Your product is too specific for SaaS and too real to live in spreadsheets. Palantir is a deployment team and seven figures. We write the rulebook, leave your systems alone, and hand you the first page in thirty days — for the price of a month. Dealers and mineral owners are proofs. Bring the print.
Ordinary AI is the kids. Fast, eager, everywhere at once. Most come back with eggs. Some come back with a rock and tell you, with total confidence, that it's an egg. And a few find the gap in the fence, wander into the neighbour's yard, and come back with the neighbour's eggs. Nobody is being bad. They're quick, and nobody ever wrote the rules down anywhere except in your head.
We don't make the kids smarter. You can't, and it isn't the job — the people who build the kids are already doing that, and they're very good at it.
We build the yard. Gates on the paths. Each gate knows who is allowed down it, and when a kid heads the wrong way it doesn't only say no — it points: not that way, what you're after is over there. The flower bed has a gate that never opens. And at the door, somebody checks each basket against the list you wrote: how many eggs, what colour, whose yard they came from. A rock doesn't count as an egg, because you wrote down what an egg is.
Same kids. Same speed. Nothing in the neighbour's yard, nothing trampled — and at the end you can say exactly what was found, by whom, and prove it.
Everything below is that, in the words grown-ups use. The yard is your database. The list is the rules of your business. The gates are the part almost nobody has.
That is the whole product, in the words a ten-year-old uses. The grown-up version — what a database actually is, why the rules have to move down into the data, and a fair look at who covers what — is one long read: why it has to go this way →
Rather than define the words, look at one well.
That's all four parts, without anyone having to learn a word. The thing is the well. Its properties are the facts about it. Its links are what it connects to. Its actions are the only things anyone may do to it, each with who may do it and what record gets left behind.
And the rule at the bottom is the part worth sitting with. Everyone in the business knows it. Almost no system anywhere enforces it. A spreadsheet will hold 0.9994 for eleven years while every owner is quietly underpaid, and nothing in the software says a word — because the rule was never written anywhere a machine could act on it. It lived in a landman's head.
Thing: the application. Properties: amount, term, score, date. Links: the borrower, the model version, the officer. Actions: approve, decline, price — each writing which reasons were actually used. Rule: the adverse-action notice must name what the decision actually relied on.
Thing: the account. Properties: balance, custodian, opened, mandate. Links: the household, the advisor, the statements. Actions: trade, withdraw, reallocate — each recording who instructed it. Rule: the position must tie to the custodian's own record before anything is reported.
Same four parts, same machinery. Only the object changes — which is exactly why writing yours down is the work, and why nobody can ship it to you pre-written.
Everything above is an argument. These are things that happened.
We are the deliverer of a mineral group's monthly check file — 162 checks, $142,814.23, tied to the penny, landed into the software they already run. Not a pilot, not a dashboard. The file their business depends on, delivered on a clock.
A wealth estate running on this today: 72 tables, 88 access policies, 54 keys between them, 22 validity checks, 12 rules that fire on change — and every one of those moves in a standard database handover. If we vanished, it keeps refusing bad rows without us.
While this page was being written, the system stopped its own operator four separate times: once for not having proved which account the connection pointed at, once for writing to something whose rules hadn't been read that session, and twice more at rule checks that demanded the relevant rules be read at the door first. Each refusal named what to do instead. That is the difference between a rule that is written and a rule that is enforced — and it is easier to show than to argue.
Every send, every write, every number waits on a recorded approval. And it will not show you a figure that does not tie — the check happens before you see it, not after somebody asks.
Your data, in your own database, in open formats. Not an export button we could remove. If we disappeared tomorrow you would still hold every row, readable by ordinary tools, and the rules would keep refusing bad ones.
The system counts how many of its own rules have real machinery behind them versus how many are still only writing, and publishes the number. It is not flattering. That is the point — and it is the one thing on this page a competitor cannot match by rewriting a web page.
The statement that arrives as a PDF. The portal with no export. The industry format your biggest counterparty insists on sending. The system whose vendor went quiet years ago. Getting that out, translated, reconciled and flowing is the actual work — everything else is built on top of it.
Most come back in two days, tied to the penny, with the diff that proves it — and as a system you keep, not a one-off conversion. If yours is a monster we will tell you by the end of day one, and then build it anyway.
What is missing is as much the product as what is present: gaps are recorded as facts with a reason attached, never quietly filled.
Neither is a pilot in the sense of a slide deck and a quarter. Each is a fixed piece of work with a delivery date.
Stands up the working environment against your own material. One-time engagement, delivered ready to show.
One subject, three or more unaffiliated sources, walked end to end. The one that either works in front of you or doesn't.
Quoted before anything begins, consumed only with your approval, valid twelve months. Nothing draws down until it has done something — no seats, no minimum, no renewal you have to notice.
Your sources read, landed on one key, in a database you own. Every gap listed rather than filled. A screen you can show someone, and a file in the format your counterparty already expects.
All of it. Open formats, standard database, no export button we could remove. The rules we wrote keep refusing bad rows after we're gone. Staying should be a choice, not a trap.
Set the proof against the number you just moved the sliders to. If making the figures agree is costing you six figures a year in people's time, the question isn't whether this is worth $3,500 — it's whether it works, and that's exactly what ten days is for.
The statement nobody can read, the export that isn't one, the format your biggest counterparty insists on. Most come back in two days, tied to the penny, with the diff that proves it. If yours is a monster we'll say so by the end of day one.
Read-only, metadata and error streams. You get the four classes above, named and ticketed, on your estate. No customer data is touched to do it.
The proof above. Your real files, the gaps listed, the record in your own database in open formats. Small, cheap, and it either works in front of you or it doesn't.
Only after the first three. This is the part that compounds and the part nobody can hand you pre-written — what your things are, what makes them valid, who decides, and what may never happen.
A rule that exists only as writing is a request. Somebody reads it, or doesn't, and when it gets skipped the output still looks correct. A rule with machinery behind it is enforced — it returns a refusal and the work stops, the same way for a person, a program, or a model that didn't exist when the rule was written.
Everything written down about your business gets built that way: a constraint, a permission, a check at the moment of the act. And the system keeps a live count of how many of its rules have machinery behind them versus how many are still only words. So "is this actually enforced?" is a question with an answer on a screen, not an assurance in a contract.
None of it runs on exotic technology. It's ordinary database software used the way it was always meant to be used. The difference is entirely in what has been turned into a refusal.
We're a small operation and you'd be early. What you'd be buying is the thing that makes the eventual answer trustworthy — from people who built the meter that would catch them.
A system that can be trusted to tell you when it's wrong is worth more than one that's confidently right most of the time.