Skip to main content
AI MAY 29, 2026

Leverage Record: May 29, 2026

Seven tasks. May 29, 2026 weighted to 32.2x leverage across 94.0 human-equivalent hours in 175 Claude-minutes. Supervisory leverage closed at 313.3x.

Seven tasks. May 29, 2026 weighted to 32.2x leverage across 94.0 human-equivalent hours in 175 Claude-minutes. Supervisory leverage closed at 313.3x.

2.4 weeks of human-equivalent throughput in 2.9 hours of Claude wall-clock. The 80.0x ceiling came from A security-scanning service Phase 1; scanner foundation (Scanner ABC + Finding model + content-addressable dedup hash + lazy registry) plus all 7 scanner integrations (semgrep/band...; the 9.6x floor sat at Origin-service rebuild finish; retired subprocess paths (orchestrator in-process + JobStore-backed synthesis/tribunal APIs + deleted SynthesisManager/TribunalManager) + standalone....

About These Records
These time records capture personal project work done with Claude Code (Anthropic) only. They do not include work done with ChatGPT (OpenAI), Gemini (Google), Grok (xAI), or other models, all of which I use extensively. Client work is also excluded, despite being primarily Claude Code. The actual total AI-assisted output for any given day is substantially higher than what appears here.

Task Log

#TaskHuman Est.ClaudeSup.FactorSup. Factor
1A security-scanning service Phase 1; scanner foundation (Scanner ABC + Finding model + content-addressable dedup hash + lazy registry) plus all 7 scanner integrations (semgrep/bandit/detect-secrets/pip-audit/safety/checkov/trivy) each with wrapper + normalizer + severity-CWE mapping + seeded-vuln fixture + golden + full test suite; built Semgrep inline as the template then fanned out the other 6 via a parallel workflow; fixed a scanner-normalizer import cycle via lazy discovery; full suite 174 passed / 2 skipped / 99% coverage; committed locally60.0h45m1m80.0x3600.0x
2Root-caused certification-content quality instability (42 cloud certification packages, 3 repair campaigns) + committed 301 engine re-stamps + regenerated 121 single-option questions for one certification domain6.0h10m3m36.0x120.0x
3Synthesis field-preservation remediation (task 1/7): raw exam_metadata passthrough + top-level fields across an origin service (runtime and service packages) so synthesis stops dropping spec fields (root cause of spec<->manifest drift); 5 files/2 repos; 607 tests green4.5h13m2m20.8x135.0x
4Remediation (contained): content_profile.lessons gating + IRT-based adaptive question selection (2PL Fisher info replacing random.choice; helper + 3 tests; 4508 engine tests green); committed on staging4.0h15m2m16.0x120.0x
5Rename an origin service across the monorepo: runtime and service packages + 2 repo dirs + CLI rename + path-deps/lockfiles across 8 repos; infra deferred; runtime 61 / service 546 / engine 4508 suites green; committed on staging9.0h35m6m15.4x90.0x
6Origin-service rebuild; shared in-process pipelinecore (runphase/runpipeline) + made lessons/questions/tribunal stub runners real + writequestion_bank + pipeline job kind + pipeline run CLI; 16 new tests; service suite 561 green; committed on staging7.0h35m3m12.0x140.0x
7Origin-service rebuild finish; retired subprocess paths (orchestrator in-process + JobStore-backed synthesis/tribunal APIs + deleted SynthesisManager/TribunalManager) + standalone CLI mode (local flags); net -415 LOC; service suite 564 green; committed on staging3.5h22m1m9.6x210.0x

Aggregate Statistics

MetricValue
Total tasks7
Total human-equivalent hours94.0
Total Claude minutes175
Total supervisory minutes18
Total tokens1,398,000
Weighted average leverage factor32.2x
Weighted average supervisory leverage factor313.3x
Human-equivalent weeks2.4

Analysis

The day's leverage distribution matters more than the headline figure. The 80.0x ceiling came from A security-scanning service Phase 1; scanner foundation (Scanner ABC + Finding model + content-addressable dedup hash + lazy registry) plus all 7 scanner integr...; the 9.6x floor was Origin-service rebuild finish; retired subprocess paths (orchestrator in-process + JobStore-backed synthesis/tribunal APIs + deleted SynthesisManager/TribunalMa.... 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 (313.3x 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 7 tasks, the day produced roughly 2.4 weeks of senior-engineer-equivalent throughput in 2.9 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.