Static evidence + bounded observation

A checked-in plan says what should exist. A validator can say what one catalog showed at one time.

TraceMap composes those records without turning either one into permission to execute, proof that a procedure worked, or release approval.

Public claim level: demo. This page uses synthetic categorical examples. It contains no raw SQL, connection material, target identities, database output, or operator attestation.

Two evidence lanes

Composition keeps provenance and claim strength separate.

Static repository evidence

Rule ID, evidence tier, coverage label, repository-relative span, commit SHA, extractor version, supporting facts, and limitations show what the scanned commit contains.

present-in-scripts means compatible checked-in evidence exists. It does not establish applied or effective state.

Observed validation summary

sql-validation-summary/v1 binds closed assertion codes to validator identity/version, repository, commit, categorical target context, observation and expiry time, and canonical digest.

The scanner reads this artifact as data. It does not connect to PostgreSQL or rerun the observation.

Human decision

Database owners resolve target selection, authorization, execution order, failures, and approval. TraceMap preserves gaps and non-claims for that decision.

Neither a static tier nor an observed status replaces change control or operator judgment.

Acceptance gates

Observed evidence is admitted only when its bindings agree.

Identityrepository and commit must match the static evidence source.

ContextEngine, server role, database role, schema role, and execution mode must match an expected categorical context.

FreshnessObservation and expiry are evaluated against an explicit deterministic as-of instant.

IntegritySchema, validator/version, closed assertion vocabulary, and canonical digest must verify.

Conflict handlingMalformed, expired, mismatched, ambiguous, duplicate, or conflicting inputs become rule-backed gaps.

Status language

Four categorical outcomes, each deliberately narrow.

`observed-pass`

One fixed catalog predicate was true for that login, target context, and recorded time.

`observed-fail`

The bounded predicate was false. This is a stop and owner handoff, not a generated repair instruction.

`observed-indeterminate`

The harness could not make the bounded observation. Provider prose is not converted into a conclusion.

`not-run`

The v1 harness did not perform that assertion. Connectivity, target-schema compatibility, arbitrary validation queries, and cleanup observation remain explicit gaps.

What this still cannot say

Point-in-time catalog evidence is not procedure evidence.

No function, job, migration, cleanup, rollback, or data movement is executed or observed.
No archive-link connectivity, correct data, replication health, continuing state, or production state is established.
No result certifies execution safety, rollout success, sufficient permission, or release approval.
No credential, connection material, local path, private target name, raw statement, scheduled body, or database output belongs in the public packet.

Continue the proof path

Review the static packet, then carry categorical observations to the owner.

SQL operator handoff Public SQL proof packet Output artifacts Evidence tiers Validation guide Limitations Validation harness documentation Runbook packet documentation