Appendix

9.1 Glossary

• Alert: an automatic message sent to the team when a pipeline run fails (section 5.5).
• Accuracy: how close an AI answer is to the expected result, scored 0–100 by the AI judge (section 7.3).
• AI judge: a model that scores AI answers against the expected result, using a fixed judge prompt.
• Advisory / enforce mode: whether the AI standards review only comments (advisory) or also fails the job on a Blocker (enforce) (section 11.6).
• Blocker: a finding that breaks a published **Rule:** block; the only finding that can fail a pipeline in enforce mode (section 11.4).
• Breaking point: the load at which the system stops meeting its targets (section 6.1).
• Checksum: a hash of a row’s or a table’s values, used to prove the data didn’t change between source and target (section 5.3).
• Consistency: how often an AI model gives the same answer to the same prompt (section 7.2).
• Expected result: what must be true for a test to pass.
• Key points: the facts an AI answer must contain to be correct.
• Load test: checks the system meets its targets at expected peak traffic (section 6.1).
• Lineage: being able to tell which pipeline run produced which rows, and from what source (section 5.3).
• Locator: how a UI test finds an element on the page (section 4.4).
• Pipeline run: one execution of a data pipeline, made of tasks run in order (section 5.1).
• Run sheet: the Excel sheet where a manual test run is recorded, with expected results filled in before testing (section 2.6).
• Precondition: what must be set up before a test can run.
• Stress test: increases load past the expected peak to find the breaking point (section 6.1).
• Soak test: normal load for several hours, to find slow problems like memory growth (section 6.1).
• Quality gate: a pipeline step that must pass before the next stage can start (section 10.2).
• remote_safe: pytest marker for a read-only HTTP check that is safe to run against a deployed environment (section 10.3).
• Service connection: the Azure DevOps credential a pipeline uses to reach Azure, e.g. to read a function key (section 10.5).
• Smoke Advisor: a pipeline step that gives an AI skill the smoke results and logs, and returns recommended fixes without changing pass/fail (section 11.2).
• Smoke test: a small, fast, read-only set of checks run after every deploy to prove the app is alive (section 10).
• Task: one step of a data pipeline, e.g. extract, transform, load (section 5.1).
• Unique key: the column (or columns) that identify one row, needed for uniqueness, per-row checksums, and lookups (section 5.3).
• Test case: one behavior to check, stored in Kiwi.
• Test plan: a group of test cases for a feature or release.
• Test run: one round of executing test cases on one build.

9.2 Change log

9.3 Acknowledgments

Maintained by the QA Engineer team. Format follows the Product Engineering standard.