What harvested documents does the twin hold a machine classification for, by corpus, document type, year and issuing body?

Document catalogue. 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/doc_catalogue.

Descriptive, sourced: no model

Basis: sourced records (URL, sha256 or locator per row), no model.

One row per sha256-verified triage row; issuing body, year and subject only through their first-page and privacy gates; overlaps with already-served document tables are flagged, not double-counted.

machine-classified (Qwen3.8-27B, first page)

Classified documents: 37,399
With an issuing body served: 11,398
With a year served: 25,969
With a model-written subject served: 33,300
Issuing bodies mapped to a registry body: 4,187
Also named in the object-documents table: 27,649

By source corpus

Documents by corpus, 37,399 in all

Corpus	Documents
Objectstorage	Documents: 26,760
Agy wave6	Documents: 7,643
Agy wave5	Documents: 2,230

By document type

The types the classifier assigns, 57 documents without one

Document type	Documents
Statistics or report	Documents: 5,224
Web page	Documents: 4,734
Office order	Documents:

By year printed on the first page

Documents by year of printing

Year	Documents
1,971	Documents: 19
1,972	Documents: 57
1,973	Documents: 14

Who issued them

The ten issuing bodies with the most documents

Issuing body, as printed on the first page	Documents
Bangabandhu Water Treatment Plant Khulna WASA	Documents: 773
BKMEA	Documents: 644
WASA, Pathorghata, Rupsha, Khulna-9240

What the gates counted

Documents also named in the object-documents table: 27,649
Documents the askdocs table also names: 303
Documents the second document layer also names: 7
Documents with no document type: 57

The catalogue

The first 200 classified documents of 37,399

Corpus	Document type	Issuing body, as printed	Year	Subject, model-written	Registry body
Agy wave6	Document type: Form	Issuing body, as printed: not available	Year: not available

Where these figures come from

The source collections

Source	How it was classified	Unit
Colab document triage: machine classification of harvested documents	How it was classified: Qwen3.8-27B bf16 (vLLM on Colab A100) from each document's first page; stored at `triage_colab_1003.jsonl`	Unit: one row per document: corpus, stored path, file sha256, `doc_type`, `issuing_body`, year, subject; every value machine-classified

Not held

Issuing bodies the triage did not mark verbatim: Served only where issuing_body_verbatim is true and the string prints on the document's first page; every other row carries issuing_body_reason.
Years outside 1971-2026 or not printed on the first page: Served only when the year (Gregorian, Bangla digits, or the Bangla-calendar equivalent) prints on the first page; else null with year_reason.
Subjects with phone numbers, personal identifiers or personal names: Every model-written subject is scanned; a hit drops the subject with subject_reason. List-type documents (beneficiary, merit, voter, payroll, roster lists) carry metadata only, never the summary.
Document text and excerpts: Never stored in the lake or served; rows carry the first-page character count and text sha256 so a reader re-extracts them.
Absolute server paths: source_file is relative to the archive disk root; production holds no copy of the disk.