What is the question bank based on?
The public Red Seal examination breakdown for the Refrigeration and Air Conditioning Mechanic. Every question is tagged with an official sub-task code (for example F-16.01, troubleshoots HVAC/R systems), and diagnostics draw questions in proportion to the official area weights. The product says it is aligned to the public occupational breakdown. It never claims to match the real exam.
Which sources are allowed?
- The public Red Seal occupational standard and exam breakdown
- Public Skilled Trades Ontario guidance
- Federal and provincial regulations, cited by section
- Public safety guidance such as CCOHS
- Manufacturer documentation that permits citation
- Original explanations of standard practice, labelled as such until reviewed
Not allowed: recalled or "actual" exam questions, copied competitor questions, unlicensed reproduction of CSA standards or paid codebooks, and technical claims generated by AI without an approved source.
How does review work?
Every item moves through draft → technical review → editorial review → approved → published. Publication needs a named reviewer, a review date, a source and locator, a rationale for the correct answer and for every wrong option, an ambiguity check and a plain-language check. The pack is versioned; an attempt is locked to the version it started on, so a later correction never changes a result you already have.
How are scores calculated?
- Multiple choice: correct is taken only from the answer key. Area scores are correct out of answered, always shown with both numbers and the content version.
- An area with fewer than three answered questions shows "not enough evidence" instead of a score.
- No speed bonus. Time is shown for your own comparison.
- No pass prediction, no percentile, no probability.
How does the voice lab use AI?
Speech is transcribed, and a language model maps the transcript to a fixed checklist written by a person. For each checklist item the model must quote the part of your answer that covers it. Our server then checks the output: unknown checklist items or extra fields are rejected, quotes that do not appear in your transcript are thrown out, and a match the model is unsure of is shown as "unclear" and earns no credit. The server, not the model, computes the score.
AI is never used to write questions for sale, decide what the correct answer is, estimate your chance of passing, or change a published item.
How is "next up" chosen?
By a fixed rule, in order: finish a session in progress; take the diagnostic; clear review items that are due; otherwise practise the area where your miss rate times the exam weight is highest. Review items are ordered by how often you missed them, the area's exam weight, and how recently you saw them.