Manager materials now carry AI claims the way they once carried Sharpe ratios: an edge from language models, a proprietary pipeline, research capacity multiplied. Some of it is real. Testing which is specialized work, and building an AI diligence desk to check one deck is not a reasonable answer. A read is the alternative: you send the claim, we score it, you decide with evidence.
1
You send the claim
As the manager presented it: the deck pages, the DDQ answer, the pitch memo, plus whatever backup came with it. No access to the manager is needed, and the manager is never contacted without your instruction.
2
We score it
Against the published SPEC standard: what is actually specified, what is evidenced versus asserted, what could be verified by a third party, and where the claim would fail under the questions a diligence process should ask. The same method behind our
published citation audit, which graded 134 claims of a widely circulated AI research report in public.
3
You get the verdict memo
What verified, what did not, the verdict in one line, and the questions to put back to the manager, with the answers that should satisfy you and the answers that should end the conversation. Delivered within ten business days of the claim pack.
4
The fee
From $5,000, one claim per read, set before work begins. Confidential on both sides; the memo is yours to use inside your process.
A read scores the claim as presented, not the manager. Its job is to tell you which questions to ask before you allocate, and which answers should end the conversation.
A read settles one claim. Most allocators who commission one are looking at a pipeline, not a deck, and the engagements behind the read exist for that:
5
The committee workshop · from $15,000
The read outsources one verdict; the workshop builds the muscle. A day with your investment committee or board, working live claims from your own pipeline through the SPEC method: what to ask, what an answer has to contain, and when to walk.
6
The retained desk · from $8,000 per month
For allocators who see AI claims every quarter. Every claim that crosses your desk gets a read, manager responses get scored, held managers get re-read when their materials change, and your committee has a standing line for AI questions. The verdict memo told you which questions to send the manager; the desk is there when the answers come back.
Tell us about the claim you want read and we will come back with scope and a start date.