Transparent methodology
Methodology
How reports become statistics — every step explained
Report → Privacy → Normalize → Duplicate → Moderate → Aggregate → PublishNormalization
Dept/service from DB taxonomy, BBS district codes, BDT integer — no UI hard-code.
Duplicates
Deterministic + fuzzy + embedding — no auto-delete, moderator Merge/Link.
Verification levels
PUBLISHED_UNVERIFIEDClaim, PII-cleared
EVIDENCE_SUPPORTEDWith sanitized evidence
MULTIPLE_REPORTSPattern across independent reports
Never “proven corruption” without verification.
Statistics rules
| Rule | Detail |
|---|---|
| Source | PUBLISHED + verified statuses |
| n≥30 | Else “Insufficient data”, no ranking |
| k≥5 | Geo suppression |
| Median | Over mean, outliers documented |
| Cache | cron → snapshots → s-maxage 60 |
ভিত্তি: N রিপোর্ট
incidentDate, not createdAt
Map
District polygons, filters, never exact pins
Index
Reported Irregularity Index — stub
Dataset
CSV/JSON, PII stripped, /data
Limitation: self-reported, selection bias — don’t treat one report as truth, look for patterns.