Excel parity validation

Tiny DSA includes an exported validation bundle that checks the generated tiny_dsa package against the source Excel workbook. The test drives the workbook with Microsoft Excel through xlwings, applies the same inputs through the package’s public set_* functions, and compares the public compute_* outputs cell by cell.

Current reference result

The current reference run is PASS: 1,770 / 1,770 cell-level comparisons passed at atol = 1e-06.

• Result: PASS
• Generated: 2026-07-12T14:18:16+00:00
• Passed: 1,770
• Failed: 0
• Pass rate: 100.00%
• Acceptance bar: 100.00%

The sweep covers 1,770 cell-level comparisons against Excel.

What Was Tested

The validation checks the exported standalone library, not just the extraction graph. It imports tiny_dsa.api, creates a fresh context for each scenario, sets inputs through the records-shaped public setters, computes the exported output series, and compares those values against the workbook’s calculated output cells.

Inspect Or Re-run

The validation bundle is shipped in the source repository under tests/. Because the golden-master oracle uses Microsoft Excel through COM automation, reruns require Windows with Microsoft Excel installed.

• Reference parity report: tests/results/reference/parity_report.txt
• Validation bundle README: tests/README.md
• Differential test harness: tests/differential/differential_test_exported_library.py

To re-run the validation from the exported project:

uv run --project . --group validation python -m tests.differential.differential_test_exported_library --layout exported