Two tasks. June 27, 2026 weighted to 8.3x leverage across 5.0 human-equivalent hours in 36 Claude-minutes. Supervisory leverage closed at 30.0x.
0.1 weeks of human-equivalent throughput in 0.6 hours of Claude wall-clock. The 10.0x ceiling came from Audit a synthesis service synthesis code for silent-failure failure modes; the 6.7x floor sat at Consolidate a synthesis service synthesis failure-mode report and fix plan.
Task Log
| # | Task | Human Est. | Claude | Sup. | Factor | Sup. Factor |
|---|---|---|---|---|---|---|
| 1 | Audit a synthesis service synthesis code for silent-failure failure modes | 3.0h | 18m | 5m | 10.0x | 36.0x |
| 2 | Consolidate a synthesis service synthesis failure-mode report and fix plan | 2.0h | 18m | 5m | 6.7x | 24.0x |
Aggregate Statistics
| Metric | Value |
|---|---|
| Total tasks | 2 |
| Total human-equivalent hours | 5.0 |
| Total Claude minutes | 36 |
| Total supervisory minutes | 10 |
| Total tokens | 180,000 |
| Weighted average leverage factor | 8.3x |
| Weighted average supervisory leverage factor | 30.0x |
| Human-equivalent weeks | 0.1 |
Analysis
The day's leverage distribution matters more than the headline figure. The 10.0x ceiling came from Audit a synthesis service synthesis code for silent-failure failure modes; the 6.7x floor was Consolidate a synthesis service synthesis failure-mode report and fix plan. Tasks at the top of the distribution share a shape: tightly-scoped specifications, clear success criteria, and minimal integration ambiguity. The AI doesn't need to discover anything new; it executes against an explicit target.
Tasks at the bottom run differently. They're either bounded by review-heavy work where every step gets verified, or they involve ambiguity that demands several rounds of trial and adjustment. The factor is real and informative, not a failure mode.
The supervisory leverage figure (30.0x today) tracks something orthogonal to wall-clock leverage. It's the ratio of human-equivalent output to human prompt-writing time. It stays high even on lower-leverage days because supervisory minutes scale with task count, not with the human-hour estimate; a 20-minute task and a 4-hour task can both be specified in two minutes of human prompt-writing.
Across the 2 tasks, the day produced roughly 0.1 weeks of senior-engineer-equivalent throughput in 0.6 hours of model wall-clock. That ratio is the practical answer to the question of how much output a single operator can move per day when the model handles the execution and the operator handles the direction.