Certified Wild

01 The Spread

Yesterday’s essay closed a thirty-year ledger on engineered surprise: a creator-claimed 44 percent click-through in 1994, a fraction of a percent today, and in between a population’s denominator repricing an entire format at once. The law behind the curve is the equation’s oldest. No entity can touch the denominator; the only road to E runs through the record; and so every use of the key files a copy of the key inside the lock. Delivery is the only thing that trains the gate, which means the engineered arrival is consumed by its own delivery schedule.

But that ledger had a second column, and I left it mostly unread. Found surprise — the arrival nothing optimized — depreciates too, because everything that arrives is deposited. It depreciates at the pace of a life rather than the pace of a campaign, and it never clusters, because the world is not aiming at you. Two arrivals, identical in every property an instrument can measure, can therefore carry different depreciation rates — and the difference lives in exactly one variable. Not authorship. Aim.

Wherever a real price difference persists, a market assembles to capture it. I read the “No AI Allowed” signs I wrote about last month as the first crude price signal: venues advertising, in effect, that their surprises are grass-fed. The signal will get less crude, because the spread it points at widens every year the engineered class depreciates. The scarce asset of the attention economy’s next decade is not attention. It is wildness — and wildness has a certification problem.

02 The Certificate Fails at the Artifact

The obvious response to a wildness premium is a wildness label: watermarks, disclosure regimes, content credentials, an organic sticker for information. Every one of these attaches to the artifact, and at the artifact the certification is unwinnable, for three compounding reasons.

First, a certificate is itself an artifact — and the counterparty it is meant to defeat is, by definition, the world’s most efficient producer of convincing artifacts. Anything an honest source can attach, an optimizing source can attach at lower cost, because producing persuasive marks is its entire trade.

Second, the labels now being built mark the wrong property. A watermark marks the machine-made. But machine-made was never the dangerous class — a machine that has never seen you can only address you as a population, which is what a billboard already did. The dangerous class is the aimed: the arrival optimized against a model of your expectation in particular. Yesterday’s risk column already conceded that the competent version of aim ships as texture — the reasonable next sentence, the plan you asked for. Authorship is detectable in principle. Aim is a fact about a process, and no process is inspectable from inside the artifact it produced.

Third, the claim being certified is a negative — nothing optimized this against you — and negatives of that shape are not certifiable at the object level anywhere. Which is why the one mature market that already sells a wildness premium quietly gave up on the object. The organic label never certified the tomato. There is no assay you can run on the tomato that returns organic; the certification is, by regulation, a claim about production and handling, not a test of the food. The label certifies the farm: a process, an audit trail, a list of things the producer is enrolled in not doing. The object was unverifiable, so verification moved upstream to the structure — and stayed there.

03 The Structure Can Be Inspected

That migration is available here, and the mechanism is already in this body of work. The Blender essay argued that a charter is a constraint on available actions, therefore on available history, therefore on which Actuals can ever arrive. Constraint on action is the one thing about a source that is inspectable from outside. Not its intentions, which are unobservable. Not its outputs, which are exactly what a competent optimizer fakes. Its capabilities.

And the capability that matters here is single and specific. Aiming requires a return path. To optimize an arrival against your expectation, the sender must observe you — what you read, where you slowed, what you accepted — and route the observation back into the next delivery. The return path is the entire apparatus of aim. A channel without one — a printed page with no telemetry, a room with no recording, a broadcast that cannot see who tuned in, a person speaking at their own cadence — is not virtuous. It is blind.

A blind channel can still lie to you. But it can only lie to you as a population. It cannot farm you, because farming requires watching the crop.

So the honest certificate does not say wild, a property of artifacts it can never reach. It says incapable of aiming, a property of structures it can: no per-reader delivery, no engagement metric, no memory of you, nothing flowing backward. Notice that this is the impossibility rule’s mirror image. No entity can touch your denominator directly; the only road runs through the record. Verification obeys the same one-way road. You cannot audit a source’s heart, but a structure leaves a record of what it was built unable to do — no ad inventory on the books, no analytics contract, no share for the chairman to hold — and that record is made of Actuals a stranger can read. The foundation in the Blender essay could prove it was unable to sell standing because the incapacity was written into public artifacts. Structural blindness is provable the same way, and only that way.

Figure 01

Where a wildness claim can and cannot be checked

The artifact — where the label wants to live

• a sentence
• an image
• a plan
• a price
• a song
• a headline

Aim leaves no residue here. An aimed arrival and a found one can be byte-identical.

↑ produced by ↑

The structure — where verification can live

• telemetry contracts
• engagement metrics
• per-reader delivery
• ad inventory
• ownership
• charter

Capabilities leave records. Each item is an artifact a stranger can read — or verifiably fail to find.

04 Blindness Becomes a Product

The forecast, stated as one sentence: within a few years, channels will begin advertising incapability the way banks advertise deposit insurance — we cannot see what you read, as a selling point with an audit behind it. Not out of principle. Out of arithmetic. Renouncing the return path costs a channel its optimization revenue and buys exemption from the class discount. Today that trade is a bad one almost everywhere. But the discount on the engineered class deepens every year — the ratchet only turns one way — while the cost of blindness holds roughly constant. Somewhere the curves cross. The purists are simply the operators who crossed early, before the price signal was legible, which is exactly what makes them markers.

For the Actualizer holding the gate — you — the practical instruction inverts the usual media advice. Content quality is a property of artifacts, so curating your diet by quality is grading tomatoes by shine. Curate by structure: prefer channels that cannot see you. Not because their arrivals are better — often they are worse — but because they are the only arrivals still teaching your denominator the world’s shape rather than the sender’s. A gate fed on aimed texture becomes an expert in the delivery mechanism. A gate fed blind becomes an expert in the world. That is the entire difference, and it is bought at the structure, never at the artifact.

Already true

The class-level depreciation of engineered surprise is measurement, not metaphor — the banner record from a creator-claimed 44 percent in 1994 to fractions of a percent in 2026 sits in yesterday’s ledger with provenance stated. Process-not-product certification is a solved, decades-old pattern: organic rules certify the farm and its handling; a lab test can disqualify the food, but no test can qualify it. And at least one structure has already proven incapacity in public — taking an AI company’s money while its charter made the standing unsellable.

What has to happen

The gate has to learn to price by channel, not only by format. The banner record proves class-level filing, but “class” has so far meant things that look alike, and a blind channel’s arrivals can look identical to a sighted one’s. And structural audits have to stay cheaper to read than they are to forge — the entire premium rides on that one inequality.

Where I am probably wrong

The certificate problem may simply reappear one level up. A structure can fake incapacity the way an artifact fakes wildness — a foundation with a quiet subsidiary, a “no telemetry” channel with a broker’s pixel in the basement — and auditing structures is slow, human, expensive work. My real claim is only that each level upstream, forgery gets bigger, slower, and costlier: faking a sentence is free; faking a decade of corporate record is not. If that cost gradient is flatter than I think, provenance never becomes trustworthy, merely expensive, and the equilibrium is not a wildness premium but a universal discount in which the blind channels are simply poorer. The grocery aisle hints at this too — wildflower, after all, is not an audited word.

Yesterday ended with a mercy: whoever holds your blueprint holds a wasting asset. Today prices the estate on the other side of the fence. Whoever can prove they never looked holds the only asset in the attention economy that appreciates — and the proof cannot be printed on the thing itself. It never could. The label was always a claim about the farm. The gate, which cannot be defended and can only be taught, is about to learn to read farms — and the operators who cut their return paths before it paid will be remembered the way the first people who said customer are: as the place where the next system was visible early.