × the tag, the two axes multiply exactly into each share’s markdown. A worked read: NVIDIA sits at tag -19%, earnings +21% since its peak — multiplied, its share price is just -2%, which is why it plots near the “price flat” curve: the earnings growth paid for the tag cut. KLA, at tag -38% with earnings +0%, wears the whole cut as price: -38%. Prices to September 6, 2026 (the four giants: September 1, 2026; Applied Materials: September 15, 2026). The run, the peak and the markdown are computed from daily closes since January 2023; “its own usual” is the data provider’s five-year mean price-earnings ratio. This page refreshes with each data pull — the dots move; the two questions do not.
Then came the markdown — and here the verdict is remarkably lopsided: at 27 of the 35 companies, earnings are higher today than they were at the stock’s peak, and at none of the 35 is the tag higher. The fall was the crowd, almost everywhere — the map above names the split per stock, and the table below adds the run up, so every name shows its price, earnings and tag through the whole round trip. Whether each newly-cut tag is now actually low is the third question — the table’s last column measures it against each company’s own five-year usual.
None of this is a forecast, and a cheap-looking dot is not a recommendation — some tags are low because the market doubts the earnings under them, and some markdowns deepen. The reading, the risks and the records live in each company’s report — every dot and every row below links to one. Nothing on this page is advice.
Part three · The walker — what leadership said
The forecasts were sober. The dog was not.
There is an old picture of a man walking a dog through a park: the man crosses the park in a straight, unhurried line, while the dog tears ahead, doubles back, and circles him the whole way. The man is the business; the dog is the stock. Everything above — the 8-fold markups, the 51% markdowns — is the dog. Here is the walker: what leadership said earnings were going to be, quarter by quarter and year by year, next to what was delivered. Where we hold the filed record, the score is 14 for 14 — every scoreable forecast was delivered at or above its guided midpoint, and 8 of the 14 cleared the entire guided range. The forecasts barely moved. The prices around them moved violently in both directions. That is the whole argument of this house in one table: read the walker, not the dog.
The walker’s record — guided earnings vs delivered, where we hold the filings
Each guide is scored against the delivery that answered it, from the company’s own 8-K filings. The dog column is the share price over the same stretch — and the last column is what that stretch paid a holder who watched the walker, not the dog.
3 of 3 at or above the guided midpoint; 3 above the entire guided range; none below it
The stock (since January 2023) walked in at $129, sprinted to $258 (+101%), now $223 (-14% off the sprint)
+73% held from January 2023, sprint and swoon included
Adjusted (non-GAAP) earnings per share as each company issues them, guide and delivery on the same basis; KLA’s figures are on the pre-split share count its guides used. Scored only where we hold the filed guide-and-answer pairs: KLA 8-K accessions 0000319201-25-000004, 0000319201-25-000009, 0000319201-25-000020, 0000319201-25-000031, 0000319201-26-000006, 0000319201-26-000014; Lam 0000707549-25-000051, 0000707549-25-000068, 0000707549-25-000082, 0000707549-26-000006, 0000707549-26-000020; Republic 0001060391-23-000005, 0001060391-24-000132, 0001060391-25-000081, 0001060391-26-000093. Two honest gaps: Comfort Systems publishes no numeric earnings guidance, and Applied Materials’ guidance appears in press releases we have not put through the same records process — neither is scored, in either direction. The full quarter-by-quarter version of each record — the walker chart — lives in every company report. “Following the guide paid” is simple share-price change over the stated window — a holder who stayed while the guides kept being met; the dog column’s percentages are that stretch’s two legs, start to sprint and sprint to now. Windows start at the calendar year each record begins (not the filing dates themselves), and dividends are not counted. It is arithmetic about the past, not a strategy, a forecast, or advice.
A clean sheet is not a promise — guides can be set low, and one day one will be missed. But it points at the discipline this page is built on: the number leadership writes down moves in feet; the price around it moves in miles. When the two disagree, the house reads the filing. The picture has a chapter of its own — “The Dog and the Walker,” chapter eleven of The Quiet Room, the free GARPify book — because it is the heart of how this house reads markets.
In each leg the earnings change times the tag change is exactly the price change (price = earnings × tag) — the run’s split is measured January 2023 to the peak, the markdown’s from the peak to now. Rows grouped by theme (theme names link to their pages where one is live), deepest markdown first. In the last column, green tags sit more than 15% under the company’s own five-year usual; red more than 15% over; gold in between. A green tag is a fact about the price, not a verdict about the business — the report is the verdict’s home. Computed from our standard data pulls of September 6, 2026 and September 1, 2026 (Applied Materials: September 15, 2026).
How it adds up — the whole shelf in two paragraphs
The run was both engines. The fall was the crowd alone. Between January 2023 and the peaks — most of them crowded into June and July of 2026 — the 35 companies on this page ran between +32% and +2,871%, with the earnings doing more of the lifting at 19 of the 35 and the tag at the other 16 — growth and markup, working together. Then came the markdown: between 2% and 51% off the peaks (the deepest: STRL, −51%) — and on the way down the split was one-sided: earnings are higher today than at the peak at 27 of the 35 companies, and the tag is higher at none of them. The prices that fell were, almost everywhere, prices of businesses still growing — and where we can score what leadership had promised (Part three), all 14 of the 14 scoreable forecasts were delivered at or above the guided midpoint.
What the shelf costs now, against its own history: 12 of the 35 tags still sit more than 15% ABOVE their own five-year usual (the furthest: LRCX, +102%), 15 sit more than 15% below it (the furthest: AMZN, -69%), and 8 sit within 15% either way. Those are facts about prices, not verdicts about businesses — a low tag can mean a bargain or a doubt, and telling the two apart is what the company reports are for. The dots move with every pull; the identity never does. Nothing on this page is advice.
The Pullback Map is a GARPify standing page — it refreshes with each data pull, and every name links to its full company report. The run, peak and markdown are computed from daily closes since January 2023; “its usual” is the data provider’s five-year mean price-earnings ratio. GARPify publishes research, not recommendations. Nothing on this page is advice.