TestSheet — pytest for Excel

Capture a golden-master baseline from your Excel model, then automatically flag any cell whose value or formula drifts in future versions.

TestSheet is a command-line regression-testing tool for spreadsheet models. It works the same way pytest works for code: you record what should be true, then run the suite on every change to verify nothing broke.

Features

• Golden-master baseline — snapshot every cell's value and formula in one command
• Drift detection — six drift kinds: value_only, formula_only, both, new, deleted, error_introduced
• Invariant rules — six built-in rule types you describe in a YAML file (range bounds, error checks, totals tie-outs, monotonic sequences, cross-cell relationships, no-hardcode guards)
• Multiple reporters — rich console table, JSON, JUnit XML (for CI), standalone HTML
• Structural-change heuristic — warns when too many cells drift at once (likely a layout change, not real model drift)
• Configurable float tolerance — set rel_tol / abs_tol in a config file
• GitHub Actions integration — single composite action, reports uploaded as artifacts

Installation

pip install testsheet

Dependencies: Python 3.10+, openpyxl, typer, rich, pyyaml, jinja2, formulas

5-minute quickstart

1. Capture a baseline

Navigate to the folder containing your workbook and run:

testsheet baseline models/q4_forecast.xlsx

This creates .testsheet/baseline.json next to your workbook. Commit this file to version control.

models/
  q4_forecast.xlsx
  .testsheet/
    baseline.json        ← commit this

2. Run the checks

After any edit to the workbook:

testsheet run models/q4_forecast.xlsx

TestSheet prints a table of drifted cells and exits 0 (pass) or 1 (fail).

TestSheet — running checks on q4_forecast.xlsx
┌──────────┬─────────┬───────────────────┬───────────┬───────────┐
│ Sheet    │ Address │ Kind              │ Baseline  │ Current   │
├──────────┼─────────┼───────────────────┼───────────┼───────────┤
│ Summary  │ B2      │ value_only        │ 460.0     │ 999.0     │
│ Model    │ D2      │ error_introduced  │ 120       │ #DIV/0!   │
└──────────┴─────────┴───────────────────┴───────────┴───────────┘
FAIL — 2 cells drifted

3. Add invariant rules (optional)

Create .testsheet/rules.yaml:

rules:
  - id: revenues_positive
    type: range_bound
    range: "Model!A1:A12"
    min: 0

  - id: no_formula_errors
    type: no_error
    range: "Model!A1:Z200"

  - id: total_ties_sum
    type: totals_tie
    total_cell: "Summary!B10"
    sum_range:  "Summary!B1:B9"
    tolerance: 0.01

  - id: margins_increasing
    type: monotonic
    range: "Model!C1:C4"
    direction: increasing

  - id: no_hardcoded_drivers
    type: no_hardcode_in_range
    range: "Assumptions!B1:B20"

  - id: gross_profit_nonneg
    type: relationship
    expression: "Summary!B5 >= 0"

Rules are evaluated on every testsheet run. A rule failure counts the same as cell drift for the exit code.

CLI reference

testsheet baseline <workbook>

Captures a golden-master baseline.

testsheet run <workbook>

Runs regression checks against the baseline.

Rules reference

Configuration

Create .testsheet/config.yaml to override defaults:

diff:
  rel_tol: 1.0e-6   # relative float tolerance (default 1e-9)
  abs_tol: 1.0e-9   # absolute float tolerance (default 1e-12)

structural_change_threshold: 0.5  # warn when this fraction of cells drift (default 0.5)

Float tolerance — useful for models with rounding differences between Excel versions. Set rel_tol: 0.01 to ignore differences smaller than 1%.

Structural change threshold — when the fraction of drifted cells exceeds this value, TestSheet emits a warning that the workbook layout may have changed (rows/columns inserted, sheet renamed), rather than flagging hundreds of individual cell drifts.

GitHub Actions

Add to your workflow:

- name: Run TestSheet
  uses: ./action.yml         # or the published action path
  with:
    workbook: models/q4_forecast.xlsx
    junit-output: testsheet-report.xml
    html-output:  testsheet-report.html

- name: Publish test results
  uses: dorny/test-reporter@v1
  if: always()
  with:
    name: TestSheet
    path: testsheet-report.xml
    reporter: java-junit

Full example workflow (.github/workflows/testsheet.yml):

name: Excel regression gate

on:
  pull_request:
    paths:
      - "models/**"

jobs:
  testsheet:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Run TestSheet
        uses: ./
        with:
          workbook: models/q4_forecast.xlsx

      - name: Publish JUnit results
        uses: dorny/test-reporter@v1
        if: always()
        with:
          name: TestSheet
          path: testsheet-report.xml
          reporter: java-junit

How it works

1. baseline — loads the workbook twice (once with data_only=False for formulas, once with data_only=True for cached values), then runs the formulas pure-Python evaluator to recompute values from scratch. Writes .testsheet/baseline.json.

2. run — repeats the same parse + evaluate step on the current workbook, then diffs cell-by-cell against the baseline. Classifies each changed cell into one of six drift kinds. Evaluates any rules.yaml invariants. Exits non-zero if any drift or rule failure is found.

3. Evaluator — TestSheet uses the formulas pure-Python Excel evaluator (no LibreOffice required). A LibreOffice headless evaluator is also available via --evaluator libreoffice for models that use functions not yet supported by formulas.

Project layout

testsheet/
├── src/testsheet/
│   ├── cli.py              # Typer CLI (baseline + run)
│   ├── parser.py           # openpyxl workbook parser
│   ├── baseline.py         # golden-master capture + load
│   ├── diff.py             # drift detection + structural-change heuristic
│   ├── config.py           # .testsheet/config.yaml loader
│   ├── evaluator/
│   │   ├── formulas_eval.py   # formulas library evaluator (default)
│   │   ├── pycel_eval.py      # pycel evaluator (legacy / Python ≤3.11)
│   │   └── libreoffice.py     # LibreOffice headless evaluator
│   ├── rules/
│   │   └── engine.py       # rules.yaml loader + 6 rule handlers
│   └── reporters/
│       ├── console.py      # rich terminal table
│       ├── json_reporter.py
│       ├── junit.py        # JUnit XML
│       └── html_reporter.py   # standalone HTML
├── tests/
│   ├── test_smoke.py
│   ├── test_evaluators.py
│   ├── test_diff_m2.py
│   ├── test_rules_m3.py
│   ├── test_rules_e2e.py
│   ├── test_reporters_m4.py
│   ├── test_ci_m4.py
│   └── test_hardening_m5.py
├── action.yml              # GitHub Actions composite action
└── pyproject.toml

License

MIT