Availability, as of September 2026. A Corpus Fit is run by us, on our infrastructure, and delivered to you as documents. The containerised version, where the engine runs inside your own environment, is not available yet. If your records cannot leave your environment, say so when you write and we will tell you honestly whether we can serve you today rather than after you have paid.
Here is what the engine does with a corpus of your type. It reads it blind and names the seats it can carry: the people, each with a job, whose work your records would support, and the work each would do there. Then, for the seat you name, it says how far your records let that seat go and where it must stop. It has been pointed at patent filings, federal regulations, research literature, security guides, grant records and a year of storm reports, and it read each of them differently, because they are different. What comes back is what your records can carry, and for whom.
To start, email TheArchitect@livingedenframeworks.com with a corpus sample. If you already know what you need built, send that too, and the Assessment is run against it.
A Corpus Fit Read: given no ask, on purpose, it names the seats your corpus can carry, each with the work it would do there.
A Corpus Fit Assessment: given the seat you bring, your ask, it says what your records bear out for that seat and what the tool must never claim.
| Corpus Fit Read | Corpus Fit Assessment | |
|---|---|---|
| You bring | a corpus sample | a corpus sample and the seat you want built |
| Direction | we tell you what seat(s) are there | you indicate the seat you need or want |
| The question | which seats can my corpus carry | for the seat I ask for, how well and far, do my records allow that seat to exist. What’s missing in my corpus that would support the seat |
| Reach | broader | deeper |
Chain the runs by taking your CFR run putting it through a CFA. Those CFR seats become the ask the Assessment is then ran against.
SAMPLES →One Read and one Assessment, both on public records, delivered as a client receives them.
Either route is a Corpus Fit engagement. Both field licenses reference that term, and the enclave deployment terms make a paid Corpus Fit engagement a prerequisite to a Secure Enclave deployment. That fee credits in full toward the Base Integration Fee where the enclave agreement is signed within thirty days of delivery. The routes, the terms and what is and is not licensable are set out on the licensing page.
CFR: A Proposal names the seats your corpus can carry and, for each, the tool that seat would use: a person with a job, an analyst, a curator, a planner, a billet, never a department. It says what each tool would deliberately not be, and whether anything like it already exists and where: on one corpus, ten of twenty-one proposed functions already existed elsewhere. It arrives in three parts: the page you decide on, the tools, and the audit, which carries what the engine withdrew, any claim it made and then ruled against itself for want of evidence. That is not an appendix; it is the point. Each tool is a named path, specified and not yet built, and it is stated as that: never as a blank, and never as something on a shelf.
CFA: A Specification names the shape the tool takes for your ask, and then where your ask and your records disagree: what you say about your records, checked against a read taken before your ask was seen. It closes with what the built tool must never claim, split into limits that hold on any corpus and limits that lift if your data improves.
Both state what could not be seen as a boundary rather than leaving it as a silence, and every artifact opens by naming who reads it.
The evidence behind every line of the document above, written so the finding can be checked rather than believed — by you, or by your own AI. It carries which stages ran and which did not, the corpus profile the read was made from, and the provenance of each claim, so a result can be re-derived rather than taken on trust.
Your engagement is run twice, end to end, by two independent AI houses. Each one takes it the whole way: the Corpus Fit Read or the Corpus Fit Assessment, and the build that follows from it, with everything put into its own handover folder for that build. Anthropic Claude is the floor, the house this engine is built and worked in. xAI Grok runs the same flow from the start, with its own read, its own documents and its own words. Neither house is shown the other’s work.
The two folders are then compared by meaning, on our own machine, with no model call and no cost carried into your engagement. Where one house named a seat the other did not, the other is asked one narrow question about that seat, and “not there” is a complete answer rather than a gap.
The split happens at the handover folder, and the one that ships to you is Grok’s, because that is the substrate the environments we serve are able to permit. The Anthropic folder ships alongside it only where Grok has not performed on par during our test phase. The point is not which house is right. The difference between them is itself a finding, and what one house did not see travels with your delivery, named, rather than disappearing into a single confident answer.
One limit worth stating plainly. Where a route checks whether a proposed tool already exists, it searches a local index of roughly forty thousand public AI models and applications and surfaces the closest matches to each proposal, which are what the judgement is made against. It is a similarity search, not an exhaustive comparison against all forty thousand. That is a proxy for the public AI tool landscape and not for the product landscape, and closed commercial products are absent from it. An absence there is not evidence that something is novel, and we do not report it as one.
A build is two halves: the engine scaled to the seat’s tool, and the tool. A licence carries two documents. The BuildMe says what the tool is, the records it runs on, what a run emits, and what it must never claim. The second says what the engine has to be at for that tool to work. The container is one delivery: those documents and the engine together. You build the tool into it, the engine is already there and wraps around the seat you built, and the two are tested together before either is called done. The availability notice at the top of this page stands: the container route is not available yet, and this paragraph says what a licence will be, not what is on offer today.
The engine works on records. A seat’s work that is not done on records is outside it: a physical build, a decision the law gives to a person, a live feed the container never sees. The first question we ask of any request, before a Corpus Fit is priced, is whether delivering it would need a person we do not have. If it would, we say so then, not after you have paid.
The BuildMe. For every tool a Corpus Fit proposes, our engine also writes an executable specification: the document that tells a developer, or an AI, exactly how to build that tool against your records. It states the functions, what each one rests on, what must exist first, and the conditions under which it is finished.
It is not a description of a tool. It is a set of instructions for building on our patented engine. Its closing section is titled What powers it, and it names the specific filed positions each function runs on, marking some of them primary, and the tool does not exist without it. The LEF Ai Engine is the base; the tool is the defined work it does inside your corpus. Take the base away and the document specifies nothing that can be built.
So handing it over is not handing over a plan. It is handing over the operating instructions for the engine itself, against the patent positions that make it work. That is licensed, not sold, and it is not part of a Corpus Fit. The second document of a build, the engine scaled to that tool, travels with the licence for the same reason. A Corpus Fit tells you what your records can support and whether it is worth building. The BuildMe comes with the licence rather than being priced separately.
You are not left holding a puzzle. The Proposal names each tool, the seat it serves, and what it must never claim, which is enough to decide with and enough to take to anyone else for a second opinion. What you do not get for $12,000 is the engine.
Living Eden Frameworks LLC | UEI EZ5VPPFP6NV3 | CAGE 1A3A2 | NAICS 541715