Legibility
Legibility is the degree to which a system's intent, judgment and causality are visible to something other than the originator — another person, a future version of yourself, an organization, or an AI system. An illegible system may work perfectly, but only while the person who built it is present to interpret it.
Definition
Legibility is the degree to which a system's intent, judgment and causality are visible to something other than the originator.
That something can be another person, a future version of yourself, an organization, or an AI system. The originator is excluded on purpose: a system is not legible because its builder understands it. It is legible when someone who was not there can.
Why the distinction matters
An illegible system may still work. That is what makes the property easy to ignore. It works for as long as the person who built it is present to interpret it — and it fails the moment scale or time is introduced, because it was resting on memory, on heroics, and on someone just knowing.
A legible system can be inspected, reasoned about, corrected, extended, and trusted appropriately. That last one is the point. Legibility does not ask for more trust; it makes the right amount of trust possible to determine.
The three surfaces the term covers
- Intent legibility — what was supposed to happen. The outcome aimed at, the constraints that mattered, the tradeoffs accepted, what would have counted as failure. Most people hold this internally and never externalize it, which works only while they are the review layer themselves.
- Judgment legibility — why something was treated as good enough, or wrong. The criteria in use, the exceptions named, taste translated into checks and boundaries. This is the surface people resist most, because judgment feels personal. But judgment that cannot be articulated cannot be delegated — only imitated, and imitation is invisible until it is wrong.
- Causal legibility — what led to this result. Which inputs were used, which version of the intent was active, what changed between iterations, what assumptions were in play. Skipping it costs nothing on the day you ship and everything six months later, when the same mistake returns under a new name.
What legibility is not
- Not transparency. Transparency tends to mean showing more of what a system contains. Legibility means showing the parts a person needs in order to act. A system can disclose a great deal and remain illegible.
- Not explainability. Explainability concerns how a model reached an output. Legibility concerns whether an operator can see the aim, the criteria, and the load-bearing evidence well enough to govern the work.
- Not a demand to expose every internal detail. The standard is bounded: enough for a person to understand the task, inspect the governing distinctions, see what is load-bearing, and know where judgment must remain theirs.
- Not a guarantee of correct outcomes. Legibility makes decisions inspectable and failures locatable. That is a different property from being right, and it is the one you can build for.
- Not a compliance or audit standard. This is a working definition, not a certification scheme, and it carries no regulatory meaning.
Where this sits
Applied to delegation, this term answers a narrower question: what must be visible before you delegate to an AI system. The broader condition it serves is governable AI action under human authority. For the larger object both belong to, see the Verse.
FAQ
- What does legibility mean?
- The degree to which a system's intent, judgment and causality are visible to something other than the originator — another person, a future version of yourself, an organization, or an AI system. A legible system can be inspected, reasoned about, corrected, extended, and trusted appropriately. An illegible one depends on memory, heroics, and the person who built it standing nearby to explain it.
- Is legibility the same as transparency?
- No. Transparency usually means showing more of what a system contains. Legibility means showing the specific things a person needs in order to act responsibly — what the system is trying to do, what it treats as important, and what shaped the result. A system can publish a great deal and remain illegible if none of it answers those questions.
- Is legibility the same as explainability?
- No. Explainability generally concerns how a model produced a particular output. Legibility concerns whether an operator can see enough to govern the work: the aim, the criteria, the load-bearing evidence, and where their own judgment resumes. A system can be technically explainable and still leave a person unable to tell what it was optimizing for.