Skip to main content
BPO

How we work

Transparency report

Corrections accepted, letters published, methodology changes, coverage gaps the desk has identified, and known limitations of the dataset. The Bangladesh Peace Observatory commits to publishing this report annually, with a rolling live counter updated as corrections move through the moderation queue.

Methodology changes

What the BPO has changed about how it codes

2026-05-21 · D1 migration. The Observatory moved from MongoDB Atlas M0 to Cloudflare D1. The lean incident schema dropped 52 audit fields not surfaced in any public lens. No row was dropped that satisfied isvalidate = 1 and ispublish = 1.

FTS5 search rebuild. Replaced the English-only porter stemmer with a Bangla-friendly tokenizer. Search recall on English terms reduced very slightly; Bangla-script queries now work.

Eventsource column drop. The lean schema dropped the eventsource foreign key. Source profile pages now match by URL host pattern, which is imperfect; the column will be restored in a future migration.

Actor refinement (2015–2018). Some pre-2015 incidents were coded against parent-party actors (AL, BNP) that the desk would today code against sub-actors (Chhatra League, Jubo Dal, etc.). Year-over-year actor comparisons across this boundary should account for the refinement.

Coverage gaps

What the BPO record is missing

  • Pre-2012 coverage is sparse — the source-newspaper net was narrower and the dataset only achieves stable per-month density from 2012 onward. Multi-year averages spanning 2011/2012 should be read as a coverage change, not a real surge.
  • SGBV is systematically under-coded versus the true underlying incidence. The BPO record is a partial sample useful for trend analysis, not for prevalence estimation. See the SGBV story for details.
  • Custodial deaths are coded when reported. Under-reporting is severe; treat counts as a lower bound.
  • Court outcomes are not tracked. The record contains arrest counts but does not currently follow cases through prosecution.
  • Casualty fill rates are 34% (killed), 25% (injured), 52% (arrested). The Methodology page has the full field-by-field table.

For editorial review

Editorial annotation pending

This page is auto-composed. The BPO research desk's prose commentary — named episodes, recent corrections of note, policy responses — is added per annual edition. Until that edition publishes, treat the structural numbers as authoritative and the framing as the data-only baseline.