OSINT Hub
Sections
← Back to Science

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

Method cards

Show a method's purpose, inputs, procedure, limitations, version and known errors.

draftVersion 1.1.0

No-network capture target preflight

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

Inputs

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

Outputs

  • policy admission decision
  • stable blocker code without a network connection

Procedure

  1. Parse and normalize the target without resolving a host or opening a network connection.
  2. Reject credentials, unsupported schemes, unsafe ports, local names and literal private addresses.
  3. Evaluate trusted runner observations when supplied and return one explicit admission decision.
  4. Replay the hostile fixture set and bind the result to a deterministic receipt digest.

Known errors and false positives

  • A public hostname may resolve to a private address after this static preflight.
  • A caller-provided DNS observation may be stale or untrusted.
  • A syntactically admitted URL may still violate source terms or collection authority.

Card 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.

Reproduce

The command evaluates twelve safe and hostile synthetic URLs without network access and returns pass when every decision matches the expected outcome.

Commandpython scripts/verify_benchmarks.py --benchmark capture-policy

Owner
owner:science
Last review
Review not assigned
Control sets
science/benchmarks/capture-policy-hostile-v1.json
draftVersion 1.1.0

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.

Inputs

  • 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 evidenceResearch material whose provenance, time, integrity state and link to an exact source version are preserved. payload and claim

Outputs

  • coverage envelopeA required record of checked sources and time bounds, search completeness and the limits beyond which a conclusion must not be applied. and evidence anchor
  • human review decision
  • verified audit chainA log where every event is linked to the previous event hash, making later undisclosed alteration detectable. and encrypted private case bundleAn encrypted portable container for case records, its manifest and audit evidence with mandatory integrity verification.

Procedure

  1. Create an isolated temporary evidence vault and seed the synthetic procurement case.
  2. Record a source versionA source state fixed at a point in time with a digest, locator and change description., bounded search attempt, coverage envelope and anchored evidence item.
  3. Create a claim linked to evidence and require a human review decision before export.
  4. Export an encrypted bundle, replay the audit chain and verify every content digest.

Known errors and false positives

  • A mutable source without a fixed version breaks later reproducibilityThe ability for another researcher to check a result using the same safe inputs and method version..
  • An unbounded search attempt cannot support a claim of absence.
  • A valid content digest does not prove identity, accuracy or source independence.

Card 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.

Reproduce

The command runs the synthetic case from source fixation through encrypted export, then replays the audit chain and verifies every evidence digest.

Commandpython scripts/verify_evidence_spine.py

Owner
owner:science-lead
Last review
Review not assigned
Control sets
none
draftVersion 1.1.0

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.

Inputs

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

Outputs

  • digest-bound ranked record references
  • positive, negative and zero-result measurements

Procedure

  1. Seed one synthetic tenant and a bounded procurement case into a temporary database.
  2. Create language-specific claims without sharing records across tenants.
  3. Run the reviewed Russian, Kazakh and English positive and negative queries.
  4. Compare exact record identifiers and bind the metrics to a deterministic receipt digest.

Known errors and false positives

  • Inflection, transliterationA rule for rendering a name in another script; it aids discovery but does not prove identity. and synonyms outside the fixtures may return no result.
  • A substring match may rank a broad case record alongside a more precise claim.
  • A zero-result response means only that the bounded indexed fields did not match.

Card 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.

Reproduce

The command runs nine queries against a temporary database and verifies exact records for six positive and three negative controls.

Commandpython scripts/verify_benchmarks.py --benchmark local-search

Owner
owner:science
Last review
Review not assigned
Control sets
science/benchmarks/local-search-multilingual-v1.json

Science route tree

Method pages have separate states and never present draft as publication.

Method cardsShow a method's purpose, inputs, procedure, limitations, version and known errors.candidate BenchmarksSynthetic control sets test the reference implementation only.candidate Dataset cardsShow corpus provenance, licence, time window, fields, freshness and limits.candidate ReproducibilityHelps repeat a safe scenario with a fixture, digest, tool version and receipt.candidate Negative resultsRecords a failed hypothesis, the reason for abstention and the next verification step.candidate Science governanceShows the owner, review and release gate for methods and data.candidate Science changelogKeeps evidenced semantic changes to methods, data and control-set versions.candidate