Governed Agent Harness — process metrics
Process telemetry of governed runs, read through the harness application layer. Values are aggregated per run and per project only; the harness produces no per-person indicators. Every metric carries its data quality (OBSERVED, REPORTED, DERIVED, ESTIMATED, NOT_AVAILABLE); token and cost values stay NOT_AVAILABLE unless the provider reports them.
27 run(s) across 17 project(s): agent-results, bdd, broken-baseline, clarification, corrections, delivery, friction, gitlab, integrity, ladder, later-change, memory, node, quickstart, review-exception, tdd, traceability.
Gate states
| Gate state | Runs |
|---|---|
| FAILED | 2 |
| NO_GATE | 4 |
| PASSED | 21 |
| Run state | Runs |
|---|---|
| BLOCKED | 4 |
| CANCELLED | 1 |
| FAILED | 5 |
| PASSED | 17 |
Duration per phase
| Phase | Median ms |
|---|---|
| INTENT | 26 |
| DISCOVERY | 597 |
| SPECIFICATION | 32 |
| PLANNING | 17 |
| IMPLEMENTATION | 104 |
| VERIFICATION | 1,971 |
| INDEPENDENT_REVIEW | 13.5 |
| DECISION | 20 |
| CLOSURE | 102 |
Attempts, corrections and human wait per run
| Project | Run | State | Gate | Impl. attempts | Correction cycles | Review cycles | Non-passed validations | ChangeSets | Human decisions | Human wait ms | Events | Chain valid |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| agent-results | 4a978526c551 | FAILED | PASSED | 1 | 0 | 1 | 2 | 1 | 1 | 4,455 | 130 | yes |
| bdd | f46bc91a13d6 | BLOCKED | PASSED | 1 | 0 | 1 | 1 | 1 | 0 | — | 151 | yes |
| broken-baseline | 1ec9bb15e734 | PASSED | PASSED | 1 | 0 | 1 | 4 | 1 | 1 | 1,726 | 88 | yes |
| clarification | 75a499abd25c | CANCELLED | — | 0 | 0 | 0 | 0 | 0 | 0 | — | 27 | yes |
| clarification | 3367a3fe35c0 | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 2,027 | 156 | yes |
| clarification | 5951219a6fd1 | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 1,987 | 167 | yes |
| corrections | 9566f9401a6e | BLOCKED | PASSED | 1 | 0 | 1 | 1 | 1 | 0 | — | 64 | yes |
| corrections | da4d9ba55fe1 | FAILED | — | 3 | 2 | 0 | 9 | 1 | 0 | — | 117 | yes |
| corrections | 4679069cb57a | PASSED | PASSED | 2 | 1 | 1 | 4 | 2 | 1 | 2,574 | 106 | yes |
| delivery | 4d0c139690cb | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 2,080 | 140 | yes |
| friction | 06bad6c3257e | PASSED | PASSED | 1 | 0 | 1 | 3 | 1 | 1 | 1,951 | 135 | yes |
| friction | 838f37508f48 | PASSED | PASSED | 1 | 0 | 1 | 3 | 1 | 1 | 4 | 145 | yes |
| gitlab | 85e347d65504 | BLOCKED | PASSED | 1 | 0 | 1 | 1 | 1 | 0 | — | 55 | yes |
| integrity | f79fb345784a | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 6,079 | 69 | no |
| ladder | da9eb5363ca7 | BLOCKED | PASSED | 1 | 0 | 1 | 2 | 1 | 0 | — | 136 | yes |
| ladder | 76e1f92c6a1b | FAILED | — | 1 | 0 | 0 | 2 | 1 | 0 | — | 103 | yes |
| ladder | 7805448011cf | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 4,039 | 168 | yes |
| later-change | 342e14218533 | PASSED | PASSED | 2 | 1 | 2 | 3 | 3 | 2 | 1,823 | 116 | yes |
| memory | d1bb8cbfcd03 | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 2,531 | 73 | yes |
| memory | 9802442f9f4f | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 2,565 | 70 | yes |
| memory | fb99a29464f3 | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 2,530 | 69 | yes |
| node | 5be43aa3a748 | PASSED | PASSED | 1 | 0 | 1 | 2 | 1 | 1 | 2,024 | 128 | yes |
| quickstart | 6414c678a191 | PASSED | PASSED | 1 | 0 | 1 | 0 | 1 | 1 | 2,065 | 135 | yes |
| review-exception | d9f1d96ed3c9 | PASSED | FAILED | 1 | 0 | 1 | 1 | 1 | 1 | 2,738 | 72 | yes |
| tdd | 368a37e1733c | FAILED | — | 1 | 0 | 0 | 2 | 1 | 0 | — | 104 | yes |
| tdd | f10e6722daed | PASSED | PASSED | 1 | 0 | 1 | 1 | 1 | 1 | 3,011 | 142 | yes |
| traceability | 22f905aec0a1 | FAILED | FAILED | 1 | 0 | 1 | 1 | 1 | 1 | 5,027 | 126 | yes |
Authorized exceptions
| Project | Run | Gate at decision | Recorded rationale |
|---|---|---|---|
| review-exception | d9f1d96ed3c9 | FAILED | Demonstration fixture: the value is not a real credential |
| Final decision | Runs |
|---|---|
| APPROVE | 16 |
| APPROVE_EXCEPTION | 1 |
| NONE | 8 |
| REJECT | 2 |
Data quality per metric
| Metric | Unit | Quality (runs) | Definition | Source |
|---|---|---|---|---|
| agent.invocations | count | OBSERVED×27 | Number of persisted agent invocation records. | agent invocation records |
| agent.transient_retries | count | OBSERVED×27 | Command-provider calls repeated after a transient failure (runtime.providerRetries). | event store |
| agent.unsupported_claims | count | DERIVED×27 | Findings agent.unsupported-claim: the agent reported success and verification failed. | finding records |
| changeset.files | count | DERIVED×27 | Unique paths appearing in persisted ChangeSets. | ChangeSet records |
| changesets.count | count | OBSERVED×27 | Number of distinct persisted ChangeSet records. | ChangeSet records |
| correction.cycles | count | OBSERVED×27 | Count of authorized transitions back to IMPLEMENTATION: REQUEST_CHANGES decisions and automatic corrections after a failed VERIFICATION. | event store |
| correction.verification_cycles | count | OBSERVED×27 | Automatic corrections after a failed VERIFICATION (runtime.verificationCorrections). | event store |
| cost.usd | USD | NOT_AVAILABLE×26, REPORTED×1 | Sum of provider-reported costs; never inferred from tokens. | resource usage records |
| duration.agent_ms | ms | DERIVED×27 | Sum of persisted agent invocation durations. | agent invocation records |
| duration.human_wait_ms | ms | DERIVED×19, NOT_AVAILABLE×8 | Elapsed time from the latest gate evaluation to the latest human decision. | gate and decision events |
| duration.phase_ms | ms | DERIVED×27 | Sum of completed phase wall-clock durations; parallel phases would be double-counted. | phase records |
| duration.tool_ms | ms | DERIVED×27 | Sum of persisted tool invocation durations. | tool invocation records |
| duration.total_ms | ms | OBSERVED×27 | Wall-clock time between the first and last persisted execution event. | event store |
| human.decisions | count | OBSERVED×27 | Number of persisted human decisions. | decision records |
| human.interactions | count | OBSERVED×27 | Events a person caused on the run: decisions, clarification answers, plan, acceptance and preflight decisions, budget raises, quarantines, contract confirmations and attached evidence. | run events with a human actor |
| implementation.attempts | count | DERIVED×27 | Count of IMPLEMENTATION phase-start events. | event store |
| replanning.count | count | OBSERVED×27 | Count of plan replacement events after the first accepted plan. | event store |
| review.cycles | count | DERIVED×27 | Count of INDEPENDENT_REVIEW phase-start events. | event store |
| tokens.input | tokens | NOT_AVAILABLE×24, REPORTED×3 | Sum of provider-reported input_tokens; no estimation is performed. | resource usage records |
| tokens.output | tokens | NOT_AVAILABLE×24, REPORTED×3 | Sum of provider-reported output_tokens; no estimation is performed. | resource usage records |
| tokens.reasoning | tokens | NOT_AVAILABLE×27 | Sum of provider-reported reasoning_tokens; no estimation is performed. | resource usage records |
| tool.invocations | count | OBSERVED×27 | Number of persisted tool invocation records. | tool invocation records |
| validation.non_passed | count | DERIVED×27 | Validation results whose normalized status is not PASSED. | validation records |