A review packet needs context, not more attachments
For a synthetic review, keep three kinds of output together: the page someone can look at, the strings they can recalculate, and the explicit assumptions used in the calculation. A screenshot alone is awkward to recompute; a bare JSON file makes it hard to see where the figures came from.
The bundled PDF and Markdown are separately authored synthetic fixtures, not a claim of automatic extraction from one into the other. In a real assignment, the analyst must verify that the selected page actually supports the table and record the correspondence.
Build the packet from the three project directories
Run each command in its named repository. Use the PDF project's virtual-environment interpreter for the first command. The out folders remain separate; copy the intended artifacts to your review folder rather than assuming a shared working directory.
# pdf-page-images
python pdf_page_images.py examples/statement.pdf --pages 1,3 --dpi 144 --format png --output-dir out
# md-table-to-csv
python md_table_csv.py examples/table-sample.md --table 2 --output out/cash-extract.csv
# cashflow-bridge-check
python cashflow_bridge.py examples/reconciled.json --json > result.jsonInclude a tiny manifest
review/
statement-page-001.png
statement-page-003.png
cash-extract.csv
reconciled.json
result.json
readme.txt # source filenames, page positions, units, mapping, toleranceFor the supplied cash example, result.json should show calculated closing 103 and difference 0. Record whether you used tolerance 0 or the default 0.01; “passed” without a threshold hides a meaningful assumption.
Keep original page numbers in image filenames and distinguish them from printed labels. Do not call the packet verified evidence: a page image does not authenticate its source and the calculator checks addition only. Permission to share source documents is a separate question. See Page Image Notes for rendering details; the packet itself has no automatic provenance guarantee.
When a review requires edits or annotations before the images are made, use the related KomoPDF editing workflow as a separate preparation step. Keep the unedited source and identify the reviewed version in the manifest. The page-image CLI is only a renderer, and an annotated export does not authenticate the original source.
← Field notes