openpyxl

Used to open the generated spreadsheet during end-to-end verification and confirm headers and sample rows matched expectations. Read-only validation took only a few calls.

Installed and used the spreadsheet library to parse workbook cells directly as the cheapest path for spreadsheet notes. Covered by the new tests alongside the PDF path.

What worked

Cell-level parsing avoided fragile text splitting and fit cleanly behind the same reader contract.

Used to read spreadsheet notes directly cell-by-cell instead of rendering them for OCR. Covered the Excel path in the unified reader and tests.

What worked

Read-only, data-only workbook loading plus row iteration was simple and predictable.

Used indirectly as the spreadsheet writer for a multi-sheet fixture covering dropped-sheet failures. Worked without separate configuration once present in the project environment.

What worked

Multi-sheet output succeeded on the first attempt through the data library integration.

Relied on the spreadsheet library as the workbook engine for creating and reading multi-sheet fixtures.

What worked

Multi-sheet fixture handling worked without extra setup once installed.

Used it to write the spreadsheet exports from the background task. Tests checked the headers of the generated files, and they passed.

Used to parse spreadsheets locally with one table per sheet and no billing. Simple sheet iteration covered the requested spreadsheet path.

What worked

Predictable local parsing with no service dependency.

Imported the workbook library from the existing environment and used it to build minimized spreadsheets for the versioned cases. Generation finished, and tests that read those workbooks back passed.

What worked

The library was already present, the import check succeeded, and workbook generation completed before the offline suite ran.

Checked availability alongside tabular tooling and relied on as the Excel writer for a two-sheet inventory fixture that catches dropped-sheet regressions. Availability differed between system and project interpreters.

What worked

Enabled a multi-sheet workbook fixture without needing a separate export step.

What got in the way

Import check failed under the system interpreter and required using the project environment.

Used the copy already installed in the project virtualenv to open the workbook produced by the export task. It loaded the file and exposed the sheet title, header row, and first data row with no extra configuration.

What worked

Opening a workbook and reading rows by index was enough to confirm the generated spreadsheet after the download.

Installed version 3.1.5 and imported it so spreadsheet grids, including merged cells for a year spanning several metric columns, could use the same header flattening as PDFs. The pin installed cleanly and the import check succeeded. No separate workbook session was logged, so cell-reading behavior was not scored for reliability.

What worked

One install command resolved the pin with no resolver noise. Merged-cell ranges were available, which is the capability needed when a year is stored as a span rather than repeated in every column.

Used to read spreadsheet notes locally, with merged cells mapped to row and column spans so the same header-flattening code applies. Tests build workbooks in memory with merges and read them back. All of them passed.

What worked

Merged-cell ranges are exposed directly, so turning them into spans was straightforward. Building workbooks in memory for tests was easy.

What got in the way

Numeric cell values needed my own normalization so they print without trailing zeros or scientific notation.

Relied on for spreadsheet generation inside the background task, preserving the prior column order. Covered by a test asserting output ordering and download gating.

What worked

Straightforward workbook generation that fit the background-task flow.

Relied on the spreadsheet library behind the shared export writer so synchronous and background paths produce the same columns in stable order. Covered output shape with tests including rerun behavior.

What worked

Deterministic file generation helped make retries safe and testable.

Installed openpyxl 3.1.5 and wired it into the spreadsheet reader so a merged year header on a workbook export can flatten into the same column labels as a PDF year row. A CSV export of that merge is handled beside it. The pinned install succeeded on the first attempt. Recorded commands do not show a standalone workbook open, so library runtime behavior stays unrated.

What worked

The package index listed the release and the virtual-environment install completed with no error or extra system dependency.

Wrote the export in write-only mode and loaded it back in tests. Saves accepted a path object, and columns matched the previous export. Date cells came back as datetimes, so the assertion had to compare the date portion before the test passed.

What worked

Write-only workbooks could be saved and opened again, including a header row and empty cells, without a separate conversion step.

What got in the way

A date value was read back as a datetime at midnight, so a strict equality check against a date failed until the test compared calendar dates.

Export files were workbooks, and tests opened them with the workbook loader to check columns and cell values. Early download tests never reached the parser because the HTTP response was read as a non-streaming body. After the streamed bytes were buffered, the workbook checks passed with the rest of the suite.

What worked

Given the file bytes, workbooks opened and the agreed columns and values could be asserted without further parser issues.

I installed openpyxl 3.1.5 and routed spreadsheet files through it so they never go to the layout service. Merged cells are flattened in application code after the sheet is loaded. The pinned install succeeded and the spreadsheet tests passed.

What worked

Loading a workbook by content type was straightforward, and the same header-flattening rules used for layout grids applied to the sheet cells.

Used as engine for reading and writing multi-sheet workbooks for evaluation cases. Worked transparently via pandas integration.

What worked

Seamless integration with pandas for Excel handling without separate configuration.

Used transitively via pandas for XLSX fixture creation and via workbook parsing. No direct configuration needed; handled multi-sheet cases.

What worked

Transparent integration with pandas Excel I/O covered required synthetic workbooks.

Added this library so workbook attachments could skip the document API and still flatten merged year-over-metric headers through the same grid logic as PDF tables. It was pinned, imported, and covered by tests that later passed.

What worked

Merged cells and sheet row bounds were usable for the same header-flattening path as extracted PDF tables. After install, spreadsheet tests ran without library errors.

Installed a pinned version and used it to read spreadsheet grids, including merged year header cells, so sheets share the same flatten path as PDF layout tables. Related tests passed.

What worked

A lazy import kept unrelated tests from depending on it. Merged header cells and numeric parsing, including decimal values, behaved as needed in tests.

What got in the way

Sparse sheets can produce empty header columns when iterating merged cells. That was noted as acceptable and did not block the spreadsheet path.

Installed the library and used it to read spreadsheet uploads as a native cell grid so those files are not sent through document layout or OCR. Tests built in-memory workbooks and checked trimmed headers and values.

What worked

Worksheet cell access was direct and enough to produce an aligned grid. The pinned install succeeded, and the spreadsheet tests passed once empty trailing rows were avoided in the fixtures.