Predicting At-Risk Primary Students in Bangladesh: Grading Artifacts, Half-Yearly Signals, and the Class 3 Failure Breakpoint

This dataset contains anonymized academic records for 4,632 primary school students (Classes 1–5) across nine schools in the Sylhet division of Bangladesh, covering academic years 2024 and 2025. It was collected to support early identification of at-risk students using half-yearly (mid-year) assessment signals and accompanies the paper "Predicting At-Risk Primary Students in Bangladesh: Grading Artifacts, Half-Yearly Signals, and the Class 3 Failure Breakpoint."

The dataset supports three research questions:
- RQ1: Can half-yearly assessment signals predict year-end at-risk status (GPA < 2.0)?
- RQ2: What proportion of at-risk students are Single-Subject Failures (SSF), and which subjects drive that pattern?
- RQ3: How persistent is at-risk status year-on-year across matched student cohorts?

Key statistics:
- 4,632 annual student records across 9 schools
- 4,470 HY-matched records (model input)
- 22.1% pooled at-risk rate (GPA < 2.0)
- 38.3% Single-Subject Failure rate among GPA=0 stu
- 58.9% year-on-year at-risk persistence (matched pairs, 2024→2025)

Files included:
- combined_consolidated.csv — Primary dataset: 4,632 students × 363 columns (de-identified)
- tier_p_features.csv — Model input features: 4,470 matched students × 29 columns
- teachers_consolidated.csv — Teacher roster: 428 teachers across 9 schools × 19 columns
- 9 per-school raw CSVs (aghs, bbs, btri, ohs, sgphs, sjis, susc, uds, victoria)
- CODEBOOK.md — Full variable definitions
- README.md — Dataset overview and usage notes                                                                                
Student names have been replaced with SHA-256-derived pseudonymous identifiers. Teacher names are retained as publicly available information.