21 tasks. August 1, 2026 closed at 42.6x weighted leverage across 1,112.0 human-equivalent hours in 1,566 minutes of wall-clock time. Supervisory leverage came in at 725.2x.
That is 27.8 weeks of human-equivalent throughput in 26.1 hours. The ceiling was 160.0x; the floor was 7.9x. 16 of the 21 entries came from a single project.
Task Log
| # | Task | Human Est. | Claude | Sup. | Factor |
|---|---|---|---|---|---|
| 1 | Documentation consolidation: docgen pipeline (11 doc-type templates + PDF/hero/ASCII/determinism) + 17-agent cataloguing fleet over 1038 docs + 199 units rendered | 200.0h | 75m | 6m | 160.0x |
| 2 | Documentation | 320.0h | 195m | 10m | 98.5x |
| 3 | Testing | 16.0h | 18m | 4m | 53.3x |
| 4 | Documentation | 20.0h | 25m | 6m | 48.0x |
| 5 | Deployment | 140.0h | 178m | 2m | 47.2x |
| 6 | Stage 3 documentation repair: 54 docs repaired by 6 Opus agents with live verification + dedup of 50 redundant units + app-content guard | 60.0h | 80m | 4m | 45.0x |
| 7 | Deployment | 3.5h | 6m | 8m | 35.0x |
| 8 | Documentation | 7.0h | 12m | 2m | 35.0x |
| 9 | Testing | 14.0h | 24m | 3m | 35.0x |
| 10 | Deployment | 130.0h | 270m | 10m | 28.9x |
| 11 | Fleet registry + CI/CD sweep: destroyed 18 orphaned CodeBuild projects across 9 AWS accounts, migrated 28 repos from stale CodeArtifact to Verdaccio (mirrored 6 missing versions, relocked all, un-gitignored… | 40.0h | 90m | 3m | 26.7x |
| 12 | Coding | 28.0h | 71m | 3m | 23.7x |
| 13 | Research and file 3 single-domain-vertical defect cards for avian-app-web (dead-end Browse courses CTAs across 19 sites, singleDomain classification for LSAT/MCAT, USA to US Citizenship nav title) | 1.5h | 4m | 4m | 22.5x |
| 14 | Infrastructure | 11.0h | 32m | 4m | 20.6x |
| 15 | Debugging | 36.0h | 115m | 3m | 18.8x |
| 16 | Prod defect run: root-caused and fixed the certs checkout lockout (guard/unique-index mismatch), blank-card render bug, product-scoped subscribe rail, restored Apple Sign In via SSM migration + 2 ECS task-def… | 36.0h | 130m | 6m | 16.6x |
| 17 | Speed program: concurrency census + parallelize 9 serial LLM loops on the [content generation] path (shared order-preserving map_calls primitive + 18-case concurrency/equivalence battery + fleet restart) | 16.0h | 65m | 4m | 14.8x |
| 18 | Root-caused Apple Sign In popup hang across AVIAN (Apple web_message posts to the redirect_uri origin, not the opener): threaded a vetted per-origin redirect_uri through avian-app-web and avian-api, shipped… | 16.0h | 74m | 4m | 13.0x |
| 19 | Diagnosed and fixed Sign in with Apple failing on all 12 AVIAN product hosts: CloudFront CSP allowed applepay.cdn-apple.com (Apple Pay) but never appleid.cdn-apple.com (Sign in with Apple); reconciled… | 4.0h | 19m | 2m | 12.6x |
| 20 | CodeArtifact cleanup: inventoried 27.4GB domain, identified console-sim as 93.7% of storage (222 versions), deleted 316 unreferenced private versions (~25.6GB) preserving all fleet-referenced versions +… | 8.0h | 45m | 2m | 10.7x |
| 21 | Documentation | 5.0h | 38m | 2m | 7.9x |
Aggregate Statistics
| Metric | Value |
|---|---|
| Total tasks | 21 |
| Total human-equivalent hours | 1,112.0 |
| Total Claude minutes | 1,566 |
| Total supervisory minutes | 92 |
| Total tokens | 37,445,000 |
| Weighted average leverage factor | 42.6x |
| Weighted average supervisory leverage factor | 725.2x |
| Human-equivalent weeks | 27.8 |
Analysis
The highest factor of the day came in at 160.0x and the lowest at 7.9x, a spread of 20.3 times between the two. That is a wide range for a single day, and it usually means the day mixed mechanical work with work that needed real judgement.
The largest single entry accounted for 200.0 of the 1,112.0 human-equivalent hours, or 18 percent of the day. No single task dominated the total, so the weighted average is representative.
Supervisory time was 92 minutes against 1,566 minutes of execution, a ratio of about 1 to 17. Supervisory leverage of 725.2x is the figure I find most honest, because it measures the hours I actually spent rather than the hours a machine spent on my behalf.
Every figure here is recorded at the time the work is done rather than reconstructed afterwards. The human estimate is my own judgement and carries the uncertainty that implies; the minutes and tokens are measured. The full dataset, including this day, is available for download.