The Framework

Sensor Properties

We don't rank sensors as "good" or "bad." Every sensor is characterized along six dimensions that determine when it is useful, what it can and cannot detect, and what evidence it produces for an agent or human.

The important thing is that no single sensor measures correctness. Each sensor measures one thing. Coverage measures execution. Mutation measures test sensitivity. Types measure a particular class of structural inconsistency. Contracts measure boundary assumptions. Observability measures what actually happened and preserves enough dimensionality to investigate unknown unknowns.

The question becomes: what independent observations would cause us to update our belief that this software is correct?

01

Oracle strength

How confidently does it know that something is wrong?

compiler error
10/10
type error
10/10
test assertion
9/10
mutation
9/10
linter
8/10
coverage
4/10
complexity
2/10
code review
6/10

A compiler has maximum oracle strength because the implementation cannot argue with it. A complexity metric has low oracle strength because high complexity doesn't prove anything is wrong — it just suggests increased risk.

02

Independence

Can the thing being evaluated manipulate the sensor?

This is extremely important for agents. A model writing tests/ is allowed to write tests that make itself pass. The producer and evaluator should be separated wherever possible.

An instruction saying "verify this" is weaker than a gate that literally refuses to proceed unless the verification command succeeded. Computational controls rather than prose rules.
03

Scope

What level of the system does it tell us about?

Line A single line of code
Function A single function or method
Module A package or module
Service A single service or component
System The whole system, across services
User journey What the user experiences end-to-end
04

Feedback latency

How long until the sensor tells you something?

compilermilliseconds
unit testsseconds
integrationminutes
mutationminutes / hours
canaryminutes
productionhours / days
incidentweeks
05

Actionability

Does it merely say "bad" or does it tell you what to fix?

This is where guiding sensors become particularly interesting. A guiding sensor doesn't just flag a problem — it tells the agent what to do next. In Böckeler's framing, the interesting frontier is sensors where the feedback itself directs the next action.

06

Predictive vs retrospective

"This is wrong" or "this looks like things that became wrong before"?

You don't need to understand FooManagerFactoryImpl. You can observe: 27 changes in six months, 8 reverts, 4 incidents, touched by 11 teams. That's a signal — a black-box sensor of maintainability.