Skip to main content

Axis score algorithm

Apiome rolls catalog and MCP lint evidence into a multi-axis evaluation. The algorithm identity shown in the UI is:

FieldValue
Algorithm idclx-axis-v1
Algorithm version2 (implementation revision)

Source: apiome-rest/src/app/axis_score.py (ALGORITHM_ID, ALGORITHM_VERSION).

What the axes mean​

Axes include quality, protocol, security, supply chain, supportability, and compatibility. An axis that has not been scanned is not assessed — that is not a clean score.

Composite scores are withheld until required coverage is met (v1: the quality axis).

Supportability, for data contracts​

Since implementation revision 2 (FMT-5.5), the supportability axis is assessed for a data contract — an item imported from ODCS or dbt — from the governance findings the data-contract rule pack produces: is there a reachable owner, a stated service level, a freshness expectation, a documented retention window, a declared serving location. Those findings score that axis instead of quality, so they are never counted twice.

Every other subject is unchanged. A schema language — Avro, XSD, RELAX NG, CDDL, a Kafka Connect schema — shares the data_schema paradigm but has no syntax in which to state an owner or an SLA, so the pack does not run for it and its supportability axis stays not assessed: a capability limit of the format is not a defect of the document. An API-paradigm item is untouched in every respect, including its score, grade and report fingerprint.

Honesty guarantees​

  • Skipped rules (missing transcript, source, SBOM, or probe consent) never count as passes.
  • Static trust-posture findings are signals, not proven exploits, until a consent-gated probe demonstrates exploitability (MCP probes).