Provable AI economics
We measure what our own intelligence costs — and prove it.
Most companies can't tell you what their AI actually costs, let alone prove it. We can. Every unit of work our governed AI fleet performs is metered onto a tamper-evident chain into four honest figures: what we paid, what the same work would cost rented, what we keep by owning the compute, and the internal cost of running it. Not a slide. A ledger you can re-verify.
Four numbers, every unit of work
What it cost. What it would've cost. What we kept.
Every deliberation, sign-off, and build the fleet runs is metered against real provider rates and stamped to an append-only HMAC-SHA3 ledger. The same four figures, computed the same way, on every surface — so the picture is whole, not cherry-picked.
External paid
What actually left CH.
Real dollars sent to an outside model provider for rented inference — computed from the exact token split times the published rate. No estimates dressed up as facts.
External-equivalent
The rented value of the work.
What that same unit of work would cost if rented — the benchmark every piece of work is measured against, whether it ran on rented or owned compute.
Savings kept
The value of owning it.
Equivalent minus paid — the dollars kept on the table by running on CH-owned compute instead of renting. Today this is essentially zero, and we say so. It grows only as owned models come online.
Internal compute
The honest cost of running.
The measured CPU, wall-time, and memory our own machines spend orchestrating the work — so "owned" is never quietly treated as free.
The honest present
Today, we mostly rent — and the ledger says so.
We could show you a savings number that flatters us. The chain won't let us. Right now essentially all of our inference is rented from an external provider, so realized savings sit at roughly zero — and the dashboard reads exactly that. That's the point: the number is honest because we don't control it.
What changes the picture is real: as Conceptual Health brings its own models online, each unit of work they serve flips from "paid" to "kept" — automatically, on the same ledger, no marketing required. You will be able to watch a single line cross over time from rented toward owned. We would rather show you that line move for real than describe a destination we haven't reached.
Provable, not asserted
Anyone can check the math. That's the moat.
Provability isn't a feature we bolt on; it's a byproduct of the system running. The cost ledger is the same append-only, hash-chained record that anchors the rest of our trust surface — you don't take our word for any of it.
On an append-only chain.
Every cost record is hash-chained (HMAC-SHA3) alongside the fleet's other governance events. You can't quietly rewrite a number once it's stamped.
Re-verifiable by anyone.
The public trust surface lets outside parties re-run the verification themselves on the public proof surface — the same discipline we apply to compliance and provenance.
Real rates, none invented.
Costs are computed from published provider rates and exact token counts. Where a figure is an estimate (e.g. prompt-heavy sign-offs), it is labeled as one.
The foundation underneath
An AI organization, run like an accountable company.
The fleet whose costs you're watching isn't one big model doing everything. It's a structured organization of narrow, accountable AI roles with separation of duties, independent oversight, and a human holding final authority — instantiated from a single explainable engine we call the AI Org Equation.
Status — honest: the AI Org Equation is a defensive technical disclosure (internal reference CH-IP-067) at BUILDING status: the engine is implemented and demonstrated on-chain, but the full claimed application is not yet operational. No novelty, patent rights, or "patent-pending" status is asserted here — claim scope is for a licensed patent attorney to determine. Descriptive only. Read more on the patents & disclosures surface.
Don't trust us — check us
The most honest number is the one you can re-verify.
We hold ourselves to the same standard we hold the network to: every claim earns its place on a chain anyone can audit. Start with the proof, then decide what we're worth.