Skip to main content
AI31 July 2026

Coding and audit and review

44 tasks. July 31, 2026 closed at 14.9x weighted leverage across 1,253.0 human-equivalent hours in 5,032 minutes of wall-clock time. Supervisory leverage came in at 414.2x.

That is 31.3 weeks of human-equivalent throughput in 83.9 hours. The ceiling was 160.0x; the floor was 2.8x. 22 of the 44 entries came from a single project.

Task Log

#TaskHuman Est.ClaudeSup.Factor
1Quorum: name + scaffold a new fleet tool for corporate governance (board meetings, quorum, resolutions, minutes, e-signature, stock ledger); researched and selected self-hosted LiveKit track-egress for…40.0h15m3m160.0x
2Testing120.0h70m3m102.9x
3Documentation28.0h32m6m52.5x
4Deployment38.0h48m3m48.0x
5Coding30.0h40m1m45.0x
6Deployment32.0h55m7m34.9x
7AVIAN documentation consolidation comprehensive plan (census + plan doc)8.0h14m5m34.3x
8Deployment52.0h95m4m32.8x
9Deployment46.0h90m2m30.7x
10avian-app-web: nav chrome root-cause fixes (undefined DS layout tokens), avatar->profile link + design-system 0.4.4 publish, page padding sweep, prod bug-reporter allowlist, single-domain product nav/docs…26.0h51m7m30.6x
11Deployment34.0h80m2m25.5x
12Infrastructure22.0h55m1m24.0x
13Documentation18.0h46m4m23.5x
14Design and frontend28.0h75m5m22.4x
15Coding26.0h70m6m22.3x
16Phase 1b generation contract: prompt quality contract (both MCQ prompts), generator_model attribution (questions+lessons), mercury MCQ bar, lazy multi-provider escalation ladder (terra→grok→sol) + 12 tests +…12.0h35m1m20.6x
17Remediate all 8 accessibility audit findings in avian-app-web: 2 design-system releases (modal accessible-name + Avatar role), new authenticated axe sweep (18 routes) that caught a serious latent defect,…80.0h240m3m20.0x
18Audit and review22.0h72m4m18.3x
19Documentation16.0h55m2m17.5x
20Deployment20.0h78m8m15.4x
21Root-caused + fixed 33 unloadable [pipeline stage] (760 orphaned nodes): diagnosed stale materialization not missing embeddings, flipped the WP2 packages/ gate with regression tests, 2 materialize passes to…24.0h95m5m15.2x
22Wire [engine subsystem] worker daemon onto models.yaml registry (category/content-type routing + key_env-first credentials + mercury amplifier flip + tests + fleet restart)12.0h50m3m14.4x
23Staging defect batch: 15 reported issues across avian-engine/avian-api/avian-app-web + 2 shared libs — root-caused, fixed with regression tests, published libs, deployed and verified on staging40.0h178m6m13.5x
24Deployment160.0h732m12m13.1x
25Testing4.0h19m2m12.6x
26Content production64.0h310m5m12.4x
27Handoff block P1-P3: parity-check root-cause fix (id-list vs vector rows) + incident regression tests + COMPLETE contract; f675fc77 store-safe manifest projection; pointer-lag investigation (410/412, median…20.0h105m3m11.4x
28Phase 2 block: client-wedge root cause + never-wired llm_hard_call_timeout_s + client.py 85→95% wedge-guard battery; reindex exit-139 root-caused (faiss/torch dup libomp, both orders reproduced) +…22.0h120m1m11.0x
29Prod content-DB cutover + deterministic receipted release: restore content_r4 to prod RDS + content contract system + 5 release-script fixes + verify-only auditor + verdaccio-only lockfile migration +…28.0h155m12m10.8x
30Audit and review14.0h80m3m10.5x
31Phase 3 block: llm_content_pool_size knob; #17 caching/batching audit (agent-swept 39 files) + 2 cache-violation fixes + speed-program blueprint card; [engine subsystem] MCP admin-key confirmed; registry…8.0h50m0m9.6x
32Purge fixture-wp6-smoke registry row (90 rows across 7 tables) + canonical-UUID domain-id guard + regression tests4.0h25m1m9.6x
33Deployment14.0h95m5m8.8x
34Prod outage: app.accelastudy.ai was serving a staging-mode bundle (login broken 5 days) — root-caused, fixed CSP, cherry-pick-promoted the defect batch to prod, fixed 2 prod-release.sh defects, deployed and…7.0h48m3m8.8x
35Deployment30.0h213m4m8.5x
36Provision classroom prod (ungated) + hyphenated staging from scratch; build/deploy both; capture hero; fix capture script DNS and load-wait bugs4.0h32m2m7.5x
37classroom.app host + banner overlap/light-mode fixes + timeline 422; caught and restored a prod API outage (desiredCount=0) via the new deploy smoke; reverse-merged main->staging across 7 repos9.0h72m4m7.5x
38Deployment6.0h50m2m7.2x
39Content production5.0h42m4m7.1x
40Infrastructure8.0h75m3m6.4x
41Benchmark fail-open forensics + fail-closed guards (checks-lib/[engine subsystem]) + Temporal ceiling fix + 5 model registrations + 4-set Luna campaign rerun + comparison16.0h155m3m6.2x
42Promoted US Citizenship to live and shipped it to production: [engine subsystem] registry promote (admin REST, MCP operator key 403s), spec+manifest status flip preserving the immutable store, surgical…4.0h40m2m6.0x
43US Citizenship package to CANDIDATE: corpus surgery (7-goal [engine subsystem] via synthetic checkpoint) + 78/78 lessons via escalation ladder + question regen/reanchor + pair [engine subsystem] + scenarios +…40.0h720m15m3.3x
44Infrastructure12.0h255m4m2.8x

Aggregate Statistics

MetricValue
Total tasks44
Total human-equivalent hours1,253.0
Total Claude minutes5,032
Total supervisory minutes182
Total tokens25,139,000
Weighted average leverage factor14.9x
Weighted average supervisory leverage factor414.2x
Human-equivalent weeks31.3

Analysis

The highest factor of the day came in at 160.0x and the lowest at 2.8x, a spread of 56.7 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 40.0 of the 1,253.0 human-equivalent hours, or 3 percent of the day. No single task dominated the total, so the weighted average is representative.

Supervisory time was 182 minutes against 5,032 minutes of execution, a ratio of about 1 to 28. Supervisory leverage of 414.2x 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.