Agency OS Architecture & Training
Sections

What Jessie can inspect

Data Sources And Evidence

The operational sources Jessie uses for context, evidence, artifacts, state, and model-call observability.

Source Table

SourceProvidesUsed by
AsanaTask brief, comments, assignment, board sections, review trail, attachments.All Jessie skills, Scheduled Monitoring, Platform Build.
Box / client context lakeApproved client documents, context chunks, derived facts, source references.All context-aware Jessie skills.
GA4Traffic, engagement, conversion, event, source/medium, landing-page, device, and location evidence.GA4 Reporting, SEO Reporting, Google Ads, Meta Ads, Integrated Report, Generic Handler, Strategy Planning.
Google Search ConsoleSearch visibility, query, page, and organic movement evidence.SEO Reporting, Integrated Report, Generic Handler, Strategy Planning.
Google AdsPaid search account, campaign, ad group, search-term, conversion, cost, and change evidence.Google Ads Reporting, Integrated Report, Generic Handler, Strategy Planning.
Meta AdsPaid social account, campaign, ad set, ad, placement, creative, action/event, and spend evidence.Meta Ads Reporting, Integrated Report, Generic Handler, Strategy Planning.
Browser RunBounded crawl, page checks, rendered validation, and public-site audit evidence.Technical SEO Audit, SEO Reporting, Google Ads, Meta Ads, Generic Handler.
R2Context chunks, raw snapshots, technical audit packages, generated evidence artifacts.Context Management, Technical SEO Audit, Client Data Plane, Report Artifacts.
D1Run state, work records, context facts, artifact registry, source metadata, run-step traces.Control plane and all hosted runs.
AI GatewayLLM request route, provider/model metadata, latency/error observability.All live Jessie LLM synthesis paths.

Evidence Rules

  • Primary source evidence should stay labelled by source.
  • Supporting source evidence should be used for corroboration, not silent replacement.
  • Missing or partial providers should become caveats.
  • Metric movement should be interpreted against approved context and task notes.
  • The LLM should reason from the evidence capsule, not from unrestricted memory.

Common Caveats

  • GA4 attribution and channel grouping can differ from ad-platform conversion counts.
  • Paid media changes may have same-period or lagged effects.
  • SEO changes can lag and may need Search Console, GA4, SERP, crawl, and link corroboration.
  • A client-approved test may make worse-looking efficiency metrics acceptable for the period.
  • Unavailable provider access should be visible in the report.

Source References