Legal

AI & Data Disclosure

Effective September 11, 2026 · How Stubby's numbers are made

Plain answer. The math that produces your cheque estimate is ordinary, deterministic arithmetic. No AI generates your numbers. AI tools were used to help transcribe collective agreements into Stubby's data files, and a machine-learning text reader runs on your device when you scan a stub. Both are checked, and both can be wrong. This page tells you where.

1. The calculation engine: no AI

Gross pay, overtime, premiums, vacation pay, union dues, CPP, EI and income tax are computed by fixed formulas: the rules written in each collective agreement and the Canada Revenue Agency's T4127 payroll-deduction formulas (Option 1), using the published 2025 and 2026 federal and provincial tables. The same inputs always give the same output. The engine has been reconciled to the penny against real pay stubs for the agreements marked "Verified" in the app.

2. Agreement data: AI-assisted transcription, human-checked

Stubby ships structured data for more than twenty Ontario building-trades collective agreements (wage ladders, effective dates, overtime rules, dues and fund contributions). Turning several hundred pages of PDF into that data was done with the help of large-language-model tools, followed by human review against the source document and, where possible, against real pay stubs.

Every agreement carries a verification status that you can see in the app's Setup screen and on each public rate card:

Status	What it means
Verified against a real stub	The engine reproduced an actual pay stub for this agreement line by line.
Rates loaded from a rate sheet	Wage rates were taken from the union's published rate sheet; overtime and other rules are from the agreement text but not yet stub-verified.
Transcribed from the published agreement	Everything comes from a reading of the agreement PDF. Treat as a well-informed draft until verified.

Transcription errors are possible, especially for unusual classifications, shift premiums, and rules that differ by local. Agreements are also amended; Stubby shows the effective and expiry dates it is working from. If a figure looks wrong, check it against the agreement or your local, and please tell us (section 5).

3. The stub scanner: on-device machine learning

The scanner uses optical character recognition (Tesseract, an open-source machine-learning text reader) to pull rates, hours and deductions from a photo or PDF of your stub. It runs entirely in your browser; your stub is never uploaded. Recognition is imperfect: blurry photos, unusual layouts and handwriting produce mistakes. Everything the scanner extracts is shown to you for confirmation before it is used, and you should check each field.

4. No AI advice, no profiling

Stubby contains no chatbot, no generated advice and no model that learns from your data. Your inputs are not used to train anything. Nothing on this site is personalised by an algorithm.

5. Reporting an error

If Stubby's figure differs from your stub or from your agreement, the "Check" screen in the app helps you find the line that is off, and the "Send feedback" button in Setup emails us the details without your pay figures. You can also write to hello@getstubby.ca. Corrections to agreement data are published in app updates and noted in the app's release notes.

6. Not advice

Nothing produced by Stubby, including agreement data and scanned values, is payroll, tax, legal or financial advice. See the Terms of Service.