Yes, with one condition. A self registered score is only as trustworthy as the signatures behind it. If the registration, the verifier results and the score are all signed with secp256k1, a quantum capable adversary can forge every one of them later, and all of that history is public today. The score should be bound to evidence that stays verifiable under NIST FIPS 204 and FIPS 205 signatures, in line with the CNSA 2.0 transition window.
First feeling, before any risk number, is to ask who built this agent, whether its code is what it claims to be, and whether that proof will still hold in five years.
Two building blocks may help here. In the Qstamp SDK family from Quantova Inc, each agent record, credential or model output is reduced to a salted SHA3 256 fingerprint, signed with ML DSA 65 and anchored on chain, so anyone given the record can verify it while nobody else learns its contents. On the Quantova network the QVM runs contracts produced by an attested compiler, so the deployed code of an agent is cryptographically tied to the exact source that built it, and every account and validator vote is signed with ML DSA 65. Applied to ERC 8126, the four verification layers could anchor fingerprints of their findings, and the score would be recomputable from evidence that a quantum computer cannot rewrite.
Following and happy to share the fingerprint layout. or find more info on Qstamp . org website and on npm under the name Quantova