POW PDF forensics

Senate-provided HGAB POWs · parsing and title-match status

These figures describe the Senate-provided HGAB POW corpus and its separate first-page title scan. Parsing counts, scanned-file counts, and matched-project rows use different units.

Senate-provided HGAB POW files · parsed results

One row per parsed Senate HGAB POW file. The title link records the parser’s current match status; it does not verify the extracted bill rows or the project scope. Proposal PDF metadata and image-forensics above remain a separate corpus.

Files with no accepted title match remain listed by HGAB file code and are not assigned to project rows. Direct Senate file links and storage paths are not published.

What this investigation can establish

PDF properties such as /Creator and /Producer name software recorded when a PDF was made or last finalized. They may identify a scanner application or device, but they may also identify Excel, a PDF printer, an editor, or a merging service. A single document-level value applies to the whole PDF and cannot tell us whether its pages came from one scanner.

Known processing caveat: signatures were covered using AI before these PDFs were provided. That image editing and any subsequent PDF save or merge may replace the original Creator/Producer values, rewrite image streams, or flatten page content. The findings below describe the supplied, signature-covered copies. Treat every device or software label as a lead about the file’s processing history, not necessarily its original scan source.

Current finding: the corpus contains a mix of office software, PDF tools, and scanner/copier model strings. These are useful leads for selecting documents for manual review. The available structural signals do not prove that different scanners were used, and matching metadata does not prove a single scanner.

Downloader coverage in the mounted metadata catalog

1. Document-level software and device tags

Creator and Producer are reported separately because they do not always name the same software. A device-like label can be a useful lead, while an office-app or PDF-library label often reflects how the final file was exported or saved. These strings are not independently verified device identities.

2. Embedded image and structure clues

Image dimensions, bit depth, color-space declarations, and compression filters describe image objects stored inside the supplied PDFs. Signature masking can introduce image objects, change compression, or flatten pages; multiple profiles can also arise from different pages, page elements, resampling, or PDF conversion. They are not a device fingerprint. The indexed page orientation and extraction fields can identify documents worth inspecting, but they describe page layout and the extraction process rather than scanner provenance.

EXIF/XMP caution: 39 supplied PDFs contain the specific JPEG APP1 EXIF marker checked here. This does not establish that usable scanner model or device-ID metadata is present. The XMP count is a structural metadata-object marker, not confirmation of scanner details inside each page image. Signature masking may have removed or rewritten earlier image metadata.

3. Do shared metadata tags track a shared PDF workflow?

A Creator or Producer value repeated across unrelated projects and engineering offices is a useful clustering lead. It could reflect a shared contractor, consultant, scanner fleet, office workflow, PDF conversion service, or the later AI signature-masking step. Metadata alone cannot distinguish these explanations or identify the party that prepared a POW.

To help improve the parser, the table compares each repeated metadata value with extraction outcomes: how many documents have item rows, how many category-summary rows were read, and how many pages were flagged for manual review. A consistent pattern can show which document families deserve parser templates or focused review. Counts are descriptive; offices and project types may differ.

iLovePDF Producer tag: a consolidation lead

Interpretation: a repeated tag across offices supports a shared production or processing workflow hypothesis, not a contractor attribution. To test contractor origin, this would need corroboration from original unmasked PDFs, procurement/contractor records, and repeatable page-layout or image fingerprints. A tag produced by the masking tool would cluster the post-processing, not necessarily the source documents.

4. Producer and Creator tags by DEO

This groups the supplied POW PDFs by the engineering office listed in the project metadata. The leading Producer is the most frequent tag on those PDFs; the tag mix shows how consistent it is across that office’s submissions. Creator is shown separately as another clue about how files were made. These are observed document tags, not verified software inventories for the DEO: a shared contractor, PDF service, or later signature-masking step can create the same pattern.

5. Search the POW inventory

Filter by project, office, metadata tag, parser outcome, or structural signal. The office profile column shows the most frequent Producer tag for that project’s listed engineering office. Expand a row to inspect indexed page details and image-object profiles. “Mixed extraction method” means the index used more than one text/OCR route across pages; it is not a scanner-count signal.