Observation Framework

Evidence-first · local-first · inspectable

Keep the source visible.
Keep uncertainty intact.

A compact framework for collecting geopolitical source records and building explicit, reviewable interpretations over them.

This Pages site publishes documentation and portable specs only. A separate lightweight Fly service hosts the collection-only watcher, observation ledger, and dashboard. RhetorLint, assessment authoring, and optional model work remain outside that hosted runtime.
Open live ledger Explore the contracts Read the boundaries

Method

Separate records from interpretations

Each layer has a deliberately narrow job. The framework preserves exact source material and asks analytical statements to show their evidence and uncertainty.

  1. 01

    Capture

    Store a bounded, immutable Observation with provenance.

  2. 02

    Isolate

    Expose exact passages, source context, and visible actor aliases.

  3. 03

    Annotate

    Optionally mark visible language patterns with local RhetorLint.

  4. 04

    Assess

    Link scoped claims and falsifiable hypotheses to cited passages.

Published artifacts

Small contracts, directly inspectable

These files are copied from the repository during deployment. They describe shapes and identities; none is a live intelligence feed.

JSON registry Published copy

Source registry

Delivery, publisher/control lineage, roles, authority scope, and caveats. It contains no truth, reliability, or credibility score.

Open sources.json
JSON registry Published copy

Actor registry

Canonical names and bounded aliases for visible-text matching. It does not rank power, assign responsibility, or describe motives.

Open actors.json
JSON Schema Draft 2020-12

Assessment schema

A portable shape for passage-linked claims, scoped power bases, alternatives, falsifiers, contradictions, and unknowns.

Open assessment schema

Epistemic boundaries

Validation is structure, not a truth machine

A valid packet can be traced and reviewed. That does not make its contents correct.

What the framework can preserve

  • Exact visible passages and source-record lineage
  • Claimed source ID and canonical-host consistency
  • Bounded, potentially incomplete actor-alias visibility
  • Evidence links, alternatives, limitations, and falsifiers

What it does not establish

  • Truth, deception, source credibility, or authentication
  • Independent corroboration from publisher diversity
  • Power from mention volume or capacity from identity alone
  • Intent from rhetoric, capability, or automated inference
Intent belongs here only as a falsifiable, evidence-linked hypothesis with alternatives—not as mind-reading or a confidence score.