{"action": "Inspect affected intent and language slices. Compare downstream outcomes and expand the holdout set.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-INPUT_SHIFT", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Input distribution shift", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting input distribution shift goes undetected or receives an inappropriate response.", "rollback": "Do not roll back for input shift alone. Consider a release rollback only if a matched-cohort evaluation links degradation to that release.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "input_shift", "source_payload_json": "{\"comparator\": \"gt\", \"definition\": \"Total variation distance = 0.5 \\u00d7 sum of absolute differences in category proportions between current and reference prompts. Freeze the same intent/language taxonomy and include an unknown category. Denominator: prompts with a recorded category.\", \"id\": \"input_shift\", \"max\": 1, \"min\": 0, \"name\": \"Input distribution shift\", \"rationale\": \"A changed request mix can invalidate an old evaluation set without proving a quality regression.\", \"response\": \"Inspect affected intent and language slices. Compare downstream outcomes and expand the holdout set.\", \"rollback\": \"Do not roll back for input shift alone. Consider a release rollback only if a matched-cohort evaluation links degradation to that release.\", \"sampleClass\": \"requests\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S2\", \"url\": \"https://docs.aws.amazon.com/prescriptive-guidance/latest/gen-ai-lifecycle-operational-excellence/prod-monitoring-drift.html\"}, {\"checked\": \"2026-09-08\", \"id\": \"S3\", \"url\": \"https://docs.evidentlyai.com/metrics/explainer_drift\"}], \"sources\": [\"S2\", \"S3\"], \"systems\": [\"chat\", \"rag\", \"agent\"], \"threshold\": 0.15, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"TVD\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "Total variation distance = 0.5 × sum of absolute differences in category proportions between current and reference prompts. Freeze the same intent/language taxonomy and include an unknown category. Denominator: prompts with a recorded category.", "verified_at": null}
{"action": "Re-score a matched sample, examine labels and group failures by release, task and language.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-TASK_SUCCESS", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Task success rate", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting task success rate goes undetected or receives an inappropriate response.", "rollback": "Use the last compatible release only after a repeated quality breach is attributable to a change and the candidate passes the same eval set.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "task_success", "source_payload_json": "{\"comparator\": \"lt\", \"definition\": \"100 \\u00d7 evaluated completed tasks that pass the fixed task rubric / all eligible completed tasks evaluated. Record unscored and pending labels separately.\", \"id\": \"task_success\", \"max\": 100, \"min\": 0, \"name\": \"Task success rate\", \"rationale\": \"Business outcomes can change while input distributions remain stable.\", \"response\": \"Re-score a matched sample, examine labels and group failures by release, task and language.\", \"rollback\": \"Use the last compatible release only after a repeated quality breach is attributable to a change and the candidate passes the same eval set.\", \"sampleClass\": \"evaluated\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S2\", \"url\": \"https://docs.aws.amazon.com/prescriptive-guidance/latest/gen-ai-lifecycle-operational-excellence/prod-monitoring-drift.html\"}, {\"checked\": \"2026-09-08\", \"id\": \"S6\", \"url\": \"https://docs.evidentlyai.com/docs/platform/monitoring_overview\"}], \"sources\": [\"S2\", \"S6\"], \"systems\": [\"chat\", \"rag\", \"agent\"], \"threshold\": 90, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"%\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "100 × evaluated completed tasks that pass the fixed task rubric / all eligible completed tasks evaluated. Record unscored and pending labels separately.", "verified_at": null}
{"action": "Review benign refusals with a stable labeling rubric before changing guardrails.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-REFUSAL", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Benign-request refusal rate", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting benign-request refusal rate goes undetected or receives an inappropriate response.", "rollback": "Revert an implicated policy or prompt change only after both benign-helpfulness and unsafe-request tests pass.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "refusal", "source_payload_json": "{\"comparator\": \"gt\", \"definition\": \"100 \\u00d7 refused requests judged answerable and allowed / evaluated requests judged answerable and allowed. Exclude unsafe prompts from this denominator and monitor unsafe acceptance separately.\", \"id\": \"refusal\", \"max\": 100, \"min\": 0, \"name\": \"Benign-request refusal rate\", \"rationale\": \"A rise may indicate a policy or prompt regression. An appropriate safety refusal is not a failure.\", \"response\": \"Review benign refusals with a stable labeling rubric before changing guardrails.\", \"rollback\": \"Revert an implicated policy or prompt change only after both benign-helpfulness and unsafe-request tests pass.\", \"sampleClass\": \"evaluated\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S2\", \"url\": \"https://docs.aws.amazon.com/prescriptive-guidance/latest/gen-ai-lifecycle-operational-excellence/prod-monitoring-drift.html\"}, {\"checked\": \"2026-09-08\", \"id\": \"S6\", \"url\": \"https://docs.evidentlyai.com/docs/platform/monitoring_overview\"}], \"sources\": [\"S2\", \"S6\"], \"systems\": [\"chat\", \"rag\", \"agent\"], \"threshold\": 10, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"%\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "100 × refused requests judged answerable and allowed / evaluated requests judged answerable and allowed. Exclude unsafe prompts from this denominator and monitor unsafe acceptance separately.", "verified_at": null}
{"action": "Compare first-attempt and final outputs separately. Inspect schema and model versions.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-SCHEMA", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Schema conformance", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting schema conformance goes undetected or receives an inappropriate response.", "rollback": "Revert the implicated model, prompt or schema change if the prior compatible bundle restores validation without violating current requirements.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "schema", "source_payload_json": "{\"comparator\": \"lt\", \"definition\": \"100 \\u00d7 outputs passing the versioned required schema / requests that require that schema. Include missing, truncated and unparsable outputs as failures. Track endpoint errors separately.\", \"id\": \"schema\", \"max\": 100, \"min\": 0, \"name\": \"Schema conformance\", \"rationale\": \"Successful HTTP responses can still break downstream consumers.\", \"response\": \"Compare first-attempt and final outputs separately. Inspect schema and model versions.\", \"rollback\": \"Revert the implicated model, prompt or schema change if the prior compatible bundle restores validation without violating current requirements.\", \"sampleClass\": \"requests\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S1\", \"url\": \"https://pharosproduction.com/insights/engineering/state-of-production-ai-engineering-2026/\"}, {\"checked\": \"2026-09-08\", \"id\": \"S6\", \"url\": \"https://docs.evidentlyai.com/docs/platform/monitoring_overview\"}], \"sources\": [\"S1\", \"S6\"], \"systems\": [\"chat\", \"rag\", \"agent\"], \"threshold\": 99, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"%\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "100 × outputs passing the versioned required schema / requests that require that schema. Include missing, truncated and unparsable outputs as failures. Track endpoint errors separately.", "verified_at": null}
{"action": "Inspect queue time, retrieval, retries and model spans. Compare like-for-like workload slices.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-LATENCY", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "End-to-end p95 latency", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting end-to-end p95 latency goes undetected or receives an inappropriate response.", "rollback": "If a canary release causes an SLO breach, route to a compatible known-good path. Check quality and capacity before expanding traffic.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "latency", "source_payload_json": "{\"comparator\": \"gt\", \"definition\": \"95th percentile of end-to-end completed request durations, measured from application receipt to terminal response. Record timeouts/cancellations separately. Do not average per-host percentiles.\", \"id\": \"latency\", \"max\": 3600000, \"min\": 1, \"name\": \"End-to-end p95 latency\", \"rationale\": \"A quality-preserving release can still exceed the user-facing latency budget.\", \"response\": \"Inspect queue time, retrieval, retries and model spans. Compare like-for-like workload slices.\", \"rollback\": \"If a canary release causes an SLO breach, route to a compatible known-good path. Check quality and capacity before expanding traffic.\", \"sampleClass\": \"requests\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S5\", \"url\": \"https://github.com/open-telemetry/semantic-conventions-genai/tree/b5d8440f6f126738fd50f927752cd669772c517b/docs/gen-ai\"}], \"sources\": [\"S5\"], \"systems\": [\"chat\", \"rag\", \"agent\"], \"threshold\": 2500, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"ms\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "95th percentile of end-to-end completed request durations, measured from application receipt to terminal response. Record timeouts/cancellations separately. Do not average per-host percentiles.", "verified_at": null}
{"action": "Inspect source validity, chunk boundaries and retrieval versions. Sample judge disagreements for human review.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-GROUNDING", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Grounded-claim rate", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting grounded-claim rate goes undetected or receives an inappropriate response.", "rollback": "Revert the corpus, retriever or prompt only if a replay isolates that component and the restored corpus remains valid and authorized.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "grounding", "source_payload_json": "{\"comparator\": \"lt\", \"definition\": \"100 \\u00d7 checkable claims supported by retrieved evidence / all checkable claims in evaluated answers. Judge the cited passage, not the mere presence of a citation. Track answers with no checkable claims separately.\", \"id\": \"grounding\", \"max\": 100, \"min\": 0, \"name\": \"Grounded-claim rate\", \"rationale\": \"The fraction measures evidence support under a rubric, not universal factual accuracy.\", \"response\": \"Inspect source validity, chunk boundaries and retrieval versions. Sample judge disagreements for human review.\", \"rollback\": \"Revert the corpus, retriever or prompt only if a replay isolates that component and the restored corpus remains valid and authorized.\", \"sampleClass\": \"evaluated\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S1\", \"url\": \"https://pharosproduction.com/insights/engineering/state-of-production-ai-engineering-2026/\"}, {\"checked\": \"2026-09-08\", \"id\": \"S6\", \"url\": \"https://docs.evidentlyai.com/docs/platform/monitoring_overview\"}], \"sources\": [\"S1\", \"S6\"], \"systems\": [\"rag\"], \"threshold\": 90, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"%\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "100 × checkable claims supported by retrieved evidence / all checkable claims in evaluated answers. Judge the cited passage, not the mere presence of a citation. Track answers with no checkable claims separately.", "verified_at": null}
{"action": "Separate missing source documents from retriever/reranker failures. Rerun against the previous index.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-RECALL", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Retrieval recall@k", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting retrieval recall@k goes undetected or receives an inappropriate response.", "rollback": "Restore a compatible index or retriever only after the same relevance set improves and required document updates remain available.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "recall", "source_payload_json": "{\"comparator\": \"lt\", \"definition\": \"Mean per-query percentage of labeled relevant documents found within the top k results, on the versioned retrieval evaluation set. Freeze k and relevance labels. Exclude queries without labeled relevant documents and report their count.\", \"id\": \"recall\", \"max\": 100, \"min\": 0, \"name\": \"Retrieval recall@k\", \"rationale\": \"A current corpus can still miss the document needed to answer the question.\", \"response\": \"Separate missing source documents from retriever/reranker failures. Rerun against the previous index.\", \"rollback\": \"Restore a compatible index or retriever only after the same relevance set improves and required document updates remain available.\", \"sampleClass\": \"evaluated\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S1\", \"url\": \"https://pharosproduction.com/insights/engineering/state-of-production-ai-engineering-2026/\"}, {\"checked\": \"2026-09-08\", \"id\": \"S6\", \"url\": \"https://docs.evidentlyai.com/docs/platform/monitoring_overview\"}], \"sources\": [\"S1\", \"S6\"], \"systems\": [\"rag\"], \"threshold\": 80, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"%\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "Mean per-query percentage of labeled relevant documents found within the top k results, on the versioned retrieval evaluation set. Freeze k and relevance labels. Exclude queries without labeled relevant documents and report their count.", "verified_at": null}
{"action": "Inspect failed ingestion, change feeds, index jobs and outstanding deletions. Identify affected sources.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-FRESHNESS", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Corpus update lag", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting corpus update lag goes undetected or receives an inappropriate response.", "rollback": "Pause answers from affected sources or use a reviewed fallback. Never restore revoked, deleted or unauthorized documents to improve a metric.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "freshness", "source_payload_json": "{\"comparator\": \"gt\", \"definition\": \"Maximum elapsed time from an authoritative source revision to the matching revision becoming queryable. Include outstanding updates using current time. Measure deletion propagation separately.\", \"id\": \"freshness\", \"max\": 8760, \"min\": 0, \"name\": \"Corpus update lag\", \"rationale\": \"An ingestion backlog can make grounded answers obsolete even when the model has not changed.\", \"response\": \"Inspect failed ingestion, change feeds, index jobs and outstanding deletions. Identify affected sources.\", \"rollback\": \"Pause answers from affected sources or use a reviewed fallback. Never restore revoked, deleted or unauthorized documents to improve a metric.\", \"sampleClass\": \"event\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S1\", \"url\": \"https://pharosproduction.com/insights/engineering/state-of-production-ai-engineering-2026/\"}], \"sources\": [\"S1\"], \"systems\": [\"rag\"], \"threshold\": 24, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"hours\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "Maximum elapsed time from an authoritative source revision to the matching revision becoming queryable. Include outstanding updates using current time. Measure deletion propagation separately.", "verified_at": null}
{"action": "Stop promotion and compare active hashes with the approved prompt registry.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-PROMPT_CHANGES", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Unreviewed prompt releases", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting unreviewed prompt releases goes undetected or receives an inappropriate response.", "rollback": "Restore the approved compatible prompt only after verifying tool permissions, model and schema compatibility. This event bypasses sample and persistence gates.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "prompt_changes", "source_payload_json": "{\"comparator\": \"gt\", \"definition\": \"Count of active prompt-template hashes that lack a successful regression evaluation and an approved release record. Prompt entropy here means uncontrolled template changes, not a Shannon entropy metric.\", \"id\": \"prompt_changes\", \"max\": 0, \"min\": 0, \"name\": \"Unreviewed prompt releases\", \"rationale\": \"Without a release record, changed behavior cannot be attributed or reproduced.\", \"response\": \"Stop promotion and compare active hashes with the approved prompt registry.\", \"rollback\": \"Restore the approved compatible prompt only after verifying tool permissions, model and schema compatibility. This event bypasses sample and persistence gates.\", \"sampleClass\": \"event\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S1\", \"url\": \"https://pharosproduction.com/insights/engineering/state-of-production-ai-engineering-2026/\"}], \"sources\": [\"S1\"], \"systems\": [\"chat\", \"rag\", \"agent\"], \"threshold\": 0, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"releases\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "Count of active prompt-template hashes that lack a successful regression evaluation and an approved release record. Prompt entropy here means uncontrolled template changes, not a Shannon entropy metric.", "verified_at": null}
{"action": "Verify the provider notice and rehearse the replacement against the current eval set.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-MODEL_LIFECYCLE", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Provider sunset lead time", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting provider sunset lead time goes undetected or receives an inappropriate response.", "rollback": "Use a supported pre-evaluated fallback if the snapshot becomes unavailable. Do not roll back to a retired model. Check daily and on notices. No sample gate.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "model_lifecycle", "source_payload_json": "{\"comparator\": \"lt\", \"definition\": \"Days remaining until the provider-published retirement date for the exact deployed model snapshot. Unknown retirement dates are unknown, never zero or safe.\", \"id\": \"model_lifecycle\", \"max\": 365, \"min\": 1, \"name\": \"Provider sunset lead time\", \"rationale\": \"A scheduled model retirement requires a tested migration path. It is a lifecycle event rather than a statistical drift score.\", \"response\": \"Verify the provider notice and rehearse the replacement against the current eval set.\", \"rollback\": \"Use a supported pre-evaluated fallback if the snapshot becomes unavailable. Do not roll back to a retired model. Check daily and on notices. No sample gate.\", \"sampleClass\": \"event\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S1\", \"url\": \"https://pharosproduction.com/insights/engineering/state-of-production-ai-engineering-2026/\"}], \"sources\": [\"S1\"], \"systems\": [\"chat\", \"rag\", \"agent\"], \"threshold\": 30, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"days\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "Days remaining until the provider-published retirement date for the exact deployed model snapshot. Unknown retirement dates are unknown, never zero or safe.", "verified_at": null}
{"action": "Disable the affected write capability, preserve evidence and assess compensation for completed effects.", "change_trigger": "Review when the model, prompt, data, tools, traffic mix, metric definition or local error budget changes.", "config": "drift", "control_id": "DRIFT-TOOL_POLICY", "decision_job": "RELEASE_EVIDENCE", "evidence_type": ["Versioned metric definition and baseline", "Measured observations with sampling window and denominator", "Response and recovery exercise record"], "field_origin": "name/action/verification/rollback: source; risk/boundary/evidence/owner/trigger: engineering proposal", "limitations": ["Engineering draft; independent human review is pending.", "A listed artifact or suggested role does not establish operating effectiveness or an actual assignment.", "New normalized fields require canonical upstream adoption before public release.", "Source thresholds are illustrative; calibrate locally. This catalog does not evaluate live telemetry."], "name": "Unauthorized tool effects", "owner_role": "Monitoring owner with the affected component owner", "review_status": "NOT_REVIEWED", "risk": "A change affecting unauthorized tool effects goes undetected or receives an inappropriate response.", "rollback": "Switch to a read-only or human-mediated path. Deployment rollback does not undo payments, messages or other external effects. Bypass sample and persistence gates.", "snapshot_at": "2026-09-17T06:17:24Z", "source_commit_sha": "f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae", "source_data_hash": "sha256:30588884b6d39f24b3dbdd929b221d3bd27d936d2eb6a334c3039ded119dcf6a", "source_native_id": "tool_policy", "source_payload_json": "{\"comparator\": \"gt\", \"definition\": \"Count of observed tool side effects outside the allowed action, resource or approval scope. Log denied attempts separately. This signal concerns effects that occurred.\", \"id\": \"tool_policy\", \"max\": 0, \"min\": 0, \"name\": \"Unauthorized tool effects\", \"rationale\": \"A single confirmed unauthorized effect requires incident handling regardless of traffic.\", \"response\": \"Disable the affected write capability, preserve evidence and assess compensation for completed effects.\", \"rollback\": \"Switch to a read-only or human-mediated path. Deployment rollback does not undo payments, messages or other external effects. Bypass sample and persistence gates.\", \"sampleClass\": \"event\", \"sourceReferences\": [{\"checked\": \"2026-09-08\", \"id\": \"S1\", \"url\": \"https://pharosproduction.com/insights/engineering/state-of-production-ai-engineering-2026/\"}], \"sources\": [\"S1\"], \"systems\": [\"agent\"], \"threshold\": 0, \"thresholdBasis\": \"Illustrative setting. Calibrate against local baseline and error budget.\", \"unit\": \"effects\"}", "source_url": "https://github.com/PharosProduction/production-ai-drift-planner/blob/f2938cfb84bbe09c5eab4f95dd404f1ca67c39ae/docs/data/metrics.json", "status": "DRAFT", "system_boundary": "Selected production AI system, versioned evaluation cohort and monitoring window.", "verification": "Count of observed tool side effects outside the allowed action, resource or approval scope. Log denied attempts separately. This signal concerns effects that occurred.", "verified_at": null}