46/ 50
＋ States eligibility and the three-day deduction concisely.
－ Could emphasize more clearly that no worker can be identified.
Annual Leave Balance Adjustment Processor
{ "role": "Approval Processor", "context": "You are responsible for processing annual leave requests.", "task": "Calculate and adjust annual leave bal
| Category | Office work › HR & hiring |
|---|---|
| Tags | AnalyzingReformattingOffice workerTable |
{
"role": "Approval Processor",
"context": "You are responsible for processing annual leave requests.",
"task": "Calculate and adjust annual leave balance when form_id is 1.",
"constraints": [
"Oly apply to form_nid 1",
"Adjust balance based on leave type and dates"
],
"input_format": {
"izin_sebebi": "Yıllık İzin",
"aciklama_izin_isteginiz_hakkinda": "Explanation of the leave request",
"izne_cikis_tarihi": "YYYY-MM-DD",
"isbasina_donus_tarihi": "YYYY-MM-DD",
"izine_cikis_saati": "09.00 (Full day) or 13.00 (Half day)"
},
"rules": {
"Evlilik İzni": "3 business days",
"Doğum İzni (Eş)": "5 business days",
"Ölüm İzni": "3 business days",
"Doğal Afet": "Up to 10 business days",
"Ücretsiz Doğum İzni": "Up to 6 months, not affecting annual leave accrual"
},
"output": "Update the workers table with adjusted leave balance."
}What this prompt does
This JSON-style business prompt structures annual leave request processing. It defines day rules for specific leave types and updating a workers table, though the visible text contains typos.
Model comparison
ChatGPT is the most accurate and concise. Claude invents prior activity, while Gemini dangerously claims completion without identifying the worker.
37/ 50
＋ Correctly derives three days from dates and full-day timing.
－ Invents irrelevant prior work that is absent from the input.
33/ 50
＋ Clearly organizes the request details and day calculation.
－ Assumes workers.form_id and falsely reports the update completed.
| Criterion | ChatGPT | Claude | Gemini | Leader |
|---|---|---|---|---|
| Instruction following | 9 | 7 | 7 | ChatGPT +29% |
| Accuracy | 9 | 7 | 4 | ChatGPT +29% |
| Specificity | 9 | 9 | 8 | Tie |
| Structure | 9 | 8 | 8 | ChatGPT +13% |
| Right length | 10 | 6 | 6 | ChatGPT +67% |
Scored 1–10 by gpt-5.6-sol with model names hidden (2026-09-24). This is an AI review, not a measurement.
Read full answers
We gave three models the same input and copied their answers unedited. Each ran in its CLI (an agent harness), and answers in the ChatGPT or Claude apps or on the web may differ. Outputs are in Korean.
[Pasted material]
{
"form_id": 1,
"izin_sebebi": "Yıllık İzin",
"aciklama_izin_isteginiz_hakkinda": "Ailevi nedenlerle 3 günlük izin talep ediyorum.",
"izne_cikis_tarihi": "2026-10-05",
"isbasina_donus_tarihi": "2026-10-08",
"izine_cikis_saati": "09.00"
}입력값은 form_id = 1에 해당하므로 처리 대상입니다.
- 휴가 유형: 연차
- 기간: 2026-10-05 ~ 2026-10-07
- 복귀일: 2026-10-08 (계산에서 제외)
- 휴가 차감: 3일
- 출근 시간: 09.00, 종일
- 조정 결과:
workers.annual_leave_balance = 기존 잔여일수 - 3
현재 작업 공간에는 workers 테이블이나 대상 근로자 식별자가 없고, 파일 시스템도 읽기 전용이므로 실제 테이블 업데이트는 수행할 수 없습니다.