Strong intervention
Separated decision criteria from recommendation
The participant stopped the model from anchoring early and established the 18-month payback standard first.
Profile
Example result — fictional data | HL-DEMO-038
Prompt Operator
Challenge (evaluation version v0.3) | 20 July 2026
Experimental
Pilot result for learning and development; hiring and credential use are not active.
82/100Composite example
Observed signal
Controlled advantage
The participant's verification and decision framing materially improved the AI-only reference, with one avoidable iteration cost.
Elapsed31:08Model tokens6,420Above reference+17 pts
Score anatomy
Where the score came from
Human judgment88
Intervened at the decision-changing moments
Verification81
Caught 4 of 5 example evidence risks
AI direction76
Strong constraints; one unnecessary restart
Smart use of your AI budget72
Good quality for the AI budget spent; 18% of the budget left unused
Final work84
Clear recommendation with traceable support
Decision trail
Moments that changed the work
Verification
Challenged the demand proxy
They compared Source A with Source B and identified the difference between registrations and completed deposits.
Efficiency loss
Restarted analysis without using the prior evidence map
This consumed 1,140 tokens without adding a new decision-relevant insight.
Human ownership
Changed "launch" to a staged pilot
The final recommendation made the uncertainty actionable instead of hiding it.
Final artifact
Staged entry recommendation
The submitted response and evidence map would appear here, with sensitive source material hidden from public viewers.