Canon Overview
3 min read
Canon is the set of documents that define how your product is supposed to behave — PRDs, specs, standards, release exit criteria. Once a document is in Canon, every AI surface in Hawzu reads from it and cites it.
The name is the point. Canon is the authority a test case is judged against, and the promise is provenance rather than cleverness: a grounded answer names the document and section it came from.
Why Grounding Matters
Section titled “Why Grounding Matters”An AI model with no documents to read still produces confident test cases. It just writes them against how it assumes systems like yours behave.
Canon changes three things:
- Generated tests stop guessing. The model writes against your stated rules instead of a generic idea of what a login screen or a checkout flow does.
- Every claim carries a citation. A test case that cites §4.2 of a real spec can be checked in seconds. An uncited one has to be re-derived from scratch.
- Release calls use your criteria. Readiness is judged against the exit criteria you wrote down, not an industry benchmark the model supplied from memory.
Where Canon Is Used
Section titled “Where Canon Is Used”Canon is not a place you visit to read documents. It is a source that other features draw on.
| Surface | What Canon does there |
|---|---|
| AI Test Case Generator | Grounds both stages — the coverage map and the written test cases. Each draft carries citation chips naming its sources. |
| Release readiness | The AI summary judges the release against your documented exit criteria and cites them. |
| Atlas | The Discover and Find gaps proposals read Canon to draft a map of the product. |
Canon itself has one interactive surface: Ask Canon, which runs the same retrieval that grounds generation so you can see exactly what the model would receive.
Canon and Files Are Not the Same Thing
Section titled “Canon and Files Are Not the Same Thing”Canon does not store files. It points at files already in your project’s file library.
That has a few consequences worth knowing up front:
- Adding a document to Canon and uploading a file are the same underlying action — the file lands in the library either way.
- You can add a file that is already in the library without uploading it again, using Choose from files.
- Removing a document from Canon deletes its index. It does not delete the file.
- A file used by Canon counts as in use, so File Manager will not let it be deleted out from under the index.
The practical distinction: put a document in Canon when it states how the product should behave. Leave it in Files when it is evidence, an attachment, or reference material the AI has no business asserting from.
Reading the Canon Page
Section titled “Reading the Canon Page”Open a project and select Canon in the left navigation. Three tiles sit above the document list:
| Tile | What it tells you |
|---|---|
| Documents | How many documents are in Canon, and how many could not be read |
| Passages indexed | How much text retrieval can actually search. Documents present but zero passages means ingestion has stalled |
| Grounding now | How many documents are live in generation and release calls right now |
Grounding now is the number that matters. A document that is uploaded but still processing, or that failed to parse, is listed but grounds nothing.
Below the tiles sit the Ask Canon panel and the document table. Page-level actions — Add document, Retry failed documents — are in the action rail at the right edge.
A Canon document lives at one of two levels.
| Scope | Grounds | Managed from |
|---|---|---|
| Project | That project only | The project’s Canon page |
| Workspace | Every project in the workspace | The workspace Canon page |
A project’s AI surfaces read both its own documents and the workspace documents, because a tester asking about login behaviour wants the login spec and the company’s accessibility standard.
Because workspace documents change what the AI asserts everywhere, adding and editing them is restricted to workspace managers. See Scope and Permissions.