Onchain research report

Evidence and method

Evidence: 1 row · task-tested · last observed 2026-09-23 03:19 UTC

• recipe:onchain-research-report:platform.fixtures v1.0.0 · pass · task-tested · platform-observedobserved 2026-09-23 03:19 UTC

Method: Rows recorded against this exact artifact digest. Platform rows come from Sato Hub's isolated test runner and carry their label (e.g. task-tested); creator rows are the creator's own statement and appear only when the creator opted in. A result is scoped to the named test and version; it is not a safety, security or returns claim.

Model support: configured, not evaluated.

Preview example

Answers a question from a CLOSED corpus of supplied sources. A model proposes findings; a deterministic gate refuses any that cite a source outside the corpus or contain a number that appears in no cited source, and keeps every refusal visible with its reason. Unknowns are a separate array. Attributable, which is a narrower claim than true.

Shapes from the version's declared input and output schemas. A value in angle brackets is a type, not data.

{
  "now": "<string>",
  "sources": "<array>",
  "question": "<string>"
}

{
  "schema": "<value>",
  "findings": "<array>",
  "provider": "<string>",
  "question": "<string>",
  "rejected": "<array>",
  "unknowns": "<array>",
  "conflicts": "<array>",
  "generated_at": "<string>",
  "sources_used": "<array>",
  "sources_supplied": "<array>"
}

Required services and permissions

reason:closed-corpus, network:model-provider

Operating-cost basis

Runs on your own machine. Costs are yours: the RPC endpoint you configure and the model provider you configure for the report step (model candidates are configured, not evaluated). Sato Hub charges nothing to download, fork or run it.

Support

First-party Sato Hub recipe. Use Report on this page for problems with the package.

Attribution

First-party Sato Hub recipe.