AI work volume
Measures: Number of deterministic demo work records in the active filter.
Derived from: Count of filtered sample records.
Meaning: Context only; higher is not inherently better.
AI Work Analytics ยท deterministic management preview
See how AI software work moves and where human control is required.
Understand routing, review, verification, approval, Work Record, and Ledger activity without inventing customer usage, production telemetry, or agent performance claims.
Deterministic preview onlyNo real customer usageNo production telemetryNon-persistentGeneric Platform Guidance: Human approval required
Management filters
Slice the deterministic control sample
Time windows are fixed demo labels, not historical periods. Every filter remains local and non-persistent.
12 of 12 rows
Showing deterministic preview rows only. Time window: Full Sample. Filter coverage includes All work, Changes requested, and Low control risk through High control risk.
Local preview state
Preparing management analytics
Composing deterministic source records locally. No telemetry query or external call is occurring.
Metric truth
Know exactly what each signal means
No metric claims speed, productivity, quality, compliance, production monitoring, or customer benchmarks.
Capability fit
Measures: How closely seeded agent capabilities match the demo task.
Derived from: Deterministic AgentOS-style routing scores.
Meaning: Higher indicates closer seeded fit, not quality or performance.
Control risk
Measures: How much oversight and approval the work requires.
Derived from: Seeded task sensitivity, permissions, and work type.
Meaning: Higher means stronger controls are required, not that an agent is bad.
Verification rate
Measures: Share of filtered records with verified demo evidence.
Derived from: Verification result equals verified.
Meaning: Higher means more demo records reached evidence review; it is not production assurance.
Human intervention rate
Measures: Share of records requiring a human decision or review.
Derived from: Human-intervention flag in deterministic records.
Meaning: Neither high nor low is universally better; it shows where control is applied.
Lifecycle completion
Measures: Share of records reaching a terminal demo decision.
Derived from: Deterministic lifecycle-completed flag.
Meaning: Higher means the sample journey concluded, not that work shipped.
Linked evidence
Measures: Safe evidence labels associated with filtered records.
Derived from: Sum of evidence counts.
Meaning: More evidence can improve review context but does not guarantee correctness.
Management capabilities
Understand how AI software work moves, where human control is required, and which agents and workflows fit best under oversight.
Summarize deterministic AI work volume and source-module distribution.
Compare recommended primary and supporting agent routing without scoring agent quality.
Show capability-fit signals and control risk as required oversight.
Explain review, verification, approval, policy, block, and changes-requested outcomes.
Measure linked evidence, human intervention, Work Record state, and lifecycle coverage in demo data.
Connect management context to AgentOS, Governance, Review Runs, Verification, Approvals, Policies, Work Records, Ledger, Findings, and Readiness.
Keep production telemetry, benchmarking, persistence, and execution disabled.
Control principles
Bounded, explainable, and human-controlled
Every chart and metric uses deterministic demo data, not production telemetry or real customer usage.
Control risk means required oversight, not agent quality, speed, or guaranteed performance.
Capability fit is a routing hint and does not benchmark agents or guarantee outcomes.
Autonomy is not safety; broader capability requires explicit controls and review.
Generic Platform Guidance: Human approval remains required wherever policy, verification, or risk demands it.
No live execution, external agent call, repo access, GitHub posting, automatic merge, or persistence occurs.
Production analytics, monitoring, compliance reporting, and customer benchmarks are not live.