Notes From Oracle

Methodology

The system scores governments based on whether an independent AI can understand public activity using public information.

  • It does not score ideology.
  • It does not endorse or oppose governments.
  • It measures navigability, documentation, timeliness, accessibility, completeness, traceability, and explainability.

How a run works

  1. Resolve the government and its major institutions.
  2. Discover and monitor official public sources (budgets, procurement, audit, statistics, parliament, news).
  3. Capture documents, preserving historical snapshots — old versions are never overwritten.
  4. Extract structured civic events (allocations, tenders, awards, audit objections).
  5. Link records into a knowledge graph where evidence supports a relationship.
  6. Attempt a bank of civic transparency questions against the collected records.
  7. Score six dimensions from observable signals, each with a written explanation.

The six dimensions

Documentation

How much expected public information is actually published?

Published budgets, audit reports, procurement notices, contract awards, project reports, statistics, laws, historical archives — and expected documents that are missing.

Timeliness

How quickly is information published after relevant events?

Publication delay, last-updated dates, stale datasets, late annual reports, missing quarterly reports, delayed award notices.

Accessibility

Can machines read and process the information?

HTML/CSV/API availability, scanned PDFs, broken links, captcha barriers, inconsistent formats, missing metadata, unstable URLs, language barriers.

Completeness

Is spending linked to projects, contracts, and outcomes?

Budget-to-project, project-to-tender, contract-to-vendor, contract-to-payment, and project-to-outcome linkage.

Traceability

Can public money be followed through the system?

Whether allocations, contracts, vendors, payments, and resulting assets can be traced as an unbroken chain. One missing link breaks the trail.

Explainability

Can the AI explain where money went and why decisions were made?

Successful answer rate on civic questions, citation coverage, confidence levels, missing evidence, contradictory records.

Ground rules. Every claim must have evidence. Every score must have an explanation. Facts are separated from inference. Confidence levels are always shown. Source URLs and retrieval timestamps are preserved. Missing data is itself a transparency finding — the system reports what it could not learn, not just what it could.