Skip to main content
AI24 July 2026

Documentation and infrastructure

19 tasks. July 24, 2026 closed at 21.7x weighted leverage across 394.0 human-equivalent hours in 1,088 minutes of wall-clock time. Supervisory leverage came in at 295.5x.

That is 9.8 weeks of human-equivalent throughput in 18.1 hours. The ceiling was 77.1x; the floor was 9.0x. 18 of the 19 entries came from a single project.

Task Log

#TaskHuman Est.ClaudeSup.Factor
1Documentation36.0h28m8m77.1x
2Post-crash recovery: root-caused WindowServer abort from crash report + session transcripts, resumed and completed the 411-domain duplication adjudication campaign from checkpoint, built and published the…13.0h15m3m52.0x
3Coding40.0h55m3m43.6x
4Three self-contained HTML doc pages for Colibri git archive (timeline + AI-productivity + finding aid; custom design system; validated palette; 4 SVG charts; dual theme)28.0h40m3m42.0x
5Documentation triage across avian-architecture + avian-architecture-client: classify every doc (keep/archive/merge/delete), create Unbuilt Work Register, move 5 completed campaigns to history, rebuild…28.0h40m5m42.0x
6Doc-vs-code fact-check across avian-architecture: 10 Opus agents verified every architecture doc + all 13 ADRs against real code; applied verified corrections across 6 repos; built pre-commit drift gate;…72.0h150m8m28.8x
7Infrastructure7.0h17m5m24.7x
8avian-api absorbed-architecture cutover: 3-DB prod migration, sandbox Stripe+webhook, 4 deploy-defect fixes (notification key, RESP3 pin, JWT keys, OIDC issuer), plans seed, [unreleased product] migration…20.0h55m4m21.8x
9Widget audit follow-up: diagnosed 12 failing interaction e2e tests across 96 widgets; fixed 3 real component defects (svg-plot 1M-tick tab hang, ForceDiagram asymmetric angle grading, mixed numeral systems…18.0h53m3m20.4x
10Build action.exclude_questions versioned-exclusion ([engine subsystem]; 34 tests; audit sidecar; npz row-subset)5.0h16m4m18.8x
11Build term.case_consistency lexicon+detector in checks-lib (105 tests; sentence-initial forgiveness rule)9.0h31m4m17.4x
12Add dup-scores sidecar checks (score_present; over_threshold) + mcq.answer_length_tell lint to avian-content-checks6.0h25m3m14.4x
13Overnight war-room recovery: crash forensics+guards; Fix2+lint+4 platform fixes landed; 86-domain AP/IB burn to full promotion; 85 widget e2e tests+5 defect fixes; docs remediation; 3 staging deploys with 4…70.0h330m10m12.7x
14Build action.score_duplication ([engine subsystem]; 30 tests; 5-point registry wiring) + ADR-0013 duplication scoring6.0h30m4m12.0x
15Removed 7 policy-violating CI test workflows across 6 repos (delete pure-test, strip test steps from build/publish pipelines); fixed Session Composition dead proxy — 501 + deleted doomed EngineClient methods…9.0h45m3m12.0x
16Wave 1 pilot gate: found and fixed 3 [engine subsystem] hub defects blocking the dup campaign (missing catalog entries, missing job-kind/TTL registration, rebuild_manifest preview ignoring _source_hashes)…9.0h48m2m11.2x
17Engine session dup-exclusion scheduler (3 choke points; flag off; 46 new tests; byte-identical proof)10.0h58m4m10.3x
18Add sched_grp session-exclusion clusters to dup_scores sidecar (union-find; schema 1.1.0; 13 tests)2.0h12m2m10.0x
19chem1b_ttu exam_tips population: TTU exam-guide research (agent), spec enrichment + consolidation-leftover domain_id restamp, worker-pool prereq-flag fix, 3 platform defects found (content-gen store seeding,…6.0h40m2m9.0x

Aggregate Statistics

MetricValue
Total tasks19
Total human-equivalent hours394.0
Total Claude minutes1,088
Total supervisory minutes80
Total tokens9,838,752
Weighted average leverage factor21.7x
Weighted average supervisory leverage factor295.5x
Human-equivalent weeks9.8

Analysis

The highest factor of the day came in at 77.1x and the lowest at 9.0x, a spread of 8.6 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 36.0 of the 394.0 human-equivalent hours, or 9 percent of the day. No single task dominated the total, so the weighted average is representative.

Supervisory time was 80 minutes against 1,088 minutes of execution, a ratio of about 1 to 14. Supervisory leverage of 295.5x 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.