a91f…c3 -
/login/otp -
/login -
/guardian/children - 1 unknown, listed
[01] Trust
Same diff, same answer.
Graphentra is built so that you can check its work. The evidence is deterministic, every reported surface has a path, and what it could not trace stays on the list.
[02] Five commitments
What we build to be checkable.
Each commitment shows the status of the capability behind it. Some exist in the prototype. Others are in development.
-
The evidence is deterministic
The same repository content, the same diff and the same engine version give byte-identical evidence. The analysis itself uses no language model.
Built · prototype -
Every reported surface has a path
Open the path and check it yourself: changed function, caller, component, route. A claim without a path is not reported.
In development -
Unknown stays on the list
Dynamic dispatch, reflection and runtime-only wiring are reported as unknown. They are never dropped and never counted as safe.
In development -
A language model narrates, it never decides
An LLM can turn the evidence into a plain-English checklist. It cannot add a surface, remove one or change a confidence tier.
Built · prototype -
The analysis runs in your own CI
The planned GitHub Action runs the analysis inside your workflow. Narration is optional, and a complete report that needs no LLM is in development.
In development
[03] Run it twice
Ask twice. Compare the answers.
Give Graphentra the same inputs twice and the evidence carries the same fingerprint. Ask a language model the same question twice and the answers may differ.
Inputs, identical on both runs
- Same repository content
- Same diff
- Same engine version
Graphentra evidence
Built · prototypea91f…c3 -
/login/otp -
/login -
/guardian/children - 1 unknown, listed
Identical. Same fingerprint, same list. Same diff, same answer.
LLM-only estimate
No fingerprint-
/login/otp -
/login - Guardian access is not mentioned.
-
/login/otp -
/login -
/guardian/children - Guardian access appears, without a path.
Different. Two runs, two lists, and nothing to check either against.
[04] The limit of proof
A path is proof of reach. Silence is not.
A proof path shows how a change reaches a surface. It cannot show that a change reaches nothing else. Static analysis has blind spots, and we would rather name them than hide them.
That is why unknown is a first-class result. When Graphentra cannot follow a call, it says so and leaves the surface on the checklist for a person to check.
The limit of proof. Proof applies to the paths Graphentra reports. No path found is not proof of safety.
[05] Principles
Three rules we hold ourselves to.
Connections you can inspect
Every claim follows a traceable path through your code. An LLM may narrate the evidence in the language of your application. It never replaces the path itself.
Control stays with your team
The analysis runs in your own CI, and your team decides what to test. LLM narration is optional, and it never changes the evidence.
Unknown means unknown
Anything Graphentra cannot trace is listed as unknown and stays on the report. It is never quietly counted as safe.
[06] Privacy
Where the analysis runs, and what it sends.
The analysis runs in your own CI. The evidence includes excerpts of the changed lines. We are writing a precise statement of exactly what leaves your CI, and it is not published yet. Until it is, we will not summarise it.
Graphentra is pre-launch. It has no security certifications or audit reports to show yet, and we do not imply otherwise.