Loading
A provenance grade describes the record behind a number, never the number. A strategy that lost money with a complete, pre-registered, on-chain record earns a higher grade than one that made money on a claim nobody can check. Inverting that would make the grade a second performance score, which is the thing it exists to counterbalance.
Stated first, because everything below is an invitation and you should know how early it is before reading it.
2 decision commitments exist, 2 of them revealed, across 1 anchored batch and 2 anchors on chain. That is the pipeline working end to end on our own test traffic and nobody else’s. If you would be the first, that is the accurate description of what you would be first at.
Three rungs, and you are already on the first one whether you asked to be or not.
Sign a declaration of a strategy before it goes live — what it trades, what is deciding, the model and its version, hashes of the prompt and configuration, leverage and drawdown limits. We publish its record from that day forward whatever happens to it.
This is free because the thing we need is the denominator, not a fee. Mandatory pre-registration eliminated publication bias in clinical medicine without anyone confessing an old failure, and it works for the same reason here: nobody volunteers a confession, but almost anyone will register something they have already decided to run. The cost to you is a signature. The cost of not doing it is that your good years are indistinguishable from somebody’s surviving account.
Register late and you get grade B, not A−, and the reason is published on the strategy: everything from the declaration onward is attested and unchangeable, nothing before it is, and the decision to register at all cannot be separated from how the strategy had done by then.
Your agent posts a hash of each intent — instrument, side, size, and the stated reasoning — before it executes, and reveals the contents afterwards. We match reveals to on-chain fills, audit the sequence for gaps, and anchor a Merkle root of the day’s commitments on Base.
Each piece is there to prevent one specific thing: committing before execution stops the record being written once the outcome is known; a monotonic counter stops twenty intents being committed and only the winner revealed; the daily anchor stops the database being edited afterwards; the rationale being inside the committed payload stops the reasoning being rewritten to fit what happened.
One gap in the sequence drops the grade to B permanently, because a gap cannot be filled in later. Grade A also requires the operator-level covenant: you register the whole book, or you are not certified. Registering twenty strategies and disclosing the survivor is the exact failure the grade exists to catch.
Grade A− takes one signature and no conversation with us. Sign the declaration from the address the strategy trades from; we recover the signer and check it matches.
For grade A, the commit and reveal endpoints exist and the SDK does not. If you want to be among the first operators attesting per decision, say so through the contact form — the honest position is that we would be building the integration with you rather than handing you a library.
The grading rules are not editorial. They are a single function with the conditions written out, and the methodology page states them alongside everything else we measure.