Can a change pass every test and still be wrong?
Yes. A developer can satisfy the immediate ticket while breaking a caller, changing a business rule, or leaving critical behavior untested. Verity connects intent, implementation and executable evidence.
An independent engineering proposal by NANDAX. Explore NANDAX Publications and the developer's LinkedIn profile. Twelve controlled cases are in the private benchmark suite; the public preview shows one recorded case. Full-scale integrations are proposed for a custom pilot.
Correctness has three sides.
Passing tests is only one observation. A trustworthy review also checks what the change was meant to do and what existing behavior it must preserve.
Six steps, one evidence trail.
From the original requirement to an auditable report. Select a step to see its information, tools and limitations.
Gather context
Collect the original requirements, code and tests.
Inspect the change. Challenge the tests.
Browse all twelve implemented benchmark cases. One is open with recorded evidence; eleven are private and offer a reviewer-access request on selection. The local Python edition executes trusted reference tests; this browser preview does not run Python or LLMs.
Private cases are available for a guided evaluator review. Their original code, regression tests and reference corrections are not published here.
Return contract regression
A changed function disrupts checkout
Edits trigger fresh static analysis in local mode. Only bundled, byte-identical fixtures are executable.
What behavior changed beyond the original request?
Findings, with source evidence.
Not every static concern is a proven defect. Observed test failures are recorded separately.
What did the tests actually prove?
The original suite and an independent regression suite are executed separately in controlled benchmark mode.
The local test runner operates on built-in developer-owned fixtures. It is not an isolation boundary for arbitrary customer repositories.
A correction should survive scrutiny.
Review a known reference patch before applying it in this temporary workbench.
A reference correction is not an AI-generated patch. Execute the tests again to see what it changes.
The evidence, in a reusable report.
Observed outcomes, static findings, source citations and remaining uncertainty.
Follow the evidence, not the claim.
Explore the system architecture, the execution pipeline, and the links between requirements, changed code and tests. Every node has an explanation.
Select a stage to see whether it ran for the current case. A missing stage is labeled—not silently marked successful.
A clickable trace is a navigational aid. Source-line existence alone does not establish semantic correctness.
Select any node to examine its purpose and available evidence.
A viable prototype. A credible path to scale.
The private prototype demonstrates controlled, executable tests. Scaling to a customer repository requires real integration and security work. Each phase below has an explicit acceptance gate.
Working assessment
From an assessment prototype to
your engineering workflow.
Verity is not a generic chatbot or a claim of a finished enterprise platform. The core is demonstrated on controlled Python cases. A custom Myco pilot would connect real repositories and requirements to that same evidence-first review process.
One agreed repository, one clear requirement source, baseline CI evidence and independently checkable findings—with a documented false-positive review.
Discuss a custom pilot ↗Capability map
Select any module to see what is built versus what needs custom project engineering.
Evidence Core · Implemented
Python AST checks, benchmark fixtures and source-linked reports.
Scope distinction: The public site contains a single recorded case. Twelve fixtures and the deterministic runner are in the separate private engineering build. GitHub/Jira integration, hosted LLM inference, secure execution of third-party code, multi-tenant SaaS and enterprise authentication are not deployed; they form the proposed custom-project work.
Show the evidence.
Make the decision.
A serious review system is explicit about what was observed, what was inferred and what is still unknown.