Skip to main content
AI MAY 15, 2026

Leverage Record: May 15, 2026

19 tasks. May 15, 2026 weighted to 21.1x leverage across 378.0 human-equivalent hours in 1075 Claude-minutes. Supervisory leverage closed at 238.7x.

19 tasks. May 15, 2026 weighted to 21.1x leverage across 378.0 human-equivalent hours in 1075 Claude-minutes. Supervisory leverage closed at 238.7x.

9.4 weeks of human-equivalent throughput in 17.9 hours of Claude wall-clock. The 73.8x ceiling came from an Android client repo skeleton: README, CLAUDE.md, and four parity docs (requirements, design, design-system, testing-strategy) translating the iOS Swift/SwiftUI client to Kotlin/...; the 7.2x floor sat at Recovered 5 misdirected re-synth packages (scripts/data/domains -> data/domains); diagnosed and fixed engine bug at loop.py:460 (prevalidate_nodes string-not-dict crash) mirrorin....

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
1an Android client repo skeleton: README, CLAUDE.md, and four parity docs (requirements, design, design-system, testing-strategy) translating the iOS Swift/SwiftUI client to Kotlin/Compose/AppAuth/Wear OS16.0h13m2m73.8x480.0x
2an Android client Phase 3 data layer: EngineApi (single Retrofit interface, all endpoint groups), 7 DTO files, EngineClient facade with HttpException/SerializationException/IOException → EngineError mapping, EngineError sealed class, AuthInterceptor + TokenProvider, EngineEventClient (OkHttp WebSocket → Flow<...30.0h28m1m64.3x1800.0x
3an Android client Phase 8 dashboard + rings: DailyRingsState + RingTargets, DailyRingsRollover (pure-function rules with 7 test cases), DailyRingsStore (SharedPreferences-backed with StateFlow + recordAnswer/recordActivity/rolloverIfNeeded), DailyRingsModule, DailyRingResetWorker (HiltWorker periodic 1-day fl...22.0h25m1m52.8x1320.0x
4an Android client Phase 2 domain logic: LcsDiff (with iOS-bug fix), DeterministicShuffle (DJB2+Mulberry32+Fisher-Yates), BehavioralRingsComputation + RingArc + RingConstants + BehavioralRings, ProficiencyColor + TimedRecallTimer, Base64Url with PKCE helpers (RFC 7636 verified), full AppModels with kotlinx-ser...14.0h16m1m52.5x840.0x
5an Android client Phase 7 onboarding + initialization: OnboardingViewModel (3-step state machine with DeterministicShuffle-seeded calibration quiz, 8-question SAMPLE_BANK, SavedStateHandle restoration), OnboardingScreen (tier cards + progress-tracked quiz + completion), KnowledgeTier 5-tier enum, Initializati...14.0h16m1m52.5x840.0x
6an Android client Phase 6 catalog + exam info: CatalogViewModel (StateFlow combine + EngineError-to-message mapping), DomainCatalogScreen (adaptive LazyVerticalGrid 1/2/3 cols, badges, top app bar with refresh + sign-in/profile, loading/empty/error states), ExamInfoViewModel (SavedStateHandle for domainId), E...18.0h21m1m51.4x1080.0x
7an Android client Phase 4 authentication: TokenStore + EncryptedTokenStore (AES-256-GCM Keystore), PendingEnrollmentStore + Encrypted impl, PkceVerifierStore + Encrypted impl with 5-min TTL, OidcConfig, OidcAuthService (AppAuth Custom Tabs orchestration with suspend code-exchange), AuthResult sealed class, Au...16.0h19m1m50.5x960.0x
8an Android client Phase 5 app shell + state machine: 28-state Phase sealed interface (all @Parcelize), ActivityModeKey + WatchPhase, AppState, AppStateHolder (StateFlow Singleton), AppViewModel (HiltViewModel with SavedStateHandle restoration + auth bootstrap + startStudying decision + handleBackPressed), Pha...14.0h17m1m49.4x840.0x
9an Android client Phase 1 design system: HslColor + an inference engineColorScheme (light + dark, 1:1 parity with web tokens.css), an inference engineBrand runtime accent override, an inference engineTokens public surface (Composable getters + Spacing/Radius/Motion/TapTarget/Elevation/FontSize), an inference...18.0h22m1m49.1x1080.0x
10an Android client Phase 0: phased build plan (18 phases) + Gradle multi-module skeleton (app/wear/design-system/domain/data/testing), Kotlin 2.0 + AGP 8.5 + Compose BOM, Hilt+KSP, version catalog, Hilt Application + Compose MainActivity for phone+Wear, manifest with OIDC + App Link intent-filters, network-sec...12.0h18m1m40.0x720.0x
11Two funding-strategy documents (pre-revenue SAFE path and growth-bridge + priced-seed path) covering consumer + a recruiter product + enterprise markets with branded PDFs16.0h28m8m34.3x120.0x
12an inference engine: retire @pytest.mark.slow tests, add 30s default timeout + pristine RNG seeding, lift 14 of 16 packages to >=85% unit-test coverage with 1,342 new fast tests across 20 files (5,010 pass / 0 fail / 75s wall-clock, pristine across 3 back-to-back runs). Built per-module coverage gate, fixed L...80.0h240m15m20.0x320.0x
13Third funding-plan variant (SAFE + 2 equity-comp founding hires + native Android September 2026); PDF tooling improvements (DOC_DATE override, H2 page-break removal)6.0h22m6m16.4x60.0x
14an inference engine Phase 3 service-layer extraction: 16 endpoints across 12 service modules (sequencing, interaction, atom_service compose v1+v2, autopilot lifecycle/create/composite/list-due, operations+telemetry batch, entity self-report + seed-from-mastery, session audio/hint/end/upload-resume/next-challe...32.0h130m4m14.8x480.0x
15a simulation harness: audited UI vs post-April app-web rebuild, fixed Postgres auth + 22 stuck workers, added 4 frontend polish fixes (SSE wiring, sidebar grouping, cloud filter, per-provider calibration facet), remapped 8 Playwright page objects (onboarding, dashboard, exam, mcq, library, session_config, aut...22.0h95m6m13.9x220.0x
16Diagnosed pair-orphan engine bug (regenerate_nodes returned only new nodes, caller looked up stale pairs by NEW id; pairs hold OLD id so intersection always empty); fixed signature + caller, added regression test #19; archived 9 affected packages; relaunched orphan-fix batch with 3-way parallel concurrency; s...5.0h35m5m8.6x60.0x
17Consolidated advisor-ready funding plan 02c (5-person team, $6.5M SAFE, profit-sharing, patent-adjusted valuations, 5-year comp tables) plus HoRO/CFO + Marketing Director job description PDFs32.0h240m35m8.0x54.9x
18Patent + diagram audit clean-up bundle for 7 follow-on filing working drafts (working drafts): reverted regression edges to deterministic-classifier verdicts (a follow-on FIG 2, a follow-on FIG 7), fixed EE 710 cross-fig conflict (subgraph carries numeral, PROV unnumbered, spec corrected), added success-path...8.0h65m2m7.4x240.0x
19Recovered 5 misdirected re-synth packages (scripts/data/domains -> data/domains); diagnosed and fixed engine bug at loop.py:460 (prevalidate_nodes string-not-dict crash) mirroring synthesizer/engine.py defensive coercion; added regression test #18; relaunched resume batch (3 syntheses + 5 QBs)3.0h25m3m7.2x60.0x

Aggregate Statistics

MetricValue
Total tasks19
Total human-equivalent hours378.0
Total Claude minutes1075
Total supervisory minutes95
Total tokens5,130,000
Weighted average leverage factor21.1x
Weighted average supervisory leverage factor238.7x
Human-equivalent weeks9.4

Analysis

The day's leverage distribution matters more than the headline figure. The 73.8x ceiling came from an Android client repo skeleton: README, CLAUDE.md, and four parity docs (requirements, design, design-system, testing-strategy) translating the iOS Swift/Swift...; the 7.2x floor was Recovered 5 misdirected re-synth packages (scripts/data/domains -> data/domains); diagnosed and fixed engine bug at loop.py:460 (prevalidate_nodes string-not-.... 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 (238.7x 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 19 tasks, the day produced roughly 9.4 weeks of senior-engineer-equivalent throughput in 17.9 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.