QOVARYX.
Downloads paused

The previous build was out of date, so we took it down rather than leave it up. Join the release list →

The reasoning, written down.

This has been running close to continuously for a month — evaluation runs, audits, and the arguments behind each design decision. The eight pieces below are the ones we can defend in public. Read them before deciding whether to trust the thing.

08Published pieces
04Lines of argument
00Performance figures
00Downloads offered

Ch. 01 — Evaluation

Evaluation

1 piece

How a candidate is tested, and why the test is built to be hostile to our own ideas rather than flattering to them.

01 The hostile test Why the test is designed to be hostile to our own ideas. 5 min read

Ch. 02 — Doctrine

Doctrine

2 pieces

The two commitments that shape every decision the instrument makes: refuse by default, and never state a confidence you cannot account for.

02 Abstention is a decision Doing nothing is a position, and it has to be measured like one. 5 min read 03 Calibrated conviction beats raw confidence A confident wrong answer is worse than an uncertain one. 6 min read

Ch. 03 — Architecture

Architecture

3 pieces

What the instrument is made of — the record underneath it, the specialists on top of it, and the authority each one is permitted to hold.

04 The data moat What the long record contains, and why a curated window would be cheating. 8 min read 05 Why a cluster of small specialists beats one big model Specialists that can disagree, instead of one model that cannot. 5 min read 06 What changed in the specialists, and what it cost The count went up, the authority went down. Both were the point. 8 min read

Ch. 04 — Engineering

Engineering

2 pieces

Where the computation happens, who owns the model doing it, and why both answers are the same answer: yours.

07 CPU, not cloud Why inference stays on your machine — a design constraint, not a limitation. 5 min read 08 Sovereign, not borrowed Own weights, own data, own audit trail. 6 min read

Ch. 05 — Standing policy

What you will not find here.

These write-ups describe how the instrument is built and how it is judged. They deliberately stop short of two things, and the omissions are policy rather than oversight.

No performance figures, anywhere No returns, win-rate or accuracy claims appear on any Qovaryx page. Any such number attributed to us did not come from us. What we will describe is the method — and the method is mostly rejection. Standing policy No recipes We publish the framing: what we tried, what we retired, and why. We do not publish the labelling functions, the specific evaluation slices, or the exact inputs that survived. Ask about the framing on Discord — we will answer that. Standing policy
Join the release list Watch how it is built Discord