Skip to main content
AI MAY 22, 2026

Leverage Record: May 22, 2026

22 tasks. May 22, 2026 weighted to 27.3x leverage across 425.2 human-equivalent hours in 935 Claude-minutes. Supervisory leverage closed at 447.6x.

22 tasks. May 22, 2026 weighted to 27.3x leverage across 425.2 human-equivalent hours in 935 Claude-minutes. Supervisory leverage closed at 447.6x.

10.6 weeks of human-equivalent throughput in 15.6 hours of Claude wall-clock. The 161.5x ceiling came from Full an inference engine accessibility audit (50 repos, deterministic Phase 0 + 4 parallel LLM agents, ~288 findings) followed by full compliance audit (12 sections, 4 parallel age...; the 2.5x floor sat at a marketing site PMP card+banner: set available_at, update category/course-page templates to render Available May 25th.

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
1Full an inference engine accessibility audit (50 repos, deterministic Phase 0 + 4 parallel LLM agents, ~288 findings) followed by full compliance audit (12 sections, 4 parallel agents, 1 CRITICAL + 5 HIGH gaps, consolidated SOC 2/GDPR/CCPA report)70.0h26m2m161.5x2100.0x
2Accessibility zero-disruption HIGH sweep: 4 parallel agents fixed ~135 HIGH findings across 30+ repos — Phase 0 went from 60 to 0 verified by deterministic checker; 488 scope=col + 15 aria-modal added across 21 tools; canvas/SVG/input aria-label additions; outline:none replacements; aria-grabbed deprecated to...80.0h35m1m137.1x4800.0x
3Full an inference engine readiness audit: Phase 0 canonical + 4 parallel agents across 60 repos (core+services, clients+libs, 21 tools, docs+sites+infra), consolidated report at audit-report-2026-05-22.md with 10 HIGH + 13 MEDIUM (4 self-fixed in-flight) + 10 LOW; 9,157+ tests verified green40.0h25m1m96.0x2400.0x
4Readiness audit rerun: 4 parallel agents verified today HIGH fixes landed clean (admin, electron, infra) + audited 42 previously-uncovered repos; consolidated to audit-report-2026-05-22-rerun.md with 2 new systemic findings (6/8 automation Lambdas + 9/10 study product sub-sites are local-only with no GitHub r...12.0h14m1m51.4x720.0x
5Readiness rerun3 + security audit (5 parallel agents): verified today HIGH fixes clean, agents auto-fixed 9 test failures + 1 real h1->h3 heading-skip a11y bug, surfaced 1 CRITICAL (ElevenLabs key) + 3 HIGH (RDS 3306 open — user fixed; engine pipeline fail; IAM wildcards), 6 inline fixes shipped across an API...35.0h50m3m42.0x700.0x
64 parallel readiness remediation agents: pushed 6 automation Lambdas + an origin service to GitHub, cleaned an infrastructure repo (6 commits — CLAUDE.md, lock files, plan.bin removal, 5 new marketing stacks, tfvars examples), registered an origin service port 8005 + real CodeBuild buildspec, fixed a web clie...14.0h22m1m38.2x840.0x
7Playwright a payment processor-free subscription lifecycle e2e: incomplete→invoice.paid→active w/ entitlement granted, cancel-at-period-end + reactivate, immediate cancel revokes entitlement, service-token gate. Uses test-ops /subscriptions + /webhooks/a payment processor/simulate; full live suite now 10/10 g...10.0h18m1m33.3x600.0x
8Security audit HIGH fix: revoked 0.0.0.0/0 + ::/0 tcp/3306 ingress on prod-ascloud-rds-sg (sg-07e500306cd69710e) — Aurora MySQL no longer reachable from public internet; verified internal app/admin paths still intact via VPC CIDR 10.10.0.0/16 + admin IP 66.182.197.254/32 + self-reference1.0h2m1m30.0x60.0x
9Playwright live-stack e2e suite for AuthModal + enrollment: register-verify-signin, signin happy path, forgot-reset-signin, dup-email error, enrollment + DB-verify, unverified-blocked. Captures emails via a notification service log API, verifies DB user records via an API gateway test-ops, uses Gmail+UUID ali...16.0h34m2m28.2x480.0x
10Phase 5 origin-extraction wiring: discover stub-runner gap, build runtime-to-service DomainSpecification adapter, real synthesis runner + three math content runners (workedexamples, misconceptions, representationpacks), env-gated registration to keep tests green, start an origin service with an inference en...16.0h35m3m27.4x320.0x
11Consolidate auth+purchase under an API gateway gateway and build in-modal auth UI (sign in, register, forgot/reset, MFA TOTP, verify email, Apple/Google social) replacing the hosted OIDC SPA; strip 12 legacy env vars and 14 per-service gateway argument call sites18.0h42m5m25.7x216.0x
12domain-difficulty-factor engine work: 4 decoy fixes (headless default, composite circuit-breaker, maxdays terminal event, catalog status from spec) + foundation-phase + alpha-saturation tuning landed in autopilotranker/orchestrator/autopilot_service; PMP+CAPM spec/manifest patches; boot cache rebuild ×2; en...32.0h90m6m21.3x320.0x
13GAP-06 fix: per-email rate limit on /forgot-password (Redis ZSET sliding window) and per-IP rate limit on /reset-password in an authentication service, with regression tests; 459 tests pass4.0h12m1m20.0x240.0x
14post-PMP-fleet morning session: AZ-500 root-cause (snapshot serializer dropped goalweights/goalsimilarity for entire v3 schema lifetime; engine fell into legacy 0.85 clamp); fixed serialize+deserialize+tensor-dispatch + bumped schema to v4; 6 new round-trip unit tests; audit guardrails for difficulty-on-spe...24.0h90m8m16.0x180.0x
15decoy daily proficiency snapshots — DailySnapshot model, alembic migration, dialect-agnostic upsert in workerpool, EOD autopilot fetch + daycompleted payload expansion in headlessrunner, GET /students/{id}/proficiencyseries endpoint, 9 new tests across workerpool/studentmanager/api8.0h30m1m16.0x480.0x
16Release-test a web client stack: fixed purchase-route DB binding (9 files hitting wrong DB) + rewrote purchase JWT verifier to use local public key (self-JWKS deadlock under single-worker uvicorn); verified end-to-end register/login/entitlements/subscriptions/plans3.0h14m3m12.9x60.0x
17Cloud-wide regression sweep: 44 students across AWS/Azure/GCP/PMP (3-batch parallel via decoy CLI). 43/44 passed; mean predicted 89.8%, mean actual 99.7%, mean gap +9.9pt. PGWA flagged with same empty-goal_weights bug as AZ-500.18.0h90m4m12.0x270.0x
18Compliance L1 (admin-service role check), L2 (audit-log profile updates), M16 (Dependabot for an inference engine + a notification service); 888 tests pass across an authentication service + admin-service; readiness audit dispatched (4 parallel agents); accessibility remediation plan (7 waves, 14-17 eng-days)6.0h35m2m10.3x180.0x
19Readiness blockers H5 (self-assign bug), H6 (eslint-plugin-react-hooks load + Sparkline conditional useEffect fix), H9 (commit infra VPC doc comments); ESLint 13 errors -> 0 errors across an admin client + a desktop client, 3 commits pushed3.0h18m1m10.0x180.0x
20Autonomous blueprint-anchor diagnosis + content-aware re-anchor script (139 domains fixed); full 47-profile confirmation sweep (43/44 passed); PGWA deep-dive identified borderline 74.7% reserved-pool accuracy with weak-goal-biased practice exam as root cause beyond blueprint fix.12.0h180m3m4.0x240.0x
21a marketing site: PMP nav entry with Coming Monday badge, catalog search box with JSON index + JS filter, PMI June dates, refactor templates, build+deploy staging+prod2.5h55m4m2.7x37.5x
22a marketing site PMP card+banner: set available_at, update category/course-page templates to render Available May 25th0.8h18m3m2.5x15.0x

Aggregate Statistics

MetricValue
Total tasks22
Total human-equivalent hours425.2
Total Claude minutes935
Total supervisory minutes57
Total tokens8,145,000
Weighted average leverage factor27.3x
Weighted average supervisory leverage factor447.6x
Human-equivalent weeks10.6

Analysis

The day's leverage distribution matters more than the headline figure. The 161.5x ceiling came from Full an inference engine accessibility audit (50 repos, deterministic Phase 0 + 4 parallel LLM agents, ~288 findings) followed by full compliance audit (12 sect...; the 2.5x floor was a marketing site PMP card+banner: set available_at, update category/course-page templates to render Available May 25th. 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 (447.6x 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 22 tasks, the day produced roughly 10.6 weeks of senior-engineer-equivalent throughput in 15.6 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.