Classification

Classification assigns each golden record its ServiceNow CI class. Rules are data, not code: an ordered rule set evaluated per record, authored during onboarding, refined through governed review, and auditable after the fact — every record carries the rule that placed it and the evidence it matched, and can show its full rationale on demand (see Evidence & reasoning).

Deployable leaf classes only

Automatic classification targets the most specific deployable class the evidence supports. Generic parent classes — the abstract layers of the CI hierarchy — are rejected as rule targets three times over: at rule-authoring time by the validator, at deployment, and at runtime, where a match against a banned class routes the record to review instead of writing a bad class silently.

Collision-guarded rules

Substring predicates are the classic classification trap: a rule matching one product string can catch an unrelated one (the industry has real examples of a mobile OS string colliding with a network operating system). Rules built on Rule authoring requires contains-style predicates to be checked against the observed values of every source, and consumer-platform rules to carry a corroborating condition before they ship.

Doubt routes to review

• →Records the rules cannot place with confidence go to a review queue with the evidence attached — never into a catch-all default.
• →Live discovery evidence outranks the incumbent CMDB class when they disagree; the disposition report shows the reclassification and why.
• →Every rule change is regression-tested against pinned full-run baselines before it ships.