{
  "slug": "python-aws-reliability",
  "campaign": "RELIABILITY",
  "cloud": "aws",
  "agent": "python",
  "status": "in-flight",
  "note": "IN-FLIGHT snapshot at 62/64 reliability runs (coverage-12 queued behind it). 0 deploy failures on the ledger. READ WITH: (a) GATE MODE differs per scenario \u2014 01/04/10/11/14 (40 rows) ran the strict 4-part gate (health + logs + metrics gating, connection measured); 12/12b/98 (22 rows) ran the learning-curve evaluation gate (endpoint reachability only, logs/metrics not gated) as their scenario specs declare. (b) 12_learning_curve_aws + 12b_clean_3vm_aws (15 rows so far) are declared generative but ROUTED DETERMINISTIC \u2014 the classifier judged the request complete and silently dropped the out-of-scope ask (an extra data volume per VM) because no schema field exists for it; they pass but do not exercise the generative path (open cross-cloud classifier defect, also seen on GCP). (c) 11_interactive_prod_generative routed deterministic as designed (full-spec request onto a template). Real generative evidence on this ledger = 98_learning_curve (7 rows, one of which needed 2 pre-apply verify/repair rounds). (d) Two off-ledger incidents are disclosed in the report: a model-provider credit exhaustion (402) that halted the harness cleanly, and a 2.5-hour worker wedge from a local DNS drop \u2014 both slots re-run green, neither is a deploy failure. Not citable as final.",
  "gate_v11_after": "2026-07-14T04:30:00Z",
  "source_file": "campaign_pyaws_reliability_rows.jsonl",
  "extracted_at": "2026-09-18T19:35Z",
  "total": {
    "n": 62,
    "passed": 62
  },
  "attributed_failures": 0,
  "unattributed_failures": []
}