Accuracy
How reliably people detect factual, numerical and operational errors before downstream use.
THE MANDORKA INDEX
● CALIBRATION PHASE
Measure the quality of AI-assisted work.
The Mandorka Index is intended to become a public benchmark of observed work-control signals by role and track. We will not publish a dramatic number just because a dashboard can calculate one.
No index score published yetVersion-aware cohortsTrack-level reportingUncertainty disclosed
WHAT WILL BE MEASURED
Not “who uses AI.” Who controls the output.
Judgment
How consistently they prioritize material risk, impact and constraints under pressure.
Instruction following
Whether explicit requirements survive AI-assisted drafting and execution.
Verification
Whether unsupported claims are challenged instead of converted into confident output.
Communication
Whether the final message is clear, bounded and appropriate to the situation.
Assessment conditions
Published cohorts will distinguish legacy records from current session controls and assessment versions.
PUBLICATION GATE
No benchmark until the benchmark deserves trust.
Enough usable observations
Cohorts must be large enough to make the summary meaningful rather than anecdotal.
Comparable versions
Materially different assessment versions will not be silently pooled into one headline number.
Quality checks
Abuse, incomplete sessions and data-quality problems must be excluded by documented rules.
Validation context
Where outcome studies exist, descriptive benchmark data will be separated from predictive-validity findings.
DATA FLYWHEEL
Every valid proof can improve the instrument.
More completed assessments create better item diagnostics and calibration opportunities. Better calibration can improve employer utility. Employer use can create stronger validation data. That loop is the long-term Mandorka moat — not selling raw participant data.