How the Consistency & Bias Engine actually scores a case
"Flags deviations, never a verdict" sounds good on a features page. It's a fair question to ask what it actually means in practice. Here's the mechanism, end to end.
It starts with two numbers, kept apart
Every case in Rectifia gets a severity score and an evidence score. These are never merged into one number, on purpose. A severe allegation with thin evidence and a moderate allegation with a paper trail a mile long are different problems, and collapsing them into a single "priority score" would hide that difference from the person who has to act on it. The investigator sees both, separately, from the moment the case lands.
Neither score is a conclusion about what happened. They're inputs that route the case to the right handler and shape what the intake process asks for next. A high evidence gap on a serious allegation might trigger a follow-up question before the case even reaches a human. A low-severity, well-documented case might route straight to a handler without extra steps. Nothing here decides guilt. It decides queue position and next questions.
Closed cases become reference points
Here's where it gets more interesting than a scoring rubric. When a case closes, it doesn't disappear into an archive. It becomes a reference point: category, severity score, evidence score, department, and whatever action was actually taken. No names, no narrative text, no identifying detail. Just the shape of the decision.
Over time, that builds into a record of how your organization has actually handled things - not what the policy says should happen, but what has actually happened, case by case, for real.
The comparison happens before a case closes, not after
When a handler proposes closing a new case with a specific action, the engine looks for similar reference cases: same category, severity within a set range, comparable department tier. If there are enough of them to say something meaningful, it compares the proposed action against that pattern.
If the proposal is unusually lenient or unusually harsh relative to what similar cases got, it's flagged right there, before the case is finalized - with the reasoning attached, not just a red flag with no explanation. The handler sees why it was flagged, what the comparable cases looked like in aggregate, and can then make an informed call: proceed anyway, with the reasoning documented, or reconsider.
That's the whole mechanism. It's a mirror pointed at your own history, held up at the one moment a decision is still reversible.
What it deliberately doesn't do
It doesn't tell anyone what action to take. It doesn't say "this is wrong." It doesn't rank investigators or track who's "too lenient" as a performance metric - that would turn a fairness tool into a surveillance tool, and it would make handlers defensive instead of thoughtful, which is the opposite of what you want when someone's job or reputation is on the line.
It also doesn't force a comparison when it doesn't have enough to go on. Below a minimum number of comparable reference cases, it simply says there isn't enough data yet, rather than manufacturing a pattern out of two prior cases that happen to share a category. A flag based on insufficient history is worse than no flag at all - it either gets ignored, or worse, it gets trusted when it shouldn't be.
And it flags in both directions. An organization that only gets warned when it's being too lenient will drift toward harsher discipline over time without meaning to, simply because that's the only signal it's ever getting. Flagging leniency and severity outliers equally is what keeps the tool honest.
Why this exists at all
Two similar cases getting two different outcomes rarely happens because anyone decided to treat people unfairly. It happens because every case gets judged in isolation, by a person, without a memory of the fifty cases that came before it. Different investigator, different week, no visibility into the pattern.
That inconsistency is invisible from the inside - every individual decision felt reasonable when it was made. It only becomes visible when someone lines up ten closed cases side by side, and by then it's usually evidence in a dispute, not a flag someone got to see before the case closed. The Consistency & Bias Engine exists to surface that comparison at the point where it can still change the outcome, instead of six months later when it can only be explained after the fact.