[01] Why deterministic
Different tools, different questions.
Several kinds of tool touch a pull request. Each answers a different question. This page compares Graphentra by category, not by vendor, because products within a category differ.
[02] By category
Seven questions to ask any tool.
Cells for other categories are hedged on purpose: “Usually”, “Varies”, “Needs history”. The Graphentra column is the target at launch, and each cell shows the status of the capability behind it.
| Question | AI code reviewers Comment on a pull request using a language model. | Test-impact tools Choose tests from coverage or test history. | Build-graph “affected” tools Choose packages from the module graph. | AI QA and UI-regression tools Write, replay or compare end-to-end tests. | Graphentra Target at launch |
|---|---|---|---|---|---|
| Works on day one, with no history or recordings | Usually | Needs history or coverage data | Usually | Needs recordings or a test suite | Starts from the repository and the diff. Built · prototype |
| Same diff, same answer | Varies. Model output can differ between runs | Often | Usually | Varies | Byte-identical evidence for the same inputs. Built · prototype |
| Symbol-level precision | Varies | File or method level | Package-level | Usually behaviour, not code | Named functions today. Wider coverage next. Built · prototype In development |
| Names affected screens and APIs | Varies | Usually test names or IDs | Package-level | Screens it has observed | Surfaces are part of the evidence. In development |
| Proof path per claim | Varies. Some cite code | Rarely | Usually a package-level reason | Usually a replay or a visual diff | Changed function, caller, component, route. In development |
| States what it could not trace | Varies. Often a general disclaimer | Rarely | Rarely | Sometimes | Listed as unknown, never dropped. In development |
| Guides manual QA | Rarely | Varies | Rarely | Varies | A ranked, plain-English checklist. Built · prototype |
The limit of proof. Proof applies to the paths Graphentra reports. No path found is not proof of safety.
[03] How they fit together
Alongside, not instead.
AI code reviewers are useful for commenting on code. Test-impact tools can shorten a test run when you have the history to drive them. Build-graph “affected” tools narrow work to the packages a change touches. AI QA and UI-regression tools help write and replay end-to-end journeys.
Graphentra is designed for a narrower job: to say which screens, APIs and flows a change can reach, with a path for each and a list of what it could not trace. Its evidence is designed to feed the other categories, and none of that integration exists yet.