The
trial.
An industry that appraises claims, a cost structure that just collapsed, and an assay office for company formation.
A cupel is the vessel used in fire assay. The ore goes in and the answer comes out.
For six hundred years, before anyone would accept a claim about what a piece of ore was worth, the ore was destroyed. Heat, bone ash, and a small porous cup. Base metal is absorbed into the vessel and what remains is the part that was always precious. The assay office did not argue about the sample. It ran it.
We took the name because the instrument is the company. A company is a claim about what buyers will pay for, and there is exactly one way to settle a claim like that. You put it in front of the buyers, under a fixed protocol, on a fixed clock, and you read what comes out.
Venture appraises claims. We test them. The distinction is the entire business.
The judgment happens before the evidence exists.
An early-stage investor sees thousands of companies a year and can seriously evaluate a few dozen. Under that constraint, the job stops being measurement and becomes appraisal: verdict a founder, a market, and a deck, and forming a view faster than the evidence could be produced.
Appraisal is not stupid. It is what you do when running the assay is more expensive than guessing. For most of venture's history that was true. Producing the artifact that would settle the question, a working product a real buyer can use and pay for, cost more than the check being decided.
But appraisal has a signature failure. When you cannot measure the thing you care about, you price the proxies. And the proxies venture reaches for are the ones a founder arrives holding: a built product, a funded runway, a warm introduction. Every one of those is cheap to observe and expensive to acquire, which is exactly what makes them a wealth test wearing a diligence costume.
Appraised
instead of
assayed.
The standard path never runs the assay. It appraises. A first check is priced off proxies that are legible before anyone forms a view of the business, and the founder pays for those proxies in advance.
Time held in reserve. A fundable early-stage company takes a year or two to assemble. Counting foregone salary and living costs, attempting one costs a mid-career professional somewhere between $150,000 and $250,000, paid in advance, non-refundable, before anyone with capital spends forty minutes on you.
A loss-tolerant address book. The first stage on every diagram of the funding pipeline is not an institution. It is a requirement that the founder personally know people holding tens of thousands of dollars of loss-tolerant capital. There is no version of that which is not a wealth test.
A social graph. Most funds source through warm referrals and cold decks convert near zero, so the third requirement is a social graph that happens to touch investors. This is the proxy the industry noticed, and softening it accomplished little, because the founder who could not afford the runway never reaches the open application either.
None of this requires anyone to hold a prejudiced belief. It is what appraisal costs when nobody runs the assay, charged to the founder, in advance.
Our claim is deliberately narrow. Not that investors are biased. Something more mechanical and easier to check: the proxies are gated on wealth, they are read before evaluation begins, and the founders they remove are never seen at all. Their absence from the funnel then gets read, in good faith, as an absence of ambition.
The assay got cheap enough to run instead of estimate.
Nobody needed permission to fund runways and replace friends-and-family rounds. It simply never penciled, because manufacturing a fundable company from a standing start cost roughly the same amount the founder was being asked to absorb. Carrying that per company, across a portfolio, is a charity budget rather than an investment strategy.
That constraint broke on one specific dimension. We hold the claim narrowly: software engineering in general did not get ten times cheaper. But the particular artifact a founder needs to clear the evidentiary bar, a working product a real buyer can use and pay for, went from a technical team and a year to a matter of weeks for a studio running modern tooling. We know, because building them is what we do.
When the cost of the experiment falls below the cost of the guess, appraisal stops being rational. You run the assay. That is the whole opening, and it will not stay open forever.
Register. Run. Read. Publish.
A trial is not a program. It is a protocol with the order fixed in advance, and the order is the part that makes the result mean anything.
Register. Before the clock starts, the criteria are written down: the buyer list, the wedge, the revenue and usage thresholds that will produce each verdict, and the version of the predictor being tested. Criteria set after the data arrives are not criteria.
Run. Ninety days. Thirty to build, sixty to sell. Cupel carries the build, the entity, and the operations so the founder's job compresses to the one activity that tests the hypothesis: selling to the buyers they said they could reach. No curriculum. No demo day. No audience other than the buyers.
Read. On day ninety the registered thresholds are applied to real billing data and a verdict is issued. The verdict is computed, not negotiated.
Publish. Every verdict goes on the record, including the ones that embarrass us. A trial whose failures are private is a marketing exercise.
Paying customers, a repeatable motion, and a founder the buyers respond to. Cupel capitalizes the next stage and the company raises on a record rather than a deck.
Revenue exists but the motion has not compounded yet. The clock extends on stated terms, with the extension and its reason published like any other result.
Something structural is out of phase: a budget cycle, a regulatory gate, a buyer who agrees but cannot sign this year. The work is preserved and the verdict says why.
The hypothesis was tested and it failed. The company is wound down, the founder keeps the artifacts and the record, and the result is published with the rest.
One vocabulary note, because we are strict about it internally. The day-ninety output is a verdict. The month-nine durability test, which asks whether the customers stayed and the motion held, is a verdict. They answer different questions and we do not let one stand in for the other.
Buyer access, v1. Stated so it can be beaten.
Every trial needs an intake hypothesis: the thing you believe predicts the outcome, written down before you select anyone. Ours is buyer access. Not credentials, not pedigree, not the pitch. Whether the founder can reach, by name and this month, the people who hold the budget.
It carries a version number for a reason. v1 is a guess with an argument behind it, and the calibration ledger exists to tell us how good a guess it was. When the data says a component does not predict, that component comes out and the version increments in public.
Not the segment. The buyers, by name and company, reachable this month. The threshold is derived from your market's deal size and realistic warm conversion. In most mid-market B2B it lands near two hundred names. In a concentrated segment with six-figure contracts it might be twelve. A founder who can produce only a category has a plan about a market, not access to one.
Former colleagues. Past customers. People who owe you something. Two founders who look identical on paper diverge enormously on this question, and almost nobody asks it.
If every sale is independent, acquisition cost never falls and the company scales only with spend. If each customer produces a reference, a referral, or a pull, the curve bends.
Each question demands an artifact rather than an assertion. That is deliberate. A test you can talk your way through is an interview.
Quitting is a wealth test wearing a commitment costume.
The founder with the deepest buyer access is frequently the one who cannot afford to leave: fifteen years inside an industry, a mortgage, and the exact relationships the trial is designed to test. Demanding resignation as proof of seriousness selects for savings, not conviction.
So Cupel 180 runs the same protocol on a part-time clock. Everything is async. There is no mandatory daytime attendance. The registered thresholds are calibrated to part-time hours rather than borrowed from the full-time track, because a threshold that assumes forty hours is a resignation requirement with extra steps.
Before day one we clear employment and IP with the founder, in writing. A trial built on a conflict the founder has not resolved is not a trial, it is a liability.
The assayer owns the ore.
We hold equity in the companies we grade. Any assay office in that position should be distrusted by default, and we would rather state the problem than let someone else discover it.
The defenses are structural. Thresholds are registered before the run. The verdict is computed from billing data rather than argued in a room. KILL is published with the same prominence as GO, and the calibration ledger tracks whether our GO verdicts survived, which is the number a dishonest assayer would most want buried.
The economic argument matters more than the procedural one. A single flattered position is worth less than a record people believe. The instant the verdicts are understood as marketing, they price at zero and so does everything downstream of them.
A verdict is
worthless until
someone grades it.
Every verdict we issue is tracked to month eighteen and published annually, including every GO that died and every KILL that thrived. That second column is the one that matters. Any grader can look accurate by grading generously.
The ledger is empty today, and saying so is part of the protocol. Cohort 01 has not run. Zero verdicts sit on the record. Treat this document as a pre-registration rather than a result, and hold us to the interpretation rules we publish before the first company is admitted.
Seven conditions, written before the data arrives.
A thesis that cannot fail is not a thesis. We published the conditions that would retire this one: access failing to predict revenue, the part-time clock producing systematically weaker verdicts, our GO verdicts not surviving to month eighteen, and four more.
Two limits belong next to them. The comparison group is small, roughly one admission in ten, so a fifty-company record can reject a strong effect and cannot establish a subtle one. And tracking of declined applicants is consent-based and therefore self-selected, which biases it in a direction we will name rather than correct away.
We know of no one else in venture willing to publish that list.
Where this applies, and where it does not.
This thesis applies to business software sold to identifiable buyers who hold budgets, which is most of what venture funds and all of what we build. It applies weakly where the technology itself is the barrier and there is no distribution problem yet. In frontier research and regulated science, capability is the constraint, credentials may genuinely proxy for it, and we make no claim there.
And a limit on the idea itself: access explains who reaches customers, not who keeps them. Relationships produce the first hundred. Product depth and durability decide what survives. That is why the month-nine verdict exists, and why we will never pretend a day-ninety verdict says more than it does.
An assay office
for company
formation.
We source where the proxies were never available: operator communities, vertical practitioner networks, regional markets, the professions that produce access instead of credentials. We select on measured buyer access, blind to wealth and identity alike. We carry the build and the entity, and we run the operations. Ninety days later a verdict is computed from real billing data and published whether it flatters us or not.
Appraisal is our protection, which is the irony the position rests on. Anyone can read this and agree with it. Acting on it requires an operating studio, a sourcing channel into communities that take years to earn, and a willingness to publish failures. The industry's economics push against all three. We expect to be copied eventually and we expect the copies to be late, because the hard part was never the insight.
Venture never decided that family wealth should gate who gets to build. It built a system where that is what happens, three times over, before anyone is evaluated at all. We took the gates out, kept the evaluation, and put a verdict on the record every ninety days.
Fifteen years inside a market.
A list of buyers.
We'd like to see it.
If you have been told to come back when you have traction, the trial is how you get it.