How the assessment is built, scored and kept honest.
This page sets out the actual mechanics behind Workplace Judgement — not marketing claims about them. If a claim below isn't true of the product as it stands today, it isn't on this page.
Behaviours in the framework
25
Responsibility levels
5
Validated scenarios in the bank
22
The behavioural framework
Every behaviour in the framework — Judgement & Decision Making, Accountability & Ownership, Delegation & Empowerment, and so on — has a plain-English definition, a description of what good performance looks like, positive and warning behavioural indicators, and notes on how it shifts in emphasis at more senior responsibility levels. None of this is generated on the fly: it's written and maintained by our assessment team, reviewed like any other content asset. Scenario complexity and the behaviours available scale with responsibility level — a Strategic Responsibility assessment draws on a genuinely different scenario set to a Foundational one, not the same content reworded.
Scenarios, actions and scoring
Each scenario presents a realistic workplace situation with four possible actions, rated independently as Counterproductive, Ineffective, Fairly Effective, or Effective. Nothing is marked simply right or wrong: every rating is compared against a pre-defined expert effectiveness benchmark using a graded distance formula, so a rating close to the benchmark scores well and the score degrades gracefully — never a cliff-edge — the further it sits from that benchmark. The weighting behind this (how much situational judgement counts versus the behavioural-preference section, and how role-critical behaviours are weighted higher) is fixed and disclosed to employers before any assessment is issued.
Evidence and consistency, not just a score
Every behaviour score in a recruiter report is shown alongside how much evidence actually supports it — how many scored decisions, across how many distinct scenarios — so a score built on one response is never presented with the same confidence as one built on a dozen. Where a candidate's responses vary more than typical across situations that probe similar underlying judgement, that's surfaced as a consistency indicator, never as a claim about honesty. Named judgement-risk patterns work the same way: they only ever appear when the underlying evidence — a real, counted share of diverged responses — actually supports them, gated behind a minimum amount of evidence so a single unusual answer never becomes a flag.
Question governance
Every scenario moves through a Draft → In Review → Approved lifecycle before it can be used in a live assessment, with the approving reviewer and approval date recorded. A scenario can be flagged for re-review or retired at any time — including specifically for a fairness or bias concern — and retired scenarios are replaced with a like-for-like alternative rather than silently dropped from assessments already in progress.
Fairness monitoring across the whole bank
Beyond individual scenario review, we monitor two signals across every scenario in the bank: how often candidates who reach a scenario actually complete it, and how much candidate responses diverge from the expert benchmark on that specific question, pooled across every employer who has used it. Both signals are gated behind a minimum sample size — a scenario with too little data reports as such rather than being given a false clean or false concerning reading from one or two responses.
Deterministic, versioned scoring
The scoring engine is a fixed, deterministic calculation — never a generative model deciding a score. The exact scenario set, weighting and scoring configuration used for a vacancy is frozen the moment the first candidate is invited, so every candidate on that campaign is scored against precisely the same standard, and later changes to the question bank or scoring configuration never silently alter a result that's already been issued.
If something goes wrong
A link stopping working, a session getting interrupted, or a result looking wrong are things we can actually investigate — not guess at. Every meaningful step of an assessment attempt is recorded: when an invitation was opened, when a session started, and when it was submitted, or wasn't. Access is controlled through unique, single-use links rather than shared passwords. Any correction — extending an expired link, unlocking a stuck session, or authorising a genuine reattempt — is deliberate, requires a stated reason, and is recorded against that reason. None of it ever edits a candidate's existing answers or score. If you report a problem, you get a case number, and we can see the same trail you're describing.