Operate Runway
Use the Runway console to manage governed Databricks delivery, environment metadata, release analytics, logs, evidence, integrations, and recovery diagnostics.
The Runway console is the platform team's operational home. Its information architecture borrows the most useful pattern from modern deployment platforms—start from a release, then move into its logs, resources, evidence, and diagnostics—but keeps the Fabric visual language and Databricks delivery semantics.
Start with Overview
Overview answers the operator's first question: what needs attention now? It shows deployment volume, gate pass rate, the human approval queue, failures, stale evaluation evidence, and recent governed activity. Every row links to the exact deployment rather than an aggregate dead end.
Follow a deployment
The deployment list can be filtered by application, deployment ID, version, environment, and state. A deployment detail page brings four evidence families together:
| Section | Source of truth |
|---|---|
| Artifact identity and digest | Runway deployment projection |
| Databricks App and Unity Catalog resources | Fabric Harness adapter evidence |
| Gate verdict, source, run ID, and freshness | Fabric Experiments |
| Actor, invocation, correlation, and lifecycle events | Fabric Platform |
The page does not calculate authorization in the browser. An approve, gate, rollback, or teardown button submits the registered Runway action to Platform Host, where actor authorization, policy, schema, state validation, durable invocation creation, adapter delivery, and event production run.
Inspect and register artifacts
Artifact Registry uses the shared Fabric family tree-and-detail workbench. Releases are grouped by application, searchable by version, artifact ID, digest, or source revision, and open into an exact provenance view containing the Unity Catalog Volume reference, SHA-256 tree digest, source revision, Harness version, actor, time, and workload manifest.
The system registration guide and form live in the same workbench, but registration remains a governed supply-chain action—not a deployment shortcut. Fabric Platform authorizes and audits the command, and the Harness Volume edge materializes the staged bytes and verifies the digest before Runway issues an immutable artifact ID.
Logs, Analytics, and Release insights
- Logs is a tenant-scoped query surface over Platform events. Filter across Platform, Harness, Experiments, and Runway without creating a second logging write path.
- Analytics derives release throughput, gate success, promotion success, rollback pressure, and application breakdowns from governed completed-day rollups plus uncovered current-day events, with selectable 7/30/90-day or all-history windows and an exact accessible data table.
- Release insights turns the same evidence into an operator decision aid. Its confidence score never authorizes promotion; policy and human approval remain authoritative.
- Observability shows component health, durable action status, attempt count, actor, failure, and safe operator guidance. Failed invocations expose a reason-bound reconciliation request when the user has operator permission.
Environment and integrations
Environment Variables provides a Vercel-inspired key matrix across Preview, Development, and
Production plus bulk updates. Only key names, presence, scope, author, and sync time are visible.
Values enter an encrypted tenant-bound command store and the governed runway.secrets_sync action
carries only an opaque command ID.
Integrations shows the composed Fabric and Databricks dependencies: Platform Host, Harness, Experiments, Lakebase, Unity Catalog Volume, and Databricks Secrets. It deliberately has no direct connector mutation or generic retry control. New configuration mutations must first become typed, registered Platform actions.
In the hosted product preview, the status Preview adapter means the real governed action ran but its external effect was handled by a local adapter; it cannot modify a customer workspace. In a customer-installed production App, Connected means the corresponding live binding is active.
Governance settings and evidence reports
Governance Settings is the tenant-admin surface for evidence freshness, strict Experiments
provenance, audit retention, trusted worker identities, the business-unit label, and memberships.
Each save invokes runway.policy_set, runway.tenant_upsert, or runway.membership_set; no settings
table is edited out of band. RUNWAY_TENANTS_JSON remains the first-start seed. Once a tenant has
event-sourced memberships, those assignments become authorization truth for that tenant.
Every deployment links to a Release Evidence Report. It assembles the artifact digest, evaluation request and exact verdict, human approver, trusted-worker execution, and immutable event IDs into an auditor-readable chain with a canonical report digest. The daily audit export remains the durable full-ledger recovery artifact in Unity Catalog.
Recovery semantics
The production worker automatically claims pending or interrupted idempotent invocations using a
lease. A terminal external-effect failure is different: an effect may be partial, so Runway shows
the correlated invocation and requires an operator reason. runway.reconciliation_request records
that decision; the trusted worker then issues the domain-specific exact follow-up and records its
new invocation id. This is reviewed recovery, not an invocation-row reset or blind retry.
One mutation pipeline
Humans, CI, API clients, and agents use different surfaces, but every write converges on the same
platform.fabric.pro governed mutation pipeline. Console analytics and logs are read-only
projections over that pipeline.