AI software reliability
GeekieNews publishes essays, audits, and tool releases on deterministic verification for agent-generated code:static analysis, never the model.
audit
A hand-labelled corpus said our repository context was good. Replaying a real project's own commits said it was not. The gap between the two numbers is the useful part, and it is why Solumbe 4.8.0 reads dependency work differently.
release
The trust harness behind the verification, routing and merge-gate work on this site has a new name. Here is the short history of why, what changed for anyone using it, and why the version numbers skip from 1 to 3 to 4.
audit
A router that sends every task to the most expensive model is not a router. Two defects explain why our first one did, and both were found by running it rather than reasoning about it.
audit
A merge gate should prove that valid changes pass, unsafe changes fail for the right reasons, and local evidence never impersonates hosted review or deployment proof.
essay
Stronger coding agents do not make the harness obsolete. They make independent evidence about scope, validation, ownership, CI, and human review more important.
