Your AI says it’s done.
Ranex proves whether it is.
Every AI coding tool swears the work is finished. Ranex is the one that checks — reading the real code, running tests the AI can’t touch, and refusing to pass anything without evidence.
Open-source kernel · MIT licence · pre-release
▶ This loop plays automatically — it’s what a real Ranex run looks like. The agent’s cheerful report is thrown away; the verdict comes from checks.
Tools like Lovable, Replit, and Base44 help the AI throw better darts. Ranex is the judge that says where they actually landed.
The AI grades its own homework.
Of course it gets an A.
When AI builds your software today, the same AI writes the code, writes the test that checks the code, runs it, and tells you the result. Nobody independent ever looks. That’s a student marking their own exam. Watch the difference:
How today's tools score
Self-scoredThe test was written after the code, by the same AI that wrote the code. It always passes — because the target is painted wherever the dart happens to land.
How Ranex scores
Frozen targetThe target is drawn and frozen before the throw. A miss is recorded honestly — and after three misses, Ranex stops and asks you instead of quietly moving the target.
Ranex doesn’t try to make the thrower better — plenty of tools compete on that. It fixes the scoring: target locked first, no self-grading, every point comes with a receipt.
One loop, five rules, zero trust in anyone’s word.
Here is the whole system in one picture. Follow the pulse: your idea becomes an approved map, the map becomes a locked target, the AI builds, and a small piece of ordinary, inspectable code — the kernel — delivers the verdict. No AI anywhere in the judging.
● The pulse is one attempt travelling the loop — a miss loops back, a pass is merged by the kernel, never by the AI.
You approve the map
Before anything is built, Ranex turns your idea into a plain-English map of what the software must do. You read it, fix it, approve it. That approved map is the target — nothing else counts.
The target locks
The map becomes tests, and the tests freeze before any code is written. The AI that builds can never edit the tests that judge it.
The AI builds — Ranex watches the work, not the words
Agents write the code in an isolated workspace. Ranex reads what actually changed on disk and runs the frozen tests. The agent’s own summary goes straight in the bin.
Every claim needs a receipt
“It works” only passes with evidence tied to the exact version of the code it describes. A claim with no evidence doesn’t default to fine — it blocks.
Misses stop the game
Three failed attempts and Ranex stops and asks you — instead of quietly lowering the bar. “This can’t be hit as specified” is an honest answer, and you’re the one who decides what happens next.
I’ve spent 17 years watching systems fail.
Most of that time was spent on the unglamorous end — platform support, incident work, four years at Pantheon on sites where downtime was measured in money. You learn a specific lesson doing that.
The dangerous failure isn’t the loud one. It’s the green light that proves nothing. A check that never ran. A test that was skipped. A report nobody verified.
AI coding agents made that failure cheap to produce at scale. So I build tools that prove things instead of claiming them — Ranex is one of them.
Anthony Garces
Senior Principal Lead Architect · Founder, Anito
- in software
- 17 yrs
- lead architect
- Principal
- 4 years, platform WebOps
- Pantheon
Ranex is pre-release and its README says so before it says anything else. I’d rather you evaluate the real thing.
Same problem, different corners of it.
Ranex judges whether work was actually done. These take the same question earlier and later in the chain. All are open source, all are early, and each says where it stands.
Software you can trust without reading a line of code.
Vision
Anyone can commission working software — and know it works, with proof, not promises.
AI has given millions of people the power to build. It hasn’t given them the power to verify. Until it does, every non-technical builder is one confident “all tests pass!” away from shipping something broken to real customers. We think the verification gap — not the building gap — is the thing actually holding this revolution back.
Mission
Make “done” mean something: every claim carries evidence, every miss is visible, every hit is provable.
We build the honest scoreboard for AI-built software: targets locked before the work starts, no self-grading ever, verdicts delivered by open-source code anyone can inspect. Misses become visible and cheap. Hits become provable. That’s the whole job, and we refuse to claim more.
What Ranex won’t tell you
Ranex does not make your AI write better code. Not by one degree. The AI you use today will be exactly as good tomorrow. What changes is what you know about its work — and what you can prove to yourself, your customers, and your team.
When a tool promises you both a better thrower and an honest scoreboard, ask which one it actually measured.
The ideas that shaped this thing.
Ranex came out of specific failures — checks that didn’t run, tests that vanished, green lights that proved nothing. I write those down as I hit them.
Read from the repo, not written for you.
Ranex ships in slices — closed only when the criteria are proven. This list comes straight from the kernel’s own record, so it can’t drift from what landed.
Full changelog- Judgment identity and verdict read channel
SLICE-02013 AugEvaluation records now carry structured five-kind causes and self-approval; projection composes refused and unattributable rejections. Dedicated verdict signing, validated atomic publication, and a total closed-state reader provide the kernel-owned channel consumed by later UI/board work.
- Host qualification as gate evidence
SLICE-01913 AugThe landing gate now consumes host qualification as signed, subject-bound evidence under the existing `EVIDENCE_DOMAIN`; admission deeply validates the closed report and re-reads durable boot, machine, LSM, userns-sysctl and parent- namespace uid/gid anchors, refusing absent, stale, mismatched, ambiguous or self-approved evidence. Qualification runs as a host operation through `cmd_run`, while the kernel remains byte-exact. The real-operator e2e is honestly guarded and skips without delegated cgroup controllers; `cmd_run` confinement and RISK-06 remain open.
- Confinement of the bound command
SLICE-01712 AugIts 47 qualification gates qualified the strict-local host and byte-reproducible GNU C17 native launcher, binding LSM state, user-namespace sysctls, boot ID, machine ID and delegation identity. ADR-006 and RISK-06 remain open for SLICE-018/019; `cmd_run` integration is explicitly deferred to SLICE-019.
The judge is public. Watch us build it — or help.
A trust product you can’t inspect is just another promise, so the Ranex kernel — the part that decides pass or fail — is open source under MIT, and the project is built in public. Every rule below is enforced by code in the repo, not written in a policy document.
I write down what breaks. You can read it.
Notes from building a system that judges AI-written code by evidence — the mechanisms, the failures, and what a green light actually proves. No cadence promises, no filler.