# Agent workflow decision memo

Experiment ID/version: __________
Decision owner/date: __________
Decision: limited pilot / revise and retest / stop / collect more evidence

## Question tested
__________

## Result table
| Measure | Baseline | Candidate | Denominator / limits |
|---|---|---|---|
| Accepted tasks | | | All eligible cases |
| Acceptance rate | | | Accepted / eligible |
| Elapsed time | | | Same complete pairs only |
| Critical errors | | | All eligible cases |
| Direct cost per accepted task | | | Only if complete cost data available |

## Findings tied to evidence
1. Finding: __________  Case IDs: __________
2. Finding: __________  Case IDs: __________
3. Finding: __________  Case IDs: __________

## Missingness and limitations
Missing measurements and why: __________
Excluded cases and reasons: __________
Changes beyond the connection itself: __________
Sampling / execution-order limitations: __________
Claims this run does not establish: __________

## Decision and next work
Reason for decision: __________
Change to make: __________
Owner: __________
Retest cases plus fresh cases: __________
Next review date: __________

## Filled interpretation of the fictional teaching dataset
Both conditions accepted 6/8 tasks (75%). On seven complete timing pairs, baseline time was 65 minutes and candidate time was 51 minutes, a 21.5% reduction on that subset. Candidate case 5 used stale evidence and case 8 remained unresolved; its completion time is missing. No all-case time saving or cost saving can be calculated.

**Example decision:** revise source-version selection and investigate the unresolved run before expanding. Preserve this run and repeat the affected cases with additional new cases. The eight-case example does not establish statistical significance or general reliability.
