Skip to main content
AI MAY 25, 2026

Leverage Record: May 25, 2026

Nine tasks. May 25, 2026 weighted to 25.3x leverage across 59.5 human-equivalent hours in 141 Claude-minutes. Supervisory leverage closed at 142.8x.

Nine tasks. May 25, 2026 weighted to 25.3x leverage across 59.5 human-equivalent hours in 141 Claude-minutes. Supervisory leverage closed at 142.8x.

1.5 weeks of human-equivalent throughput in 2.4 hours of Claude wall-clock. The 33.6x ceiling came from Synthesis pipeline: prompt caching + Anthropic Batches API integration across synthesis scripts in core/an inference engine (1649 LOC, 5 files); the 12.0x floor sat at core/an inference engine: autopilotservice legacy coverage-damping ceiling lifted + bulkamplifyfleet and bulkbackfillrecall customid format fix with error logging.

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
1Synthesis pipeline: prompt caching + Anthropic Batches API integration across synthesis scripts in core/an inference engine (1649 LOC, 5 files)28.0h50m5m33.6x336.0x
2Investigate and backfill missing a metrics tracker leverage records since May 22: surveyed git logs across 80+ an inference engine repos, identified 7 commit-clusters across May 23-25, diagnosed root cause (process discipline gap, not infra), reconstructed and POSTed 7 records to a metrics tracker-api4.0h8m4m30.0x60.0x
3Audit 29 ~/.claude/skills/ manifests for missing leverage-POST step on free-form task path; identify 16 tool-loader skills (a marketing platform, a calendar platform, a knowledge base, an email platform, a defect tracker, a portfolio browser, a metrics tracker, a newsletter platform, a time-tracking app, a CM...3.0h6m2m30.0x90.0x
4Invert leverage tracking policy: CSV first then cloud second both mandatory; patched global CLAUDE.md Rules block, /fix skill Step 6k, 16 tool-loader Step 4 blocks, and /leverage-post Phase 2 reconciliation4.0h8m2m30.0x120.0x
5/leverage-post reconciliation Phase 1+2: backfilled 139 CSV rows across 12 days (5/14-5/25), verified all in sync, 0 stragglers remaining1.5h4m1m22.5x90.0x
6core/a simulation harness rebuild: brain answerer switched to direct Anthropic SDK with prompt caching (257 LOC), all zero/pmp sweep profiles flipped to omniscient:false (46 files), headless runner surfaces non-200 from /next-pair-mcq with context10.0h30m4m20.0x150.0x
7Cross-fleet prompt-caching micro-sweep: cachecontrol on a relationship CRM sonnet system block, an API gateway a recruiter product/llmnormalizer streaming call, automation-resume-refinement SYSTEMPROMPT, an origin service atoms/generator system + toolschema3.0h10m2m18.0x90.0x
8an audit toolchain: content-audit P4.1 pair-density check with PGWA-class detection (61 LOC), a configuration file headline counts bumped to 2026-05-25 audit snapshot, per-activity-format trackers added (scenarios, flashcards, etc.)4.0h15m3m16.0x80.0x
9core/an inference engine: autopilotservice legacy coverage-damping ceiling lifted + bulkamplifyfleet and bulkbackfillrecall customid format fix with error logging2.0h10m2m12.0x60.0x

Aggregate Statistics

MetricValue
Total tasks9
Total human-equivalent hours59.5
Total Claude minutes141
Total supervisory minutes25
Total tokens928,000
Weighted average leverage factor25.3x
Weighted average supervisory leverage factor142.8x
Human-equivalent weeks1.5

Analysis

The day's leverage distribution matters more than the headline figure. The 33.6x ceiling came from Synthesis pipeline: prompt caching + Anthropic Batches API integration across synthesis scripts in core/an inference engine (1649 LOC, 5 files); the 12.0x floor was core/an inference engine: autopilotservice legacy coverage-damping ceiling lifted + bulkamplifyfleet and bulkbackfillrecall customid format fix with error.... 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 (142.8x 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 9 tasks, the day produced roughly 1.5 weeks of senior-engineer-equivalent throughput in 2.4 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.