keepwhatworks
A chaotic field of golden and blue sparks streaming from the left, filtered through a series of luminous ring gates, converging into a single clean beam of light toward a bright horizon — infinite opinion, one calibrated verdict kept.
Essay · Jul 7, 2026 · Judge with veracity, not ferocity.

Everyone's a Critic

Creation went to zero. Criticism went to zero right behind it. The last asset standing isn't the critic. It's the proof of whose judgment still tracks the truth. And for the first time, that proof is measurable.

There's a comforting story going around. It's half right, which is the most dangerous kind of story. It goes: creation just got cheap, so the value moves to the critic. Anyone can generate a thousand images, a hundred landing pages, a full app from a sentence. So the scarce skill now is taste, judgment, the ability to say that one, not that one. Critics are cheaper than creators. Everyone's a critic. And now, at last, it's all measurable.

The first half is true. It's the thesis of this entire site. The second half, "everyone's a critic," is where the story quietly walks off a cliff.

The cheap seat

Start with why criticism was always cheaper. The asymmetry is older than AI and deeper than fashion. It is easier to check a proof than to find one, easier to taste a dish than to cook it, easier to know a sentence is wrong than to write the right one. For a vast class of problems this isn't a vibe. It's the P-versus-NP intuition wearing an apron: solutions are expensive to produce and cheap to verify. Machine learning keeps rediscovering the same slope. Discriminators converge before generators. Verifiers outperform the models they check. The entire modern alignment stack works by pointing cheap preferences at expensive generation. It works because humans can reliably say which of two answers is better long before any system can reliably produce the better answer on demand. Criticism has always been the cheap seat in the house.

What changed in twenty-four months is only that the other seat got cheap too. Generation fell toward zero. And by the plain logic of scarcity, when the thing that used to cost everything goes free, value should flee to whatever's still scarce. Which, the story concludes, is the critic. Taste is the moat. The verifier is the asset. Fine. That much I'll sign.

The word that breaks it

Now watch the word doing all the damage: everyone. The slogan assumes every critic's verdict is worth the same. It never was. "Everyone's a critic" was always an insult, precisely because cheap criticism is worthless. When creation was scarce, the worthlessness didn't hurt. There were few creators, so even a mediocre critic had real signal to point at. Scarcity on one side covered for the noise on the other.

Kill that scarcity and the cover is gone. Zero-cost generation means infinite creation. Infinite creation means infinite reaction. And infinite reaction means the price of an opinion collapses to zero right alongside the price of a creation. This is the part the story refuses to say out loud: everyone being a critic does not make critics valuable. It makes them worthless in the exact same way creators just became worthless. The floor fell out from under both.

The commoditization of creation doesn't crown the critic. It commoditizes the critic next. Same mechanism, one beat later.

So the sentence eats its own tail. If cheap creation devalues creation, then cheap criticism devalues criticism. And "everyone's a critic, now measurable" describes a world drowning in free opinions, not a world where judgment is scarce.

Volume is not value

The slogan makes the oldest mistake in the book, the same one that sinks every KPI dashboard. It confuses volume with value, the proxy with the truth. A click is a proxy. A rating is a proxy. An opinion is a proxy. The truth they're standing in for is judgment that actually predicts what's good. When everyone can rate, the rating stops meaning anything. That's not a bug in the star system. That's Goodhart collecting his rent. A billion reviews is not a billion units of taste. It's noise with a five-point scale bolted on.

The scarce verdict

Here is the thing that survives the collapse, and it is not criticism. It's calibrated criticism. The specific critic whose verdict is coupled to reality. Everyone can have an opinion about the movie. Almost no one can tell you, in advance, which film returns its budget. The Amazon review is free. The review that forecasts your five-year retention is priceless. Your taste in prose is free, and worth roughly what everyone else's is. Your taste in which parcel actually appreciates is the entire edge. It's a verdict that keeps being right when money is on the line.

Calibration is just coupling, pointed at a person: a judgment whose "yes" reliably moves with the truth it names. That is the one thing infinite free opinion cannot manufacture. You can only earn it by being right, repeatedly, in a way the world confirms and you can't fake.

The palate test

Is a person's judgment even the kind of thing a machine can carry? We ran the experiment. Look honestly at where taste data comes from: corrections. The moment you rewrote the answer, rejected the framing, reached for one of two options. Every atom has the same shape: here is what was offered, here is what I kept. That means taste data is selection data, by construction, all the way down. Nobody's history contains the ghost of the answer they would have written from scratch. It contains the verdicts they rendered on answers that existed.

So we measured both directions. A taste model fit on one person's corrections, shown held-out choices it had never seen (the thing they kept versus the thing they rejected), picked the owner's side nine times out of ten (p = 0.011 against chance). Then we injected the same validated signal into generation to make the machine write like its owner. The output moved sixteen out of thirty, indistinguishable from chance. The sample size was committed before the first result. The same judgment that identifies its owner's choices at ninety percent cannot be heard in what the machine generates.

That split is the oldest asymmetry confirming itself one more time: judging is cheaper than making, all the way down to the data. A palate is a discriminator. You can taste that the soup needs salt without being able to cook it. Now that's not a proverb. It's a measurement. The product implication follows immediately: the thing worth building from a person's taste is not a ventriloquist that fakes their voice. It's an oracle that renders their verdicts. It chooses as they'd choose, admits when it can't tell, and proves its accuracy on their own live decisions.

Now measurable

Which is where the story's smartest word turns out to be its best one: measurable. But it means the opposite of what the story thinks. Measurement doesn't make everyone's criticism valuable. It makes it possible, for the first time, to sort. To separate the couplings that hold from the couplings that rotted, and to find the rare critic whose verdicts track the truth. Measurement is not the coronation of the crowd. It's the guillotine for cheap opinion and the crown for the proven one.

The tell is what kind of number counts. A public leaderboard everyone can see is worth almost nothing. It's an opinion with a spreadsheet. The number that matters is the one only you can produce: a per-axis score on your own hardest questions. "This model gave me what I actually wanted 84% of the time." That's a benchmark nobody else can publish, because no single vendor ever sees your judgment across all of them. Measurability doesn't flatten judgment into a crowd average. Used correctly, it does the reverse. It lets one calibrated verdict be told apart from ten thousand cheap ones, with data instead of deference.

Be the verifier

So scarcity didn't vanish when the machines learned to generate. It relocated, one level up. From having an opinion to having an opinion that's provably right. It's the identical move we keep watching. When cognition commoditizes, value flees to atoms and taste. When creation and criticism both commoditize, value flees to calibration. Same physics, higher floor.

Which reframes the whole engine. The thing worth building was never a creation machine. Infinite free supply. It was never a criticism machine either. Infinite free supply, arriving one beat later. It's a calibration machine. The instrument that captures the only signal left scarce: whose judgment tracks reality, on what, and how reliably. Judge with veracity, not ferocity. The loud verdict is free. The accurate one is the entire asset. And it comes down to a single, unfair, and completely earnable line:

Everyone's a critic. Almost no one's a verifier. Be the verifier.

Intellectual lineage

  • P versus NP — the formal root of the cheap seat: verifying a solution can be radically cheaper than producing one.
  • David Hume — Of the Standard of Taste (1757) — taste as trainable discrimination: the delicacy that tells the difference is earned through comparisons, not composition.
  • Goodhart's Law — when everyone can rate, the rating stops meaning anything. Volume-as-value is Goodhart collecting rent.
  • Philip Tetlock — Superforecasting — the empirical proof that calibrated judges exist, are rare, and are findable only by scoring verdicts against outcomes.

Appendix — the science of scoring judges

→ Trinity Local

A calibration engine, not a criticism engine

Trinity Local doesn't add to the pile of creations or the pile of opinions — both are free and infinite now. It measures the one thing that stays scarce: whether a given judgment is coupled to the truth. Your lens is a measured critic, built from the verdicts you already rendered in your own transcripts — the palate validated at nine-in-ten on held-out choices, used exactly where the measurements say it works, with a prospective registry scoring it on every new choice you make. And when a new model lands, eval-run scores it against your rejection corpus, not a public leaderboard: calibration turned into a number only you can produce. Everyone gets to react. Trinity keeps score of who was right.

See Trinity Local →


Part of an ongoing series on durable systems.