Audit pack

What this report is — and the decision it supports

Most assessment tools can tell you what a candidate scored. Very few can prove, two years later, who scored it, against which rubric version, under what weighting, with what assistance, and why the decision followed. The audit pack is built for that second question. It is not a summary of the record — it is the record: assessment structure, methodology, scores, notes, integrity events, provenance, and decisions, packaged so that it stands on its own with no access to the platform required.

The decision it supports is usually someone else's: a compliance officer confirming the process matches policy, a legal team responding to a challenge, a procurement reviewer vetting your assessment practice, or an internal auditor sampling hiring decisions. For all of them, the pack answers the same question — can this decision be reconstructed and defended from the evidence? Because it is self-contained, the answer does not depend on logins, tribal memory, or anyone still being at the company.

When to reach for it

The moment everyone thinks of first is the challenge: a rejected candidate disputes the outcome, a regulator or works council asks how a decision was made, or litigation makes the hiring file discoverable. At that point you cannot improve the record — you can only produce it. Teams that export an audit pack when the decision is made are producing a contemporaneous file, not reconstructing one under pressure.

The quieter moments matter just as much. Procurement due diligence, where a prospective client asks how your assessments actually work before signing. Internal audit cycles, where a sample of decisions is reviewed against policy. Consultant handoffs, where an engagement ends and the client keeps a complete record of what was done in their name. In each case the pack replaces weeks of email archaeology with one file exported from the reports page.

• The moment a decision is finalized — export while the record is contemporaneous.
• When a candidate, regulator, or works council challenges an outcome.
• During procurement or client due diligence on your assessment practice.
• In internal audit or compliance reviews of hiring decisions.
• At the end of a consulting engagement, as the client's permanent record.

What's inside, section by section

The pack is ordered the way an auditor works: first how the assessment was constructed, then how it was scored, then what happened during it, then what was decided. Every claim in the later sections is checkable against the earlier ones.

• Assessment structure: section weights, rubric criterion weights, and skill mappings — the full skeleton the scores hang on.
• Plain-language methodology: how scoring works, stated so a non-specialist reviewer can follow it without a statistics background.
• Every score, attributed: each rubric score with its named evaluator and timestamp, alongside the evaluator's notes.
• Integrity events and resolutions: what was flagged during sessions, who reviewed it, and how it was resolved — flags as human-review context, never automatic rejection.
• Score rollups: the arithmetic from individual criterion scores to competency and overall results, reproducible by hand.
• AI provenance: the model and version behind any AI-assisted output, so assisted summaries are never confused with human judgments.
• The decision log: what was decided, by whom, when, and on what stated basis.

How it's produced

Nobody assembles an audit pack. It is exported from the reports page, and everything in it was captured as a side effect of running the assessment: the structure comes from the role blueprint, the scores and notes from evaluators as they reviewed, the integrity events from the session itself, the provenance from SkillCort's model and version logging, and the decision log from the decisions your team recorded. A record built this way cannot be quietly edited into a better story after the fact — which is precisely what makes it defensible.

The scoring math inside is the same transparent rollup used across the platform: evaluator scores average within a criterion, criteria combine by rubric weights, and tasks roll up by skill weight times section weight. An auditor with the pack and a calculator can reproduce every rollup. Access to the export is role-controlled, so the complete record is available to the people who need it and only to them.

How to read it well

Read it in the direction of the question you are answering. Verifying an outcome? Start at the decision log and trace backward — decision to rollups, rollups to attributed scores, scores to rubric and weights. Verifying the process? Start at the structure and methodology and read forward, checking that what was declared is what was applied. The pack supports both directions because every layer references the one beneath it.

Two misreadings to avoid. First, an integrity event in the record is not a stain on the candidate — the point of the resolution entry is that a human reviewed it and dispositioned it, and an event resolved as benign is evidence the process worked. Second, the AI provenance section does not mean AI made decisions; it exists to prove the opposite. It shows exactly which outputs were machine-assisted so that every judgment and every decision can be attributed to the named person who made it.