Two tasks. July 5, 2026 weighted to 7.3x leverage across 20.0 human-equivalent hours in 165 Claude-minutes. Supervisory leverage closed at 85.7x. A focused two-task engineering day: content-remediation tooling (a confirmation-sweep harness across models, question re-anchoring, node regeneration) and a pluggable provider-registry refactor for the retrieval-answering service.
0.5 weeks of human-equivalent throughput in 2.8 hours of Claude wall-clock. The 8.0x ceiling came from Content-remediation tooling: confirmation-sweep harness across candidate generation models, live-package question re-anchoring, critical-goal promotion...; the 6.0x floor sat at Pluggable provider registry for the RAG answerer (mirrors the content-generation pipeline's provider-loading logic), adding a new candidate generation model....
Task Log
| # | Task | Human Est. | Claude | Sup. | Factor | Sup. Factor |
|---|---|---|---|---|---|---|
| 1 | Content-remediation tooling: confirmation-sweep harness across candidate generation models, live-package question re-anchoring, critical-goal promotion, confirmed-error node regeneration (4 new tools plus fixes) | 14.0h | 105m | 10m | 8.0x | 84.0x |
| 2 | Pluggable provider registry for the RAG answerer (mirrors the content-generation pipeline's provider-loading logic), adding a new candidate generation model provider; refactored hardcoded per-model branching into style-dispatched dispatch; caching-path generalization to support an additional provider's compatible endpoint; provider key-resolution fix (dedicated key wins over generic key); 44 tests; ran a 5-package non-omniscient sample (4526 answers, 0 auth errors) | 6.0h | 60m | 4m | 6.0x | 90.0x |
Aggregate Statistics
| Metric | Value |
|---|---|
| Total tasks | 2 |
| Total human-equivalent hours | 20.0 |
| Total Claude minutes | 165 |
| Total supervisory minutes | 14 |
| Total tokens | 870,000 |
| Weighted average leverage factor | 7.3x |
| Weighted average supervisory leverage factor | 85.7x |
| Human-equivalent weeks | 0.5 |
Analysis
The day's leverage distribution matters more than the headline figure. The 8.0x ceiling came from Content-remediation tooling: confirmation-sweep harness across candidate generation models, live-package question re-anchoring, critical-goal promotion...; the 6.0x floor was Pluggable provider registry for the RAG answerer (mirrors the content-generation pipeline's provider-loading logic), adding a new candidate generation model.... 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.
Both tasks were build-and-verify work backed by test suites, which lands them in the moderate-leverage band; it was the smallest day by human-equivalent output in this stretch, about half a week.
The supervisory leverage figure (85.7x 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 0.5 weeks of senior-engineer-equivalent throughput in 2.8 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.