Who it’s for
Wherever capital is committed against evidence.
EliAI™ MRF is in market for funds that purchase medical receivables at a discount and collect from the settlement. The architecture underneath is not specific to that asset — it is specific to the problem of paying for a claim whose value turns on facts nobody at the desk is qualified to read.
The three readers
Three people have to believe a mark before it is worth anything.
A valuation that convinces the desk and not the committee has not converted into anything. Eli is built so the same artifact satisfies all three, without anyone re-doing the work in a different format.
The principal
You sign, and you carry the mark. You need a reason to believe a number before you fund against it.
Positions that re-mark themselves on every event, and a valuation you can re-derive in front of a committee years later.
The underwriting desk
You read files nobody at the desk is qualified to read end to end. The arithmetic is the last five minutes of a week.
Four specialist readings on every file, attributed, allowed to disagree, with the evidence attached.
Audit and the trustee
You are asked to stand behind a valuation somebody else struck, on evidence you did not see.
An append-only trail, controls enforced in the ledger rather than in the process, and a deletion certificate that names what survives.
Is this your desk
Five things that have to be true.
If four or five of these describe your week, the conversation is worth an hour. If one does, it probably is not.
- You commit capital against evidence somebody has to read — a file, not a form.
- The thing that decides whether you get paid is a fact, not a rate: whether the treatment was defensible, whether the security is enforceable, how the counterparty behaves.
- More files reach a human than humans can read properly, so the marginal file gets the thin version of the analysis.
- You are asked to defend a mark to somebody who was not in the room — a committee, an auditor, an LP, a trustee.
- Your book re-values on events you do not control and do not learn about on a schedule.
Where it is not the answer
Three desks we would send elsewhere.
Saying this on a sales page costs us the meetings that were never going to close, which is the cheaper side of the trade for both of us.
You want a faster spreadsheet.
The arithmetic was never the slow part. If your files are homogeneous and the inputs arrive clean, a model is the wrong shape of spend.
You need a number on every file, always.
Eli refuses to price a cohort that will not carry a price, and that refusal is enforced in the database rather than in a guideline. If a blank is unacceptable, this will frustrate you by design.
You want the model retrained on your book.
There is no training and no fine-tuning anywhere in this product. Valuation is retrieval-grounded over comparable settled matters; your history deepens the corpus, it does not move any weights.
For the practitioner
If you are the one who will actually use it.
Everything above is written for the person who signs. This is the path for the person who will have to live with it.
Bring Eli a file you already know the answer to.
Tell us what your desk underwrites. We will run a worked valuation on a matter of that kind — synthetic, so nothing of yours leaves your building — and you tell us whether the answer is one you would have signed.
A person reads it and replies. No sequence, no drip, no form acknowledgement.