Service operations and monitoring
Monitor platform health, investigate queued work and manage operational AI hosting, blockchain and banking work items.
- Check service healthOpen System Health and review queues, failed jobs, scheduler heartbeat, backup recency, disk usage, cache and queue reliability.
- Find the affected work itemUse the relevant managed-project, blockchain service, banking request or finance operation list and verify its current state.
- Assign and record workUse the available assignment, stage or status controls and enter accurate operational notes or external status where requested.
- Retry only when appropriateRetry a failed job only after understanding its failure and checking for duplicate side effects. Use the work item history to avoid repeating completed work.
- Confirm closureRe-open the record and confirm terminal status, owner, required deliverables and audit trail. Escalate unresolved system-level faults.
System Health
System Health brings together queue depth, failed jobs, disk usage, cache status, queue reliability, scheduler schedule and next due times, scheduler and log heartbeat, and last successful backup. It is a diagnostic dashboard, not a general server shell. A failed-job retry action requeues work; it does not repair the underlying cause. Check the exception summary and queue before retrying one job or all jobs. High disk use, absent heartbeat or a failed cache indicator should be escalated promptly.
- Open System Health and compare the current indicators with recent observed values.
- Inspect the failed job queue and failure summary; identify whether retry could repeat an external action.
- Retry only the justified job and refresh the page to confirm whether it completes or fails again.
Scheduler, queue and backups
Scheduled commands show their schedule expression and next due time. The page reports when scheduler and log heartbeat files were last written, giving a signal of whether background processing is alive. Queue reliability also summarises dead-letter and idempotency issues. The backup panel shows a recent-success timestamp or reports that none was found. If an expected run is missing, correlate the heartbeat, queue and backup state before raising an incident; do not invent or run shell commands from this dashboard.
- Check last scheduler and heartbeat times, then inspect due times and queue depth.
- Compare dead-letter/idempotency counts and backup recency with the expected service schedule.
- Report the non-secret indicators and time observed to the platform operator if processing is stale.
AI hosting project operations
Managed Projects is used to track accepted AI service work. The list and project detail show quote or job information, assignment and progress. Staff can update status and assignee, upload deliverables, record model versions, submit change orders and complete defined stages; quote approval is separate. Use only the customer project linked to the task, and confirm that a stage is ready before marking it complete. A quote, imported catalogue record or configured model does not itself prove a live deployment.
- Open the managed project and review its quote, current status, assignee and incomplete stages.
- Assign an authorised owner and attach the correct deliverable or model-version record.
- Complete a stage only after its acceptance evidence is present, then verify the new project state.
Blockchain service operations
Blockchain Operations lists service orders and their tasks. Eligible support and admin staff can assign, start and complete tasks, and record external or governance status. These controls record operational progress; do not claim they execute an on-chain transaction unless the relevant task evidence confirms it. Keep the order detail history up to date and ensure the correct task is selected. Catalogue changes are handled separately in Blockchain Commerce and belong to God-only administration.
- Open the service order and inspect task assignment, invoice links and existing status history.
- Assign the task, mark it started, and record the verified external or governance status when applicable.
- Complete the task only after its outcome is evidenced; re-open the order to confirm the timeline.
Banking request operations
Banking Administration is God-only and provides a request list, request detail, quote creation and lifecycle transition. The page shows request events and associated service order and invoice where present. A customer-submitted request or administrative quote is not evidence of funds movement. Check the terms, supported currency and request state before quoting or transitioning. If a request needs more information, use the appropriate workflow status and keep the reason recorded in the event history.
- Open the request and inspect its submitted details, quote, service order and event history.
- Prepare the approved quote with the supported currency and accurate price and terms.
- Use the authorised transition and check the updated state and event entry; escalate any mismatch.
Finance operation alerts
Finance operations pages expose health, queued runs, start/succeed/fail/replay actions and alerts that can be acknowledged. These are operational controls for finance processing, not instructions to alter ledger entries manually. Review the run and alert context, use the appropriate state transition and record the actual outcome. Replay may repeat a workflow, so check idempotency and prior result first. Keep permission boundaries and approval requirements intact.
- Open the finance operation health view and inspect the run state and related alert.
- Choose start, succeed, fail or replay only when the recorded evidence supports that transition.
- Acknowledge the alert after the condition is understood and verify the run’s final status.
Frequently asked questions
Does retrying a failed job fix its cause?
No. It requeues the job. Inspect the failure and possible side effects first, then escalate repeated failures.
What does queue reliability show?
The dashboard reports dead-letter and idempotency failure indicators alongside the queue and failed-job state.
Does marking a blockchain task complete send a transaction?
The control records task completion. Confirm separate transaction evidence before claiming an on-chain action occurred.
Does a banking quote mean payment was made?
No. A quote and request transition are workflow records, not proof that funds moved.
When can an AI project stage be completed?
When its required work and acceptance evidence are present. Confirm the project record before changing the stage.
What should I send when escalating health issues?
Share the relevant status, time observed and non-secret failure indicators through the approved operations channel.