IMP-013 - Prove runtime readiness and publish freshness lineage¶
Priority / effort: P2 / L
Outcome: Post-publish tests and one operator surface prove live workspace
usability and freshness.
Acceptance: A successful deployment validates required items, bindings,
KQL objects, schedules, pipeline execution, and freshness signals.
Implemented locally:retail-setup verify and
deploy/scripts/verify_readiness.py provide the fixed profile-aware check
taxonomy, read-only default, explicit exact-run pipeline option, redacted
atomic report, and deployment-journal integration. Pagination, definitions,
task flow, KQL sets, schedules, job states, freshness, aggregation,
redaction, CLI, and journal behavior have unit/contract coverage.
Live evidence completed: A full-demo workspace has proven required items,
definitions and bindings, exact setup/ML runs, Lakehouse SQL access,
Reporting, ontology, Data Agents, and task flow. SQL endpoint metadata was
synchronized for every returned table.
Remaining live boundary: Start a bounded stream-events run and retain
recent Silver watermark, Eventhouse ingestion, and checkpoint-identity
evidence. These three checks are optional because streaming is intentionally
manual; they are the only remaining live evidence for this item.
ENH-008 - Use more platform-native monitoring and governance¶
Priority / effort: Idea / M
Outcome: Fabric workspace/Eventhouse monitoring, Monitoring Hub,
governance, catalog, labels, and Purview/DLP touchpoints are part of the
operator story.
Prefer Fabric-native monitoring where it provides the required signal.
Run history and failures are evidence, not disposable console output.
Active-path tests use discovery and fixture-driven markers. Spark tests run
in bounded process-isolated shards, while Windows, E2E, documentation,
notebook drift, and repository contracts feed one required release gate.
Destructive recovery must validate the live target and preserve audit context.
Required deployment steps, task-flow publication, setup, and the exact-run
required ML gate fail the deploy and persist an atomic run journal.
Setup-03 is the single Silver publication boundary; historical Silver and
Gold candidates stage and validate before promotion, with compensating Delta
restore/drop rollback on partial failure.
Streaming Gold stages all ten outputs before promotion and restores prior
Delta versions if an attempted promotion fails.