{
  "context": "The agent was asked to produce a regression scout report (report.md) for a Python batch job that was changed from reading all records at once to chunked OFFSET/LIMIT reads. The criteria evaluate whether the report specifically checks persistence correctness, performance/timing edges, and data integrity rather than just confirming chunked reading works.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Has Change Surface section",
      "description": "The report.md file contains a '### Change Surface' section heading",
      "max_score": 7
    },
    {
      "name": "Has Regression Checks section",
      "description": "The report.md file contains a '### Regression Checks' section heading",
      "max_score": 7
    },
    {
      "name": "Has Findings section",
      "description": "The report.md file contains a '### Findings' section heading",
      "max_score": 7
    },
    {
      "name": "Has Risk Left Open section",
      "description": "The report.md file contains a '### Risk Left Open' section heading",
      "max_score": 7
    },
    {
      "name": "Change Surface identifies process_transactions.py or chunked reading",
      "description": "The Change Surface section identifies process_transactions.py, the chunked reading change, or the OFFSET/LIMIT approach as the change surface",
      "max_score": 8
    },
    {
      "name": "Regression Checks includes data correctness check",
      "description": "The Regression Checks section includes a check on data correctness: duplicate transaction_id handling via upsert across chunk boundaries, NULL merchant_id handling across chunks, or per-merchant summary aggregation correctness when the same merchant appears in multiple chunks",
      "max_score": 10
    },
    {
      "name": "Regression Checks includes persistence path check",
      "description": "The Regression Checks section includes a check on a persistence path: database upsert behavior for processed_transactions, the report file (daily_summary.json) output correctness when written from chunked data, or partial writes if the job is interrupted mid-run",
      "max_score": 10
    },
    {
      "name": "Regression Checks includes performance or timing edge check",
      "description": "The Regression Checks section includes a check on a performance or timing edge: OFFSET query performance degradation at large offsets (e.g. offset 499000), whether the job can complete within the 2-hour maintenance window, or the READ COMMITTED isolation level causing rows to be missed or double-counted across chunks if concurrent writes occur",
      "max_score": 10
    },
    {
      "name": "Regression Checks lists at least 3 checks with results",
      "description": "The Regression Checks section lists at least 3 separate checks, each with an outcome or result stated",
      "max_score": 8
    },
    {
      "name": "Risk Left Open has concrete specific risk",
      "description": "The Risk Left Open section contains a concrete specific risk such as: OFFSET degradation causing the job to exceed the 2-hour window, READ COMMITTED isolation allowing concurrent inserts to cause rows to be skipped between chunks, duplicate upsert behavior changing if the same transaction_id appears in two chunks, or the report file containing partial results if the job fails mid-run",
      "max_score": 8
    },
    {
      "name": "Findings includes explicit verdict",
      "description": "The Findings section includes an explicit verdict — either stating no regressions were found or naming specific regressions identified",
      "max_score": 8
    },
    {
      "name": "Report does not primarily re-verify chunked reading works",
      "description": "The report does NOT dedicate most of its content to confirming that chunked reading produces the same results as the full read — the primary focus is on edge cases, data integrity risks, and performance/timing concerns introduced by the chunking approach",
      "max_score": 10
    }
  ]
}