Methods before claimsAn atomic, verifiable statement linked to evidence, a coverage envelope and separate human assessment dimensions.

Science

Versioned methods, limitations and reproducible evidenceResearch material whose provenance, time, integrity state and link to an exact source version are preserved.–not model marketing.

Every Source AtlasA research catalog of potential Kazakhstan sources with their coverage, owner, risk and mandatory admission gates. connector is disabled. Discovery is not legal or security approval to collect.

No-network capture target preflight

Reject unsafe URLs, redirects and DNS states before an isolated runner performs network activity.

  • normalized target URL
  • redirect and DNS observations produced by the isolated runner

Limitations

  • The method performs no DNS lookup, connection, redirect or content collection.
  • A DNS observation is not authoritative outside a trusted isolated runner.
  • A successful preflight is not legal or contractual authorization to collect.

Full method description →

Evidence SpineThe connected chain from source to conclusion that preserves versions, exact anchors, limitations, transformations and human decisions. contract validation

Verify that a bounded investigation preserves sources, search coverage, evidence anchorsAn exact pointer to a fragment of a document, page, image or media item that supports or refutes a conclusion., separated assessment dimensions, human reviewThe stage where an authorized professional examines evidence, method, limitations and harm risk and records a decision. and an auditable encrypted export.

  • synthetic source passportA versioned source record describing ownership, legal basis, coverage, freshness, risk and collection-admission state. and immutable version
  • bounded synthetic search attemptOne recorded search run with its query, checked sources, time bounds, results, failures and exclusions.
  • synthetic evidence payload and claim

Limitations

  • The fixture represents no real person, organization, procurement or event.
  • Local UTC is not trusted time.
  • The method does not benchmarkA test set with expected answers that measures a reference implementation, not real-world quality. live capture, OCR, retrieval or model quality.
  • Passing this method does not authorize production or private-data processing.

Full method description →

TenantA security boundary isolating one organization's users, cases, keys and policy from every other organization.-local multilingual search baseline

Measure deterministic case-scoped retrieval before selecting a production search owner.

  • allow-listed structured case fields
  • synthetic Russian, Kazakh and English queries

Limitations

  • This bounded linear baseline is not a production search engine.
  • There is no morphology, semantic, vector or cross-organization search.
  • The synthetic set does not establish relevance quality on real Kazakhstan data.

Full method description →

Archived observations

These archived pages are available in Kazakh and Russian.

Reproducible control sets

They test only the local reference implementations and are not a real-world retrieval-quality metric.

Reproducible control sets →