Abstention is a decision
The first thing a competent trader learns is when to sit out. The second is that sitting out is itself a position. Most automated tools are bad at it, because they are scored on what they say and never on what they decline to say.
Ch. 01 — Doctrine
Why declining is hard to learn
If a system is trained to reward predicting the correct side of a move, it learns to always predict some side. Declining to answer looks, to that kind of training objective, like failure. So the system finds an answer even when no answer exists in the data in front of it.
That is the mechanism behind every automated alert that fires in directionless conditions, in the window around a scheduled announcement, in the middle of a macro panic. The system was never designed to abstain. It was designed to pick a winner from the options available, and abstaining was not one of the options.
How we score restraint
We measure refusal coverage directly, against a hand-curated set of situations where the correct answer was to do nothing. The set is built from the kinds of conditions that reliably produce confident nonsense:
- The window immediately around a scheduled macro announcement on a volatile day.
- Conditions where broad market indices are pulling against each other and no coherent direction exists to read.
- A symbol sitting a couple of days ahead of a scheduled company event, with option pricing already reflecting the uncertainty.
- A setup with the surface appearance of a breakout but none of the participation underneath it.
A system that looks strong on ordinary conditions but collapses on this set is a system that will do its worst damage on the days that matter most. We would rather ship the one that looks more modest on ordinary days and holds its discipline on the bad ones.
The days a tool ruins an account are not average days. Measuring it on average days tells you nothing about them.
On why the hostile set existsA dedicated component whose only job is to say no
Restraint is not left as a by-product of the components that read direction. We run a separate specialist whose only function is to raise an objection, evaluated in parallel with the components that are trying to read the market. It has no opinion about which way anything is going. Its question is narrower and more useful: is this situation legible enough to judge at all?
The system combines both. The consequence is that it declines considerably more often than the direction-reading components alone would. Some of those refusals would have worked out. We accept that trade deliberately: smaller swings, shallower worst cases, and a tool that is still standing after a bad month.
We publish no performance figures. The claims on this page are about method and behaviour. Any return, win-rate or accuracy number attributed to Qovaryx did not originate from us.
A system that always gives you an answer is selling you certainty it does not have.
On the cost of always answering