Yandex Direct placement CSV audit

Executed local example — synthetic data

The bundled calculator ran locally. The synthetic CSV is an included input fixture, not a generated deliverable. No advertising account was connected or changed.

python3 -B direct_report_audit.py --json doctor

{"ok":true,"command":"doctor","result":{"version":"0.1.0","python":"3.9.6","offline":true,"auth_required":false,"network":false,"runtime_dependencies":"Python standard library only"}}

The audit used the bundled synthetic placement fixture with max CPA set to 1000 and the other default thresholds. The exact command is in the included SKILL.md.

Complete calculator output, formatted for readability:

{
  "ok": true,
  "command": "audit",
  "result": {
    "schema_version": 1,
    "input": {"rows": 5, "placements": 4, "delimiter": "semicolon", "synthetic": true},
    "currency": "RUB",
    "thresholds": {"min_clicks": 20, "min_cost": "1000", "max_cpa": "1000", "min_conversions": "1"},
    "totals": {"impressions": 4550, "clicks": 97, "cost": "6000", "conversions": "5", "cpc": "61.86", "cpa": "1200.00"},
    "placements": [
      {"placement": "costly.example.invalid", "source_rows": 1, "impressions": 1000, "clicks": 25, "cost": "2500", "conversions": "1", "cpc": "100.00", "cpa": "2500.00", "manual_review": true, "reasons": ["cpa_above_threshold"]},
      {"placement": "example.invalid", "source_rows": 2, "impressions": 1500, "clicks": 30, "cost": "1800", "conversions": "0", "cpc": "60.00", "cpa": null, "manual_review": true, "reasons": ["zero_conversion_spend"]},
      {"placement": "quality.example.invalid", "source_rows": 1, "impressions": 2000, "clicks": 40, "cost": "1600", "conversions": "4", "cpc": "40.00", "cpa": "400.00", "manual_review": false, "reasons": []},
      {"placement": "small.example.invalid", "source_rows": 1, "impressions": 50, "clicks": 2, "cost": "100", "conversions": "0", "cpc": "50.00", "cpa": null, "manual_review": false, "reasons": []}
    ],
    "candidate_count": 2,
    "limitations": [
      "Candidates are for manual review, not a blocklist or evidence of waste.",
      "Check attribution lag, tracking, report period, conversions, and downstream lead quality.",
      "No API calls, campaign changes, automatic exclusions, or savings estimates."
    ]
  }
}

The repeated rows for example.invalid aggregate before threshold checks. It and costly.example.invalid are review candidates; the tool has not established waste or a ready exclusion list. Before using an authorized real report, check its period, currency, chosen goal, attribution lag and downstream lead quality. Every record above is invented educational data.