Skip to content

Methods

How we handle data, make claims, and communicate certainty. These standards are codified in the platform contract and apply uniformly across every application.

▲Trust Model▲

Every data point on the platform carries three orthogonal dimensions of trust metadata, displayed alongside the data itself. Users never have to guess how much to trust what they see.

Data age

Freshness

  • Fresh — last sync within expected cadence
  • Stale — cached data beyond normal TTL
  • Cached — upstream unreachable, serving last-known-good

Confidence

Certainty Level

  • High — directly observed, validated, or complete data
  • Medium — partially observed or interpolated
  • Low — heuristic, modeled, or sparse data

Provenance

Data Origin

  • Exact — directly from authoritative source
  • Inferred — derived from repeated observations
  • Synthetic — modeled or estimated from incomplete data
▲Claim Ladder▲

Every derived claim is inspectable through a standard ladder. The path from a headline claim to the raw source record is always navigable. No claim exists without at least one indicator and one evidence point.

1 Claim what we're saying
2 Indicators quantitative signals that support it
3 Evidence specific data points underlying the signals
4 Derivation how we got from data to indicators
5 Confidence how much we trust this derivation
6 Caveats what could be wrong or what we don't know
7 Raw records the actual source data, inspectable
▲Editorial Standards▲

Language on the platform is confidence-gated. We lead with the observation, not the judgment. We never imply intent — "concentrated" is not "collusive," "unusual" is not "suspicious."

Confidence Phrasing Example
High "is", "shows", "confirms" "72% of contracts awarded with single bid."
Medium "suggests", "appears to", "indicates" "Data suggests a concentration pattern."
Low "may", "possible", "insufficient data" "Possible pattern; data is incomplete."
▲Flagging Vocabulary▲

We use

  • › unusual
  • › concentrated
  • › notable
  • › worth inspection
  • › statistically significant
  • › structurally notable

We never use

  • corrupt
  • fraudulent
  • rigged
  • illegal
  • criminal

Unless there is direct authoritative legal basis.

Platform contract

All methods above are part of the g0v.mk platform contract — 17 sections covering trust model, observation model, harvest pipeline, finding schema, evidence drawer, entity pages, export contract, editorial language, data quality indicators, bilingual handling, and correction policy.

Read the platform contract →
· ꙮ ·