Skip to main content
AI26 July 2026

Coding and deployment

18 tasks. July 26, 2026 closed at 11.8x weighted leverage across 217.0 human-equivalent hours in 1,107 minutes of wall-clock time. Supervisory leverage came in at 183.4x.

That is 5.4 weeks of human-equivalent throughput in 18.4 hours. The ceiling was 48.9x; the floor was 2.3x. 16 of the 18 entries came from a single project.

Task Log

#TaskHuman Est.ClaudeSup.Factor
1Documentation22.0h27m7m48.9x
2Design and frontend16.0h32m3m30.0x
3Deployment20.0h50m4m24.0x
46AM staging content deploy automation — staging-snapshot-deploy.sh + launchd agent + installer + docs (TCC-aware; shellcheck/plutil/dry-run validated)6.0h18m3m20.0x
5Lane 1 onesies repair: 6 families fixed w/ evidence (pair coverage 19-24pct to 100pct x4), 2 waived, 6 catalog gaps root-caused, 4 [engine subsystem] tooling bugs found19.0h65m5m17.5x
6Deploy-tooling migrations gap — fargate migrate support + gated staging-migrate.sh (700 lines) + docs; caught [unreleased product] chain 4 behind8.0h33m2m14.5x
7Testing6.0h25m2m14.4x
8Coding34.0h150m2m13.6x
9Full GO program: ship difficulty ladder + wave sizing + overtime grace + dynamic sprint clock + sprint-in-autopilot + convoy interleaving + endless afterburner + [ip-cluster] injection; web+engine prod…30.0h140m5m12.9x
10Infrastructure9.0h45m3m12.0x
11Engine boot cross-domain vector collision fix — if_exists=skip on auto-load path + 6 regression tests (10773 passed 93.14% cov)4.0h25m2m9.6x
12Lane 6 LLM-content repair: 3 catalog gaps root-caused via source, 31 domains repaired/verified, spend recon for 4 big families, [cost]actual9.0h80m5m6.8x
13Lane 5 goal-remap triage: Pre-Algebra spec-graft root cause via git archaeology, tier-normalization false positive found, cs.goal_linkage mechanism validated + 105 domains dispatched6.0h60m5m6.0x
14Lane 4 dedupe/normalize: cross-lane version-race root-caused w/ evidence, dedupe_flashcards validated, 4 families source-verified + chunked8.0h80m5m6.0x
15Catalog-overrides precedence fix (web+electron) + KaTeX/t-deps component defect + electron pre-commit gate — 16 tests6.0h65m2m5.5x
16Lane 2 manifests/packaging: 141/141 rebuilds verified clean, frozen packages-tree + stray-classes gaps found, materializer race diagnosed8.0h105m5m4.6x
17Fleet content audit re-run 411/411 + findings analysis + AP French/German candidacy-gate diagnosis + hub restart4.0h54m5m4.4x
18Land orphaned backfill_node_embeddings action: [engine subsystem] worker action + [engine subsystem] hub registration, 2 suites green, both repos pushed2.0h53m6m2.3x

Aggregate Statistics

MetricValue
Total tasks18
Total human-equivalent hours217.0
Total Claude minutes1,107
Total supervisory minutes71
Total tokens7,313,025
Weighted average leverage factor11.8x
Weighted average supervisory leverage factor183.4x
Human-equivalent weeks5.4

Analysis

The highest factor of the day came in at 48.9x and the lowest at 2.3x, a spread of 21.6 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 22.0 of the 217.0 human-equivalent hours, or 10 percent of the day. No single task dominated the total, so the weighted average is representative.

Supervisory time was 71 minutes against 1,107 minutes of execution, a ratio of about 1 to 16. Supervisory leverage of 183.4x 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.