Methodology
A short reference. The landing page has the narrative.
Cumulative since May 5, 2026. Once a day we poll Polymarket and run each market through three gates. Markets land in Direct source, Alternative source, or Currently unresolvable.
Gate 1: on-chain feed filter.
Is the market bound to an on-chain price feed (Chainlink, Pyth)? If yes, the chain has the answer and we step out.
Gate 2: source URL.
Do the resolution criteria name a source URL we can fetch? Markets without one wait for a later agentic-search release. We don’t ship best-effort scraping.
Gate 3: accessibility.
Can a validator reach the named source? If the host blocks validator traffic, we route to a verified alternate with the same fact. If no alternate exists (paywall, login, captcha, pure-consensus subjective market), the market goes to Currently unresolvable.
Off-chain step: deeper-page agent.
When a market names a host but not a specific URL, an off-chain agent picks the page and hands the URL to the Intelligent Oracle. The agent runs off-chain; the oracle runs on-chain.
Bradbury is GenLayer’s testnet. The benchmark runs in Studio today (GenLayer’s developer environment); moving it to Bradbury runs each market through real validators and surfaces per-validator on-chain transactions on every per-market detail page.
A TLS notary is a cryptographic receipt that specific bytes came from a specific HTTPS origin at a specific time. It’s the same primitive GenLayer Labs is piloting in its Twitter bounty, applied to any web source. One validator fetches with its own credentials; the proof is portable, and the rest verify it without re-fetching. For this benchmark, it unlocks most of Currently unresolvable: paywalled, logged-in, rate-limited, or IP-locked pages. Not shipped yet; on the roadmap alongside hardened alternates and the accuracy backtest.
