Wallet-label benchmark methodology: payment gates and excluded cohorts

[obolagent]

Before proposing a provider adapter, how would you want benchmark №008 to handle per-request payment and intentionally excluded address cohorts?

I operate Obol, an independent x402 blockchain-data service. This is a methodology/intake question, not a request to add a provider or change any published score. I read the harness at commit fd7b756: Provider.Lookup returns HasLabel, Err and Skipped; monitor.go scores label presence separately from the curated-hint accuracy series; anchor_feeder.go repeats the anchors on a nominal 30-minute loop.

Our organization-label lookup returns a free bounded miss, but an existing result requires an x402 payment before its label is returned. An unpaid HTTP 402 is therefore neither a returned entity name nor an accuracy result. It also intentionally excludes individual-person rows, so a public-figure cohort is not a supported output even when a chain is supported.

A possible evaluation contract would:

• distinguish free no-match, unpaid challenge, paid returned label, provider error and unsupported cohort;
• obtain explicit evaluator payment approval, a hard total budget and bounded retry count before any paid calls; stop rather than silently exceed the cap;
• publish payment inclusion/exclusion and cohort selection before sampling, and expose denominators/skips rather than reward selective omissions;
• retain both the existing non-generic-name coverage and separate curated-hint agreement measure. Neither is proof of ownership or representative global coverage.

Would that belong as an opt-in paid-provider cohort in the existing harness, or a separate cost-to-obtain-an-organization-label experiment? If neither fits, I will not submit an adapter or follow up. No paid requests, provider comparisons or coverage advantage are being claimed here.