Thirteen tasks. May 2, 2026 weighted to 17.0x leverage across 395.0 human-equivalent hours in 1393 Claude-minutes. Lab simulator dominated the day's volume. Supervisory leverage closed at 353.7x.
The day's ceiling was 60.0x (24h human in 24 Claude-minutes) on Embedded subscribe flow: PaymentIntent backend + recompute + coupon validate endpoints, new shared @the platform/subscribe-react lib (FlipCard primitive, Subscr. The floor was 3.4x on Diagnose remaining engine 401s as a frontend bug (engine.ts had its own raw fetch() bypassing Authorization header), audit all api/*.ts callsites, fix engine.ts. Median Claude-minutes per task: 35; median human-equivalent hours per task: 8.
Task Log
| # | Task | Human Est. | Claude | Sup. | Factor | Sup. Factor |
|---|---|---|---|---|---|---|
| 1 | Embedded subscribe flow: PaymentIntent backend + recompute + coupon validate endpoints, new shared @the platform/subscribe-react lib (FlipCard primitive, SubscribeFront/Back, EmbeddedSubscribeFlow, useSubscriptionIntent hook), web client migration to lib, popup checkout removed | 24.0h | 24m | 6m | 60.0x | 240.0x |
| 2 | Cloud-cert lab Batch 5 wave 1: a cloud cert exam cluster shared root cause diagnosis (stale sidebarTarget bug) lifted all 19 a cloud cert exam labs in one shot, plus partial progress on a cloud cert exam/a cloud cert exam/a cloud cert exam. 26 newly strict-pass; 32 improved-but-still-partial. Sub-agent across 1 wave. 931 -> 957 ship-ready. | 52.0h | 120m | 3m | 26.0x | 1040.0x |
| 3 | Operator-managed banners across admin dashboard/API gateway/web client with markdown CRUD, semantic color variants, audience targeting from purchase state, push-to-all-clients SSE broadcast, dismissal tracking, and CTA actions (subscribe modal + URL) | 28.0h | 65m | 6m | 25.9x | 280.0x |
| 4 | Diagnose hard hang from disk-full + memory exhaustion (Docker VM + TM + Spotlight + Maestral); archive 123 GB training chunks to new S3 bucket with byte-exact verification | 3.0h | 7m | 4m | 25.7x | 45.0x |
| 5 | Cloud-cert lab Batch 3: lifted 100 closest-to-passing labs (gap 5-30) to strict-pass via sub-agent across 11 waves. Service-cluster fixes on a cloud cert exam, a cloud cert exam, a cloud cert exam, a cloud cert exam, a cloud cert exam, plus per-lab fixes. Dashboard testIds added (bedrock onConfirm, apigateway Deploy modal, vpc CreateSgModal, cloud-spanner instance select, ResourceLocks). 706 -> 806 ship-ready. | 100.0h | 293m | 5m | 20.5x | 1200.0x |
| 6 | Add purge + bulk-purge for revoked comps across billing service, admin-service WS, and admin dashboard UI with tests | 4.0h | 12m | 3m | 20.0x | 80.0x |
| 7 | Cloud-cert lab Batch 4: harder set of 100 partials (gap 20-40, AWS 28 / Azure 24 / GCP 48) lifted to strict-pass via sub-agent across 7 waves. Heavy dashboard work to add testIds for missing modal flows. 810 -> 931 ship-ready (+104 with incidental wins from dashboard changes). | 150.0h | 577m | 5m | 15.6x | 1800.0x |
| 8 | Refactor API gateway gateway to contextvar-scoped JWT forwarding (durable, multi-instance safe, opt-in per upstream); engine + purchase now propagate user JWT; defect tracker stays service-only; regression tests both directions; fix 2 pre-existing ruff failures | 5.0h | 30m | 4m | 10.0x | 75.0x |
| 9 | Diagnose amplifier KeyError engine bug (a content generator returns pairs as dict not list, KeyError: 0/slice) and fix with coercepairslist helper applied at 4 sites; build ISACA recovery batch orchestrator (runisacarecovery.sh, wipe-and-rerun semantics for 21 specs after engine fix); free-tier batch prep (7 Security depthconstraints normalization, 1 Insider Threat 7-dup-goal-id rename, 1 AP Micro proficiencytier label backfill, 12 SecurityAwareness flipped to freetier:true, 22 specs flipped to tribunalverdicts:true); Tableau→Salesforce category merge (1 spec moved, 3 docs updated, activitiescatalog.json updated); build free-tier batch orchestrator (runfreetier_close.sh, 11 waves, full tribunal pipeline) | 12.0h | 90m | 12m | 8.0x | 60.0x |
| 10 | Fix the marketing site console errors: observability tool-rum.js schema mismatch (400), missing favicon (403), and tailwind CDN production warning. Rewrote observability tool-rum SDK to canonical backend schema, deployed via observability tool-frontend pipeline (verified live with 202 ingest). Added temp favicon to _shared-the product overlay. Replaced cdn.tailwindcss.com + inline config with a real Tailwind 3 compile pipeline (tailwind.config.js, build-tailwind.sh, compiled CSS) and rebuilt the marketing site dist locally. | 4.0h | 35m | 4m | 6.9x | 60.0x |
| 11 | Add Rime TTS as alternate provider; add voice/speed picker in Settings with Play audition; wire labs to user-selected voice; deploy backend + frontend + infra to production | 8.0h | 75m | 6m | 6.4x | 80.0x |
| 12 | Engine dual-auth: accept JWTs from authentication service JWKS alongside static a feature flag (jwtauth module port of gateway core/jwt.py + middleware refactor in restgateway.py + tests) | 3.0h | 30m | 4m | 6.0x | 45.0x |
| 13 | Diagnose remaining engine 401s as a frontend bug (engine.ts had its own raw fetch() bypassing Authorization header), audit all api/*.ts callsites, fix engine.ts request()+lesson-audio+evidence-audio paths to use loadTokens(); flag competitive.ts and bugs.ts as same-pattern issues | 2.0h | 35m | 5m | 3.4x | 24.0x |
Aggregate Statistics
| Metric | Value |
|---|---|
| Total tasks | 13 |
| Total human-equivalent hours | 395.0 |
| Total Claude minutes | 1393 |
| Total supervisory minutes | 67 |
| Total tokens | 2,684,000 |
| Weighted average leverage factor | 17.0x |
| Weighted average supervisory leverage factor | 353.7x |
Analysis
The day's leverage distribution is the part that matters more than the headline figure. 1 task cleared the 30x threshold; 1 task ran below 5x. The 30x+ tier is what produces the impression that AI changes the time-cost curve; the sub-5x tier is what reminds anyone watching that some work is still gated by human review and cannot speed up arbitrarily.
Top-of-distribution tasks tend to share a shape: tightly-scoped, well-specified, with no integration ambiguity. On May 2, 2026 the 60.0x ceiling came from Embedded subscribe flow: PaymentIntent backend + recompute + coupon validate endpoints, new shared @the platfo. The work fit cleanly into 24 Claude-minutes because the inputs and the success criterion were both explicit; the AI was not required to discover anything new. That shape is repeatable; tasks like it post 30x to 60x consistently across the recent log.
Bottom-of-distribution work runs differently. The 3.4x floor on Diagnose remaining engine 401s as a frontend bug (engine.ts had its own raw fetch() bypassing Authorization he reflects real human review per checkpoint, often serial because each step depends on the previous one. The supervisory ratio (354x weighted today) tracks differently: it captures how much human prompt-writing time the day's output consumed, and it stays high even on lower-leverage days because supervisory minutes scale roughly with task count, not with human-equivalent hours.