How we score.

Each pillar is scored on a 0 to 100 benchmark against a published rubric, and each score maps to one of five behaviourally-anchored bands: Developing, Proficient, Strong, Distinctive, Exemplary. Each band is anchored to observable behaviours, not adjectives. A "Distinctive" on Prioritisation is not "was really good at prioritising." It is, specifically: re-sequenced the day without prompting when a new input arrived, held the principal’s agenda in mind when the obvious path would have solved the smaller problem, named the trade-off in writing rather than just picking one.

The behavioural anchors are identical for the employer edition and the candidate (Certified) edition at every level. The scenario changes, the rubric does not. A Level 2 Certified is directly comparable to a Level 2 employer report. For the process end to end, from commission to report, see how the assessment works.

Every report is reviewed and signed off by Lily Shippen, across every stream. The rubric is co-authored with Emily Partridge, former EA Team Manager, and the reviewer pool recalibrates quarterly against a shared library of anchor responses. A sample of reports is double-reviewed for inter-rater reliability, and drift is corrected at the quarterly calibration.

Reviewers never see the candidate’s name, company, or CV during scoring. The assessment file is pseudonymised before it reaches the reviewer queue. The candidate is identified by level and scenario only.

Name, age, and education. Stripped from the candidate’s submission before the reviewer sees it. The reviewer scores on what the candidate did, not on where they were educated or how their name reads.

Writing style. The rubric scores decisions, not prose quality. A well-reasoned bullet list at Distinctive band outscores a beautifully written paragraph that misses the call.

Time of day. Reviewers score in batches, not in sequence, so a reviewer’s third report of the morning is not compared to their eighth of the afternoon.

Accommodations. Candidates can self-certify a +25% time accommodation for neurodivergence, disability, or medical reasons at the pre-checkout stage. No diagnosis required. Reviewers do not see which reports had accommodation applied, and the final band is not adjusted down.

The assessment is built from modules, not questions. Each module stress-tests a specific capability that our combined decades of hiring, placing, and reviewing EAs tell us is decisive at that level. Nothing is there to pad the time. The stack has been rebuilt twice since 2024 as the profession evolved.

The six core modules, Inbox Triage, Simultaneous Demands, Conflicting Priorities, Calendar Rebuild, Email Drafting, and Stakeholder Dilemma, are the immovable skills of the role. They run across every level and are recalibrated at each altitude: a Level 1 Inbox Triage is not a Level 3 Inbox Triage, even though both carry the same name. At Level 2 and Level 3 we add an AI Fluency probe, where candidates are asked to revise an AI-drafted output or describe how they would use AI for a specific task. At Level 3 we also add a Data Interpretation module: a KPI dashboard or a messy spreadsheet, with the brief to surface the three things the principal needs to know.

A short Mindset layer runs alongside the modules. Five questions, scored separately. Not a pass/fail layer, a qualitative overlay that tells the reviewer whether the candidate understands the role they are being hired into.