BDPolicyLab analysis in the twin. Part of what the twin can answer. Every figure below is the value the route serves, with its source; the raw response is at /api/twin/absorb/products.
Descriptive, sourced: no model
Basis: BPL products linked to Body or Place by exact name; counts of items, no model.
2 items.
Narrativeরোহিঙ্গা: প্রত্যাবাসনের মৃত্যু, হিসাবের জন্ম
content/narratives/2026-05-03-rohingya-niye-hisab
Ministry of Foreign Affairs: the text names it (publication_bindings: exact_name_bn+bn_case_suffix 'পররাষ্ট্র মন্ত্রণালয়', paragraph 41)
NGO Affairs Bureau: the text names it (publication_bindings: exact_name_en 'NGO Affairs Bureau', paragraph 49)
Narrativeহরমুজ যদি বন্ধ হয়, বোরো কত খাবে
content/narratives/2026-05-04-hormuz-jodi-bondho
Bangladesh Bank: the text names it (publication_bindings: exact_name_bn+bn_case_suffix 'বাংলাদেশ ব্যাংক', paragraph 9)
Department of Agricultural Extension: the text names it (publication_bindings: exact_name_bn+bn_case_suffix 'কৃষি সম্প্রসারণ অধিদপ্তর', paragraph 17)
Export Promotion Bureau: the text names it (publication_bindings: exact_name_bn 'রপ্তানি উন্নয়ন ব্যুরো', paragraph 21)
Finance Division: the text names it (publication_bindings: exact_name_bn 'অর্থ বিভাগ', paragraph 38)
Ministry of Commerce: the text names it (publication_bindings: exact_name_bn 'বাণিজ্য মন্ত্রণালয়', paragraph 40)
Ministry of Foreign Affairs: the text names it (publication_bindings: exact_name_bn 'পররাষ্ট্র মন্ত্রণালয়', paragraph 41)
Petrobangla: the text names it (publication_bindings: exact_name_bn 'পেট্রোবাংলা', paragraph 19)
72 analyzer modules, each run once offline into a pinned snapshot built 2026-10-02T07:21:38Z; results are not validated.
Kinds: `pub_module` (app/publications/pubs), `publication` (the publication store, app.publications in data/app.duckdb), `narrative` (content/narratives), `story` (the stories table in bdpolicy.db) and `analyzer` (app/analysis). Each item links to Body or Place objects only where a name matches exactly, and every link says how (`via`): - `tag`: a frontmatter or store tag, hyphens and underscores read as spaces, equals the full name_en of a national Body or the title_en of a division or district Place (case-insensitive, whole string). Acronyms and lower tiers are not matched from tags: tag `sob` means state-owned banks in a banking narrative, not Survey of Bangladesh (SoB), and a field office named "Forest" or a union named "Benapole" is a homonym of the topic. Product (DGDA medicine brands) is not matched: brands include common words (Energy, Macro, Forest). - `text_binding`: a narrative's text names the Body by its exact registry name, as bound by scripts/build_publication_bindings.py (publication_bindings.parquet, the file /api/twin/entity/{id}/evidence serves); reused here, not re-derived. - `series_source`: the item reads a series (its analyzers' required_series, or a series name printed in its own code) whose `data_series.source` names a Body. The source is compared whole, split on " / ", and with one trailing parenthetical taken as an alternative name, each part compared whole against Body name_en or acronym. A key matching more than one object is ambiguous and dropped, never resolved by guess. The index is built lazily on first use and cached; nothing here writes. Lane BRN7B: nothing here reads bdpolicy.db or runs an analyzer at request time. The series-to-source map, the stories and every analyzer's stored answer come from two manifest-pinned parquets (bpl_series_source, bpl_analyzer_snapshot; root lake_bpl_absorb) that scripts/twin_bpl_absorb/build_analyzer_snapshot.py builds offline. A missing file, or one whose sha256 is not its pin, raises SnapshotUnavailable (the routes answer 503).