Skip to main content
AI JUN 01, 2026

Leverage Record: June 1, 2026

Five tasks. June 1, 2026 weighted to 27.4x leverage across 52.0 human-equivalent hours in 114 Claude-minutes. Supervisory leverage closed at 390.0x.

Five tasks. June 1, 2026 weighted to 27.4x leverage across 52.0 human-equivalent hours in 114 Claude-minutes. Supervisory leverage closed at 390.0x.

1.3 weeks of human-equivalent throughput in 1.9 hours of Claude wall-clock. The 90.0x ceiling came from ADR 0005: novel adult-learner activity set for one content domain; triaged 16 activities into 6 tiers with intellectual-property considerations, implementation plan, validation met...; the 6.7x floor sat at Two content domains to pristine+beta (cohort 30): ~6,157 questions from 0, reweighted, pairs amplified (median->15), a competence stamp, all 30 beta zero-findings, committed+pushed.

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
1ADR 0005: novel adult-learner activity set for one content domain; triaged 16 activities into 6 tiers with intellectual-property considerations, implementation plan, validation metrics, and risk evaluation; updated ADR index15.0h10m2m90.0x450.0x
2A security-scanning service Phase 1 close-out; scan endpoint + dedup-aware persistence: 5 ORM models (repos/scans/scanresults/findings/findingevents) + Pydantic schemas + orchestrator (sequential scan execution with dedup/reopen/fixed lifecycle) + 8 REST endpoints (repos/scans/findings with status PATCH + audit events) + hand-authored Alembic migration + 13 endpoint tests; full suite 186 passed / 99% coverage; committed locally22.0h26m1m50.8x1320.0x
3Phase 0 of ADR-0005: extended activities catalog with 16 novel-activity entries for one content domain + 4 addons + academic category + deferred runtime_source + per-activity feature-flag plumbing + library mirror sync + 8 new tests + fixed 2 pre-existing test failures8.0h18m1m26.7x480.0x
4Forensic diagnosis of crashed cross-session synthesis run + safe checkpoint resume (one certification domain)2.0h15m3m8.0x40.0x
5Two content domains to pristine+beta (cohort 30): ~6,157 questions from 0, reweighted, pairs amplified (median->15), a competence stamp, all 30 beta zero-findings, committed+pushed5.0h45m1m6.7x300.0x

Aggregate Statistics

MetricValue
Total tasks5
Total human-equivalent hours52.0
Total Claude minutes114
Total supervisory minutes8
Total tokens510,000
Weighted average leverage factor27.4x
Weighted average supervisory leverage factor390.0x
Human-equivalent weeks1.3

Analysis

The day's leverage distribution matters more than the headline figure. The 90.0x ceiling came from ADR 0005: novel adult-learner activity set for one content domain; triaged 16 activities into 6 tiers with intellectual-property considerations, implementation...; the 6.7x floor was Two content domains to pristine+beta (cohort 30): ~6,157 questions from 0, reweighted, pairs amplified (median->15), a competence stamp, all 30 beta zero-findin.... 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 (390.0x 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 5 tasks, the day produced roughly 1.3 weeks of senior-engineer-equivalent throughput in 1.9 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.