SYNTHETIC PRACTICE DATA. No actual team, model run or measurement.
All minutes are active labour, including any separate reviewer's effort; unattended waiting is zero in this invented example. Cases are matched by difficulty, not identical repeats. M means manual; A means AI-assisted. All eight cases were assigned, completed and finally accepted.
Columns: case | condition | preparation | drafting/prompting | review | correction/fallback | first-pass accepted | critical first-pass error | final accepted
M1 | manual | 2 | 9 | 4 | 1 | yes | no | yes
M2 | manual | 2 | 10 | 4 | 4 | no | no | yes
M3 | manual | 2 | 11 | 4 | 1 | yes | no | yes
M4 | manual | 2 | 10 | 5 | 1 | yes | no | yes
A1 | AI | 2 | 5 | 3 | 2 | yes | no | yes
A2 | AI | 2 | 5 | 5 | 2 | yes | no | yes
A3 | AI | 2 | 5 | 6 | 3 | no | no | yes
A4 | AI | 2 | 4 | 6 | 12 | no | yes | yes
A4 required manual fallback after an invented refund promise. Its 12-minute fallback remains in the AI-assigned condition.
AI setup/training: 40 active minutes for this pilot. Manual setup/training: 0 additional minutes in this example. Licensing and any real deployment overhead are not supplied; do not assume they cost zero.
Predeclared example gates: no critical first-pass errors; final acceptance no worse than manual; at least 3 active minutes saved per completed case before allocating setup. All gates must pass. No claimed statistical significance.