Audit scoring contract
The helper accepts normalized JSON independent of any particular Tableau MCP response:
{
"methodology_version": "1.0",
"definitions": [
{"id": "def-1", "name": "Revenue", "status": "assessed"},
{"id": "def-2", "name": "Restricted", "status": "failed"}
],
"findings": [
{
"definition_id": "def-1",
"rule_id": "format.currency_unspecified",
"evidence_key": "representation.currency_code",
"severity": "medium",
"summary": "Currency format has no currency context",
"observation": false
}
]
}
Definition status is assessed, partial, or failed. Required finding keys are definition_id, rule_id, evidence_key, severity, and summary; observation is optional. Severity is critical, high, medium, or low.
The helper rejects unknown keys, duplicate definition IDs, findings against failed definitions, invalid types/enums, and methodology versions other than 1.0. It deduplicates by definition, rule, and evidence key, keeping the highest severity. Observations never affect scores. Partial definitions are scored provisionally but excluded from the overall composite.
Readiness is separate from the hygiene score
Output includes per-definition readiness and overall_readiness. Any scored critical finding makes readiness not_ready, even if the numeric grade is high. Incomplete/failed coverage produces not_assessed unless a known critical finding already makes the result not ready. Fully assessed results with no critical findings return no_critical_findings, not a guarantee of readiness. Present this status before scores and prioritize critical repairs.