Government publications by agency. 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/brain/govfull/agency/bmd.
Descriptive, sourced: no model
Basis: sourced records (URL, sha256 or locator per row), no model.
stored files, documents and chunks, and counts over them
| Upload year | Files | Bytes |
|---|---|---|
| 2016 | Files: 1 | Bytes: 1,533,665 |
| 2017 | Files: 1 | Bytes: 6,163,480 |
| Verdict | PDFs |
|---|---|
| Text | PDFs: 3 |
| Manifest line | Type | Bytes | Characters | Upload year | Upload month | Also in the office store | Text layer | Pages | Re-hash verified |
|---|---|---|---|---|---|---|---|---|---|
| 380 | Type: pdf |
| Pages | Chunks |
|---|---|
| 19 | Chunks: 18 |
Not held: Printed domain, Page, Line, Printed kind, Page, Line, Printed date, Page, Line, Printed memo, Page, Line.
| First page | Last page | Characters | Lines | Numeric tokens | Bangla share |
|---|---|---|---|---|---|
| 1 | Last page: 1 | Characters: 4,574 | Lines: 879 |
| Source | Publisher | Unit | Caveat |
|---|---|---|---|
| Government untruncated corpus harvest (stored copies, gap row A3) | Publisher: Issuing ministries, departments, divisions and boards of the Government of Bangladesh (stored copies) | Unit: one row per stored PDF document: manifest line, URL, sha256 at fetch, bytes, chars, issuing agency, portal host, upload year/month, registry entity binding, overlap with the stored files from government portals; source file is relative to the corpus root | Caveat: All 4,049 documents are full PDF publications; 241 match SHA-256 hashes of the stored files from government portals (BRN46). Agency assignment is verified against exact registry entities. sha256 verified is set only for re-hashed files (the 400 deterministic random sample and the pilot); the rest carry the harvest sha256. |
phones_detected), never stored.| Files: 1 |
| Bytes: 16,169,262 |
| 2025 | Files: 1 | Bytes: 29,708,558 |
| 2026 | Files: 1 | Bytes: 7,806,407 |
| Characters: 97,166 |
| Upload year: 2026 |
| Upload month: 5 |
| Also in the office store: no |
| Text layer: not available |
| Pages: not available |
| Re-hash verified: not available |
| 1,853 | Type: pdf | Bytes: 29,708,558 | Characters: 167,689 | Upload year: 2025 | Upload month: 11 | Also in the office store: no | Text layer: Text | Pages: 130 | Re-hash verified: yes |
| 2,240 | Type: pdf | Bytes: 16,169,262 | Characters: 109,205 | Upload year: 2024 | Upload month: 2 | Also in the office store: no | Text layer: not available | Pages: not available | Re-hash verified: not available |
| 3,372 | Type: pdf | Bytes: 6,163,480 | Characters: 89,603 | Upload year: 2017 | Upload month: 4 | Also in the office store: no | Text layer: Text | Pages: 51 | Re-hash verified: yes |
| 3,758 | Type: pdf | Bytes: 1,533,665 | Characters: 117,706 | Upload year: 2016 | Upload month: 8 | Also in the office store: no | Text layer: Text | Pages: 19 | Re-hash verified: yes |
| Bangla share: 0% |
| 2 | Last page: 2 | Characters: 4,580 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 3 | Last page: 3 | Characters: 4,576 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 4 | Last page: 4 | Characters: 4,580 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 5 | Last page: 5 | Characters: 4,580 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 6 | Last page: 6 | Characters: 4,580 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 7 | Last page: 7 | Characters: 4,573 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 8 | Last page: 8 | Characters: 4,581 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 9 | Last page: 9 | Characters: 4,571 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 10 | Last page: 10 | Characters: 4,581 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 11 | Last page: 11 | Characters: 4,580 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 12 | Last page: 12 | Characters: 4,581 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 13 | Last page: 13 | Characters: 4,575 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 14 | Last page: 14 | Characters: 4,581 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 15 | Last page: 15 | Characters: 4,581 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 16 | Last page: 16 | Characters: 4,576 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 17 | Last page: 17 | Characters: 4,573 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| 18 | Last page: 18 | Characters: 4,573 | Lines: 879 | Numeric tokens: 818 | Bangla share: 0% |
| Per-issuing-agency manifest datasets | Publisher: Built from the rows above by the inventory pipeline | Unit: one row per issuing agency, with its partition file, sha256, file counts, and text layer counts | Caveat: Each agency's inventory partition is the resumable unit of the pipeline; a rerun rewrites only partitions whose rows changed. |
| 200-chunk pilot: documents and chunks with printed fields across 14 agencies | Publisher: each document's stored PDF (PyMuPDF text layer) | Unit: one row per pilot document (fields with page and line) and per chunk (page range, counts, text sha256) | Caveat: Pilot documents were chosen to maximize agency diversity across text-layer candidates, chunked in consecutive page blocks up to 4,000 characters, hitting exactly 200 chunks total. Documents with personal data (CVs, data sheets, rosters, beneficiary lists, or >=2 mobile phone numbers on pages 1-2) are screened out before chunking. Phone numbers are never stored in any parquet or served payload. |
| Pipeline run log (measured) | Publisher: Measured pipeline runs on a 16-core machine | Unit: one row per stage run: items, bytes, seconds, workers | Caveat: Runs over files a previous run had read hit the page cache and overstate the rate; the projection uses the cold runs named in its basis. |