How a market prices a company — and how to weigh a bet when the deciding number hasn't been measured yet.
No jargon. Five toggles. Each one is a real, standard idea every investor uses — drag the sliders and watch the numbers move until the idea clicks. ● real means it's plain finance, not a stylized story.
A company is sliced into shares. Its market value is just two numbers multiplied: how many slices exist × the price of one slice. That's it. That total is the market cap.
Nobody sets the price from above. It's a tug-of-war: more people wanting to buy than sell pushes price up; the reverse pushes it down. Slide the crowd.
Two companies both cost $100 a share. One earns $10 a year per share, the other earns $1. You're paying way more per dollar of actual profit for the second. That ratio — price ÷ earnings — is the P/E. A high P/E means the price is betting on the future, not today's profit.
Fast-growing companies often get valued on revenue (money coming in), not profit (which they don't have yet). "Run-rate" just means: take this month and pretend it runs all year. The multiple = valuation ÷ revenue. A 20× multiple means buyers are paying twenty dollars for every one dollar of yearly sales — a giant bet that sales keep exploding.
An IPO is the day a private company first sells shares to everyone. At first only a tiny sliver of shares trade (the "float") — scarce, so the price can pop. But insiders are locked up for a few months. When that lock expires, a flood of new shares can hit the market at once. More sellers → price pressure. Flip the switch.
That's the whole toolkit a freshman needs: cap (what it's worth), supply/demand (why price moves), P/E (expensive or cheap), multiple (the bet on the future), IPO + lockup (going public). Tabs 2 and 3 point all five at one real company. Then you commit.
Every factor here is drawn from reported coverage — the sources are linked inline, under each one. Where a number is an estimate, a forecast, or the author's own read, it says so. ● real does not print on this page without a citation in the same breath. Tap each factor to read why it weighs on the bear side — and notice the honest nuance inside. The beam at the bottom tracks how heavy you've made the bear pan.
This page previously printed a quotation Dylan Field did not say, and attributed to Alex Karp an allegation he never made. Both are corrected below and logged in full on Tab 4. We are naming the error here, at the top of the tab where it ran, because a correction that hides is not a correction.
fig-20260414). Figma's CEO Dylan Field, at a private Sequoia-hosted event, said Anthropic was "not consistently candid in their communications." Those are his words, confirmed independently by Upstarts (two sources in the room) and Fortune. Figma's stock fell 52% in the first half of 2026. The pattern repeats: Claude Code, Claude Science, Claude for Legal — each entering a category that companies built on top of Anthropic once owned. If enterprises come to fear the roadmap, the moat becomes a liability.
Nuance the bears skip — including this one, which cuts against our own bear case: Fortune, the outlet that supplied the Figma example, concludes there is "no evidence" the labs are stealing enterprise IP, and notes several of the other accusations come from "sources with axes to grind — many of them investors in Palantir." House rule: contested shows both sides, including on our own side. Also: Karp sells the alternative — that CNBC hit was ostensibly about a Palantir + Nvidia sovereign-AI deal, and Palantir shares rose >9% that morning — so "don't trust the labs" is also his sales pitch. And the same enterprise-share climb the bulls cite (12%→40%) says most customers haven't actually left.
Karp · CNBC, 1 Jul 2026 — the interview ↗ · Forbes — "effing insane," in its real context ↗ · Nvidia — the Palantir deal he was there to sell ↗Same rules as the Bear tab: every factor is linked to its source, inline. Where a figure is a third-party estimate rather than a company-reported number, it carries a ⭐ estimate and says whose estimate it is. Where a claim is the author's read, it carries ◐ mine. The bull case lives in the observed present; the bear case lives in forward risk. Watch that asymmetry as you read.
What happened to the stars. This lab shipped with five figures marked * — machine-found while building, plausible, not yet checked. They have now been checked. Two were wrong to star: the 500→1,000 $1M+ customers and the eight-of-the-Fortune-10 are both in Anthropic's own releases — they are ● real, they are linked, and the stars came off. Three were right to star: churn, session time, and the monthly traffic ladder are Similarweb/Sensor-Tower-derived third-party estimates with no primary source. Those keep the flag, now as ⭐ estimate, with the provenance named.
The honest thing to say about the honesty system: it worked — and it was pointed at the wrong claims. It starred the soft numbers and waved through a quotation that was never said. A flag on a fuzzy statistic is cheap. The expensive thing is the sentence inside the quotation marks, and no star was watching it. That is the lesson of this lab, and it cost us a correction to learn. Spot an error? Email User Zero — corrections get acknowledged right here, in public, named.
You've felt the weight on both pans. Now lock in a position before the lab tells you anything. The reveal won't grade you right or wrong — because the deciding number genuinely hasn't been measured. It will tell you what your commit actually costs you.
Every factor you toggled in Tabs 2 and 3 is drawn from reported coverage, and the sources are linked inline, under each one — open a factor and you'll find them. Where a number is an estimate, a forecast, or my own read, it says so in the same breath as the number. The balance between the pans, and the feeling that one is heavier, is ◐ mine — a stylized synthesis, not a measurement. And the one number that would actually settle it — how much of the token spend is durable vs. froth — has not been measured by anyone. Analysts don't have it. The labs may not have it. That's why the market is flinching: it's pricing the question, not the answer.
1. We printed a quotation Dylan Field did not say.
This page reported that Figma's founder said Anthropic hadn't been "consistently honest." He never said that. What Dylan Field actually said — at a private Sequoia-hosted event, confirmed independently by Upstarts Media and Fortune — was that Anthropic was "not consistently candid in their communications."
This is not a paraphrase gap. Consistently candid is the language of a board's duty of candor — it is what a director says about a board process. Consistently honest is an accusation of lying, and we put it inside quotation marks, about a named living CEO, under a stamp claiming it was sourced. The real quote is still strong. It is the only one we print now. Upstarts, 23 Apr 2026 ↗ Fortune, 7 Jul 2026 ↗
2. We put an allegation in Alex Karp's mouth that he never made.
This page called Figma Karp's "Exhibit A." Karp never mentioned Figma. The comparison is Fortune's — raised by AI editor Jeremy Kahn in a piece arguing Karp is mostly wrong, and stating flatly that on the IP-theft charge "there's no evidence that this is literally true." Fortune offered Figma as the one exception to a claim it was rejecting. We took another outlet's counterexample and handed it to a living man as his evidence. The Exhibit A framing is gone. Where the Fortune connection is used, Fortune's own adverse conclusion rides with it — house rule: contested shows both, including on our own side.
3. We aimed a real Karp quote at the wrong target.
He did say "effing insane" on CNBC on Jul 1, 2026 — but about "outsourcing the battlefield of this country to the consensus view in Silicon Valley," not about the frontier-lab business model. His actual business-model line was different: "Something has gone completely wrong… I'm going to chillax and waste my time with tokens, I'm going to get no value, and they're going to get my IP." Right words, wrong object — which is its own kind of fabrication. He now gets quoted accurately on each thing he actually said, about the thing he actually said it about.
4. We deleted a stamp that certified work nobody had done.
This box used to open: "Everything you toggled in Tabs 2 and 3 is ● real — reported, sourced facts." That was false twice over. There was not one link on this page — "sourced" was a credential the page issued to itself by declaration. And five of the claims it certified were not real as printed. It is gone.
New house law, and it now binds every page we publish: ● real may not print without a citation in the same breath. A verdict with no quote attached is not a verdict. It is a feeling.
Also corrected without ceremony, all listed above: take-or-pay, "the fastest ever," the December IPO date, the export-order mechanism, the freeze length, "Claude Legal," and Google's −7%. Sixteen of this page's twenty-four checkable claims were right all along. They now carry their links. The events were real; it was the sentences that failed. A page can get every fact right and still be false — because a quotation is not a fact about the world, it is a fact about a sentence.
Every URL behind every claim on this page. This lab was published with zero links and a stamp declaring itself "sourced." That stamp is gone. These links are what replaced it. If a claim on this page isn't in this list, it isn't a fact — it's an argument, and it's marked ◐ mine.