Basis: sourced records (URL, sha256 or locator per row), no model.
scanned PDFs and parsed procurement documents
hed.gov.bd: procurement documents parsed
11
hed.gov.bd: procurement documents matched
11
hed.gov.bd: text-layer PDFs scanned
13
hed.gov.bd: procurement documents by kind
Kind
Documents
E tender notice
Documents: 9
Annual procurement plan
Documents: 1
Tender notice
Documents: 1
hed.gov.bd: procurement documents, first 50 of 11
Kind
Family printed
Tender number
Method
Printed date
E-GP mark
Upload folder
Pages
Annual procurement plan
Family printed: Annual Procurement Plan
Tender number: not available
Where these figures come from
The source collections
Source
Publisher
Unit
Caveat
Text-layer scan of the stored files from government portals (gap row A1)
Publisher: each document's stored PDF (PyMuPDF text layer, pages 1-3)
Unit: one row per text-layer PDF: general kind keyword, family match, personal screen, phone count; source file is relative to the archive disk root
Caveat: Only verdict=text PDFs are scanned; legacy-encoding, undecodable and image-only PDFs are counted in the remainder, never scanned. Kind keywords match Unicode Bangla in logical order or English; visual-order extracts miss, so an absent kind is not evidence of none.
Not held
Document text: Never stored in the lake or served; fields are located by page and line.
Person names and phone numbers: No name or phone field exists in either table; phones are counted at the build (phones_detected), never stored. Screened-out documents appear only as exclusion counts by reason.
A2 oracle_wave spec values: LLM-derived; used only as pointers to design the keyword patterns, never stored or served.
Non-text-layer PDFs and non-PDF files: Not scanned in this lane; counted in the remainder for later lanes.
Method:
LTM
Printed date: not available
E-GP mark: not available
Upload folder: 2024/12
Pages: 19
E tender notice
Family printed: e-GP
Tender number: not available
Method: not available
Printed date: 22.02.2022
E-GP mark: e-GP
Upload folder: 2024/12
Pages: 1
E tender notice
Family printed: e-Tender
Tender number: lice
Method: not available
Printed date: 07.04.2025
E-GP mark: e-GP
Upload folder: 2024/12
Pages: 1
E tender notice
Family printed: e-Tender
Tender number: tice No
Method: LTM
Printed date: 08.02.2022
E-GP mark: e-GP
Upload folder: 2024/12
Pages: 2
Tender notice
Family printed: Invitation for Tenders
Tender number: not available
Method: not available
Printed date: 09.02.2025
E-GP mark: not available
Upload folder: 2024/12
Pages: 2
E tender notice
Family printed: e-Tender
Tender number: not available
Method: not available
Printed date: 1.05.2424
E-GP mark: e-GP
Upload folder: 2026/4
Pages: 1
E tender notice
Family printed: e-Tender
Tender number: not available
Method: not available
Printed date: 17.05.2026
E-GP mark: e-Gp
Upload folder: 2026/4
Pages: 1
E tender notice
Family printed: e-Tender
Tender number: not available
Method: not available
Printed date: not available
E-GP mark: e-GP
Upload folder: 2026/4
Pages: 1
E tender notice
Family printed: e-Tender
Tender number: not available
Method: not available
Printed date: not available
E-GP mark: not available
Upload folder: 2026/4
Pages: 1
E tender notice
Family printed: e-Tender
Tender number: not available
Method: not available
Printed date: 24.05.2026
E-GP mark: e-GP
Upload folder: 2026/4
Pages: 1
E tender notice
Family printed: e-Tender
Tender number: not available
Method: not available
Printed date: 10.05.2026
E-GP mark: not available
Upload folder: 2026/4
Pages: 1
Not held: Memo, Package, Reference.
Procurement family: printed fields with page and line
Publisher: each matched document's stored PDF (full read)
Unit: one row per parsed procurement file: tender no, ref, memo, package no, tender id, method, first date and e-GP mark, each with page and line
Caveat: Family membership is keyword-based (tender, e-tender, annual procurement plan, award notice, IFT/RFP/REOI, e-GP marks) on pages 1-3; Bangla visual-order extracts under-match. Fields are first regex hits on pages 1-3, verbatim; an absent field means no hit, not none printed.
Whole-corpus inventory (reused BRN46 pin)
Publisher: The harvest manifest (stored copies)
Unit: one row per stored file; corpus totals are counted here from the pin
Caveat: Built by the inventory pipeline (lane BRN46); reused without re-check per the source rule for pinned tables already served.