Skip to main content
AI18 July 2026

Mostly infrastructure

19 tasks. July 18, 2026 closed at 34.3x weighted leverage across 1,143.0 human-equivalent hours in 1,998 minutes of wall-clock time. Supervisory leverage came in at 672.4x.

That is 28.6 weeks of human-equivalent throughput in 33.3 hours. The ceiling was 87.1x; the floor was 3.3x. 19 of the 19 entries came from a single project.

Task Log

#TaskHuman Est.ClaudeSup.Factor
1Build all 102 physics-cluster interactive simulations on the generalized domain-agnostic harness (build->verify workflow, physics authoring contract with real constants + numerical integration,…370.0h255m3m87.1x
2Build all 139 biology-cluster interactive simulations on the generalized domain-agnostic harness (biology authoring contract with real base-pairing/codons/Nernst/Hardy-Weinberg, rive/lottie->procedural…500.0h350m3m85.7x
3AVIAN root cleanup: 14-agent triage of 11 root docs vs codebase + 5-agent root-file dependency mapping; 4 designs converted to artifacts, 4 plans moved into repos, 6 files deleted (verified); refs repointed…20.0h50m3m24.0x
4Phase 8.1 npm-workspace elimination: Verdaccio registry stood up in supporting-services (pinned image + config + idempotent publish-libs.sh w/ frozen historical versions), 13 pkg versions published, 19…32.0h105m3m18.3x
5Feature-flag cleanup: cut 16 stale/dead+retire flags across 6 repos (gateway/web/android/supporting-services/[engine subsystem]/engine) with per-flag consumer checks + suite runs; reviewed 7 ADR ship-dark…16.0h55m4m17.5x
6Deployment10.0h35m4m17.1x
7CHEM1B web-serving audit: 10 gaps (2 engine inference bugs + catalog typo + unreachable sims/cases/scenarios + dead exam tips)18.0h65m4m16.6x
8Coverage agent B: api secondary +659 statements (214 tests; composite exam_metadata bugfix)28.0h105m5m16.0x
9WP-3.5 electron Direct-To parity (unit rail/chip/nudge/badges) + NextActionsResponse bug fixed in 2 places (53 tests)14.0h55m4m15.3x
10Phase 8.3 root CLAUDE/AGENTS stub-shrink (content -> docs/avian-architecture project-reference docs, @import stub + plain stub) + full Verdaccio pipeline proof: publish script hardened (legacy-tag retry,…6.0h25m2m14.4x
11WP-3.4 web Direct-To UI: UnitRail/FocusChip/nudge/badges + advance passthrough + NextActions shape bugfix (53 tests)22.0h100m5m13.2x
12CHEM1B content repair: 389 validated questions (9 goals; 99.74% pass) + combined gas law lesson + exam tips + corpus v42 pipeline + engine reload20.0h100m8m12.0x
13Deployment24.0h120m12m12.0x
14Catalog schema completeness (sims/cases/scenarios fields) + loud categoryKey validation (15 tests; surfaced 15-course drift)6.0h33m4m10.9x
15CHEM1B serving fixes: manifest-first category/tier + sims/case/scenario grants + exam-tips route+UI across 4 repos20.0h115m15m10.4x
16WP-2.5 electron parity: Position Fix flow + IPC 5-layer port + QA gate + tips card (514 tests; found domainId wiring bug)7.0h70m5m6.0x
17Prevention system: per-goal coverage checker + assembler density backstop + [engine subsystem] invariant + fleet census (11 true zero-question instances) + ChemFund v42 local serve (49 tests)21.0h225m6m5.6x
18Engine /domains/load symlink containment + idempotency verification; golden harness 8/8 (12 tests)4.0h45m4m5.3x
19Console-error fixes (TTS wrong-origin 404, bug-report 401, banners 502 degradation) + full activity-type vocabulary inventory (476 usages, 2 lists) + ship-now planSession compat fix so study-plan…5.0h90m8m3.3x

Aggregate Statistics

MetricValue
Total tasks19
Total human-equivalent hours1,143.0
Total Claude minutes1,998
Total supervisory minutes102
Total tokens68,892,048
Weighted average leverage factor34.3x
Weighted average supervisory leverage factor672.4x
Human-equivalent weeks28.6

Analysis

The highest factor of the day came in at 87.1x and the lowest at 3.3x, a spread of 26.1 times between the two. That is a wide range for a single day, and it usually means the day mixed mechanical work with work that needed real judgement.

The largest single entry accounted for 370.0 of the 1,143.0 human-equivalent hours, or 32 percent of the day. No single task dominated the total, so the weighted average is representative.

Supervisory time was 102 minutes against 1,998 minutes of execution, a ratio of about 1 to 20. Supervisory leverage of 672.4x is the figure I find most honest, because it measures the hours I actually spent rather than the hours a machine spent on my behalf.

Every figure here is recorded at the time the work is done rather than reconstructed afterwards. The human estimate is my own judgement and carries the uncertainty that implies; the minutes and tokens are measured. The full dataset, including this day, is available for download.