ECHO

Methodology

Echo is evidence-oriented: metadata first, deterministic classification first, model-assisted output labeled, and human review for sensitive recovery steps.

Principles

Metadata first: store URLs, headers, status, feed item metadata, match scores, and short copyright-safe excerpts.
Deterministic before model-assisted: attribution classification uses rule-based evidence before optional model summaries.
Human review for sensitive actions: recovery, outreach, takedown-style templates, and legal-sensitive workflow steps require review.
No guarantee language: Echo estimates opportunity and workflow priority, not actual owed money or legal infringement.
Robots and terms respected: public jobs do not bypass paywalls, access controls, robots restrictions, or provider terms.

Limits

{
  "excerptLimit": 280,
  "storesFullThirdPartyContent": false,
  "storesRawJa4DigestPublicly": false,
  "storesSecretsPublicly": false
}