The job search is one of the highest-stakes decisions most people make, and yet the tools that recommend roles are usually black boxes. A site shows you a job and you have no idea why. We decided early that Jobrods would not work that way. Explainability is not a feature we bolt on at the end — it is a constraint we design within from the start.
Principle one: every score is decomposable
A match score is never a single opaque output. It is always the sum of named, weighted factors that we can show you. This rules out certain modelling approaches that would be marginally more accurate but impossible to explain. We accept that trade — a slightly simpler model you can trust beats a slightly better one you cannot.
Principle two: explanations are generated from the same maths
The breakdown you see is not a post-hoc story we invent to sound plausible. It is read directly from the factor contributions that produced the score. If the explanation and the score ever disagreed, that would be a defect. Keeping them tied to one source of truth is an engineering discipline we enforce in code.
Principle three: no dark patterns
We do not boost roles because a recruiter paid us to, and we do not bury the reasoning behind a vague "recommended for you." If a role ranks highly, it is because it scored highly on factors that matter to you, and you can verify that.
Why this matters for hiring
Opaque recommendation systems can quietly encode bias and are nearly impossible to audit. A decomposable, transparent model is something we — and eventually regulators and candidates — can inspect. In a domain that affects people's livelihoods, that accountability is not optional.
Explainable AI is more work. It constrains our choices and forces us to justify the system to ourselves. We think that is exactly the point.