bench
Assay lab
Strike a specimen against live chain data. The score is computed twice — once on the server with a deterministic lexical comparator, then again in this browser with an open-weight sentence-embedding model — and both readings name the comparator that produced them.
the two comparators
- server · always available
- lexical-cosine-v1 (char 3-gram + token overlap)
- Character n-gram cosine blended with token Jaccard. No model, no network, no variance.
- browser · on demand
- Xenova/all-MiniLM-L6-v2 (int8, on-device)
- Apache-2.0 weights, 22M parameters, int8 quantized, fetched once and then cached. Runs on your CPU; the token's identity text is never sent anywhere.
engine weights
| factor | weight |
|---|---|
| Metadata collision | 0.34 |
| Market depth | 0.18 |
| Holder concentration | 0.16 |
| Market age | 0.10 |
| Metadata completeness | 0.12 |
| Claim evidence | 0.10 |
mintline-assay-v1.0.0
A factor that cannot be measured contributes zero and flags the result as degraded. Weights are never rebalanced around a gap, so a partial computation cannot masquerade as a confident one.
verdict bands
Attested origin 75–100
No meaningful collision, real market depth, and an inspectable identity.
Safe to attest. Record the claim in the registry and keep the dossier.
Unregistered 55–74
No collision found, but the evidence for a clean origin is incomplete.
Ask the issuer for a signed origin statement before accepting the identity.
Collision suspected 35–54
The identity text resembles a registered origin closely enough to matter.
Treat the name as unproven. Compare the two metadata records before trading or bridging.
Impersonation likely 0–34
Strong collision with an existing origin and thin independent evidence.
Do not treat this identity as authentic. Report the mint to the issuer of the matched origin.