{
  "title": "Living Scenarios: real Forge ingestion benchmark",
  "recordType": "Curated summary of an existing staging benchmark; not a new run",
  "evidenceDate": "2026-09-29",
  "environment": "Forge staging environment on a Jira Cloud free-plan test site, driven by the actual CLI from WSL2 in California",
  "workload": {"featureFiles": 500, "scenarios": 5000, "targetIssues": 1000, "cucumberJSONResults": 20000, "data": "Generated test workload"},
  "results": {
    "firstFeaturePushWallSeconds": 19.9,
    "unchangedRepushWallSecondsRange": [4.5, 6.3],
    "resultsUploadWallSeconds": 18.1,
    "featureUploadBatchP95Seconds": 3.7,
    "resultUploadBatchP95Seconds": 1.14,
    "finalResultsBatchSecondsRange": [5.3, 5.7]
  },
  "findings": [
    "Forge SQL rejected large derived-table updates that succeeded in local TiDB; bounding those updates to 100 rows resolved the observed finalization failure.",
    "Jira bulk property updates required completion polling before the next write.",
    "The final result batch met its 20-second handler budget but exceeded the tighter 5-second per-batch target."
  ],
  "limitations": [
    "Generated workload on a test site, not customer traffic.",
    "In-Jira panel latency, rendering, SQL request counts, and actual platform compute costs were not measured by this benchmark.",
    "The app remains unreleased to the marketplace."
  ]
}