{
  "evidence_pack_version": "1.0",
  "benchmark_id": "selective-replay-benchmark-v1",
  "status": "MEASURED_LOCAL_CAPABILITY",
  "claim_boundary": "Synthetic in-process benchmark. It is not a production SLO, certification, customer result or comparison against Temporal/Camunda.",
  "hypothesis": "When a change affects a small dependency-linked subset, selective replay should avoid recomputing preserved records while retaining deterministic results and evidence lineage.",
  "environment": {
    "python": "3.11.9",
    "platform": "Windows-10-10.0.26200-SP0",
    "machine": "AMD64",
    "processor": "AMD64 Family 25 Model 120 Stepping 0, AuthenticAMD",
    "cpu_count": 12,
    "workers": 8
  },
  "tiers": {
    "100000": {
      "records": 100000,
      "dependency_change_rate": 0.01695,
      "affected_records": 1695,
      "preserved_records": 98305,
      "baseline_full_replay": {
        "records": 100000,
        "elapsed_seconds": 0.149713,
        "throughput_per_second": 667946.45,
        "latency_ms": {
          "p50": 0.0014,
          "p95": 0.0019,
          "p99": 0.0026
        },
        "result_digest": "855e810fc79c75c0e451e70af92c867ae2cce90b15542d9b37082915eef591cf",
        "result_xor": 2947027459328139632463191211930155606963475237100147081447140926226654698696
      },
      "selective_replay": {
        "records": 1695,
        "elapsed_seconds": 0.002628,
        "throughput_per_second": 644952.63,
        "latency_ms": {
          "p50": 0.0014,
          "p95": 0.0021,
          "p99": 0.0026
        },
        "result_digest": "d508e6d29600fd61e46603703a555bb3a8134d7ecda9640f6a0504891547b1da",
        "result_xor": 58425842843789716666366899539905595363858594788459846634074966955050923184979
      },
      "concurrent_selective_replay": {
        "workers": 8,
        "records": 1695,
        "elapsed_seconds": 0.007681,
        "throughput_per_second": 220671.52,
        "result_digest": "d508e6d29600fd61e46603703a555bb3a8134d7ecda9640f6a0504891547b1da",
        "result_xor": 58425842843789716666366899539905595363858594788459846634074966955050923184979
      },
      "forced_failure_resume": {
        "crash_after_records": 847,
        "resumed": true,
        "reconstructed_records": 1695,
        "result_digest": "d508e6d29600fd61e46603703a555bb3a8134d7ecda9640f6a0504891547b1da",
        "matches_single_pass": true
      },
      "efficiency": {
        "records_avoided": 98305,
        "work_reduction_percent": 98.305,
        "speedup_ratio": 56.9684
      },
      "negative_controls": {
        "duplicate_events_rejected": true,
        "out_of_order_updates_rejected": true,
        "stale_control_held_for_review": true,
        "malformed_dependency_held": true,
        "tampered_digest_detected": true
      }
    },
    "1000000": {
      "records": 1000000,
      "dependency_change_rate": 0.01695,
      "affected_records": 16950,
      "preserved_records": 983050,
      "baseline_full_replay": {
        "records": 1000000,
        "elapsed_seconds": 1.575872,
        "throughput_per_second": 634569.47,
        "latency_ms": {
          "p50": 0.0014,
          "p95": 0.0023,
          "p99": 0.0028
        },
        "result_digest": "cb319f77d1f07f92051747c3bbc2f1579919fdd29e83fe07a20aa8872c47c1f1",
        "result_xor": 79485291388664942113678644863671944731987861076845724187759152210955750604380
      },
      "selective_replay": {
        "records": 16950,
        "elapsed_seconds": 0.029005,
        "throughput_per_second": 584386.03,
        "latency_ms": {
          "p50": 0.0014,
          "p95": 0.0024,
          "p99": 0.0027
        },
        "result_digest": "ac0f8f8cb597781a175751260acdc38b53e699f50992429c02c48393b6d4e4a6",
        "result_xor": 2211385201915048808394993846768625260212343433560495733131648757899895275279
      },
      "concurrent_selective_replay": {
        "workers": 8,
        "records": 16950,
        "elapsed_seconds": 0.033034,
        "throughput_per_second": 513104.6,
        "result_digest": "ac0f8f8cb597781a175751260acdc38b53e699f50992429c02c48393b6d4e4a6",
        "result_xor": 2211385201915048808394993846768625260212343433560495733131648757899895275279
      },
      "forced_failure_resume": {
        "crash_after_records": 8475,
        "resumed": true,
        "reconstructed_records": 16950,
        "result_digest": "ac0f8f8cb597781a175751260acdc38b53e699f50992429c02c48393b6d4e4a6",
        "matches_single_pass": true
      },
      "efficiency": {
        "records_avoided": 983050,
        "work_reduction_percent": 98.305,
        "speedup_ratio": 54.331
      },
      "negative_controls": {
        "duplicate_events_rejected": true,
        "out_of_order_updates_rejected": true,
        "stale_control_held_for_review": true,
        "malformed_dependency_held": true,
        "tampered_digest_detected": true
      }
    },
    "10000000": {
      "records": 10000000,
      "dependency_change_rate": 0.016949,
      "affected_records": 169492,
      "preserved_records": 9830508,
      "baseline_full_replay": {
        "records": 10000000,
        "elapsed_seconds": 15.727252,
        "throughput_per_second": 635838.97,
        "latency_ms": {
          "p50": 0.0014,
          "p95": 0.0021,
          "p99": 0.0028
        },
        "result_digest": "9aa04c950028a61177f4f16fd6d1781e50dde9e70366d6274a4f918682c037eb",
        "result_xor": 112411147002158359312926658989256372801733505804780234613365872947959507974926
      },
      "selective_replay": {
        "records": 169492,
        "elapsed_seconds": 0.284356,
        "throughput_per_second": 596056.28,
        "latency_ms": {
          "p50": 0.0014,
          "p95": 0.0026,
          "p99": 0.0028
        },
        "result_digest": "94f23b6e257bc591074014d66ee49e832c830c38e7218627d8e6e61d602ce93f",
        "result_xor": 18149519759288365099953621066042953197274679030340124426901388043962301517843
      },
      "concurrent_selective_replay": {
        "workers": 8,
        "records": 169492,
        "elapsed_seconds": 0.282838,
        "throughput_per_second": 599255.12,
        "result_digest": "94f23b6e257bc591074014d66ee49e832c830c38e7218627d8e6e61d602ce93f",
        "result_xor": 18149519759288365099953621066042953197274679030340124426901388043962301517843
      },
      "forced_failure_resume": {
        "crash_after_records": 84746,
        "resumed": true,
        "reconstructed_records": 169492,
        "result_digest": "94f23b6e257bc591074014d66ee49e832c830c38e7218627d8e6e61d602ce93f",
        "matches_single_pass": true
      },
      "efficiency": {
        "records_avoided": 9830508,
        "work_reduction_percent": 98.3051,
        "speedup_ratio": 55.3083
      },
      "negative_controls": {
        "duplicate_events_rejected": true,
        "out_of_order_updates_rejected": true,
        "stale_control_held_for_review": true,
        "malformed_dependency_held": true,
        "tampered_digest_detected": true
      }
    }
  },
  "long_duration_run": {
    "target": "30\u201360 minutes",
    "status": "NOT_RUN",
    "reason": "This capture measures bounded tiers; run the same command with a controlled duration in the target environment before making a stability claim."
  },
  "random_crash_schedule": {
    "status": "NOT_RUN",
    "reason": "This capture uses a deterministic midpoint interruption; randomized crash schedules require a longer controlled run."
  },
  "resource_metrics": {
    "status": "NOT_CAPTURED",
    "ram_peak_mb": null,
    "database_size_bytes": null,
    "reason": "The measured workload is in-memory and this capture did not collect process peak memory."
  },
  "postgresql_run": {
    "status": "NOT_RUN",
    "reason": "No PostgreSQL connection was supplied to this local benchmark; database-backed performance must be measured in the customer-owned target environment."
  },
  "temporal_comparator": {
    "status": "NOT_RUN",
    "reason": "No controlled Temporal implementation was installed for this capture."
  },
  "camunda_comparator": {
    "status": "NOT_RUN",
    "reason": "No controlled Camunda implementation was installed for this capture."
  },
  "adversarial_scope": [
    "duplicate events",
    "stale controls",
    "out-of-order updates",
    "malformed dependency mappings",
    "forced mid-replay failure and resume",
    "tampered result digest"
  ],
  "pass_criteria": [
    "Full and selective replay produce deterministic digests for their declared scopes",
    "Preserved records are not replayed",
    "Forced failure resumes without losing or duplicating replay work",
    "Negative controls are held or rejected",
    "Every measured tier records latency, throughput, environment and scope"
  ],
  "completed_at_epoch": 1788231141.6059148,
  "elapsed_seconds": 18.731,
  "reproducibility": {
    "command": "python scripts/run_selective_replay_benchmark.py --sizes 100000,1000000,10000000 --workers 8",
    "source": "selective-replay-benchmark-v1"
  }
}