CareerVector × Management

Purpose

Operate and support CareerVector without creating a privileged browsing surface over private workspaces or an unrecorded path for agents to alter user state.

Operator pathways

• Understand aggregate service health, realtime behavior, work queues, model availability, token use, local-execution adoption, and error classes.
• Investigate a user-reported problem with explicit, scoped evidence supplied by the user or generated by privacy-safe diagnostics.
• Manage Included AI supply, provider availability, zero-cost budgets, and model catalogues.
• Review QA, releases, incidents, migrations, and public status.
• Execute participant erasure and verify completion across product boundaries.

Invariants

• There is no general operator workspace browser.
• Private prompts, documents, keys, and workspace content do not enter telemetry by default.
• Every write to workspace truth uses the same typed operation contract as other actors.
• Support capability is explicit, scoped, revocable, and visible to the participant.
• Aggregate learning and observability cannot reconstruct a person's workspace activity.

Decisions still required

• Support-capability lifecycle and evidence-sharing UX.
• Privacy-safe telemetry tiers and redaction contract.
• Included AI capacity policy, fairness, and exhaustion behavior.
• Operator approval boundaries for model/provider changes.

Completion evidence

Resolve a real support case using only scoped evidence; prove denial of unrelated workspace access; audit every state-changing action; verify erasure and key isolation end to end.