Skip to main content
AI03 August 2026

Coding and deployment

31 tasks. August 3, 2026 closed at 15.3x weighted leverage across 637.0 human-equivalent hours in 2,492 minutes of wall-clock time. Supervisory leverage came in at 444.4x.

That is 15.9 weeks of human-equivalent throughput in 41.5 hours. The ceiling was 48.8x; the floor was 3.5x. 25 of the 31 entries came from a single project.

Task Log

#TaskHuman Est.ClaudeSup.Factor
1[engine subsystem] publish integrity: Meta [restricted] store fix + 98-domain registry repair + candidacy/publish-on-promote enforcement + prod orphan deletion w/ parity gate (V1 Opus agent)100.0h123m2m48.8x
2Debugging20.0h28m2m42.9x
3Coding24.0h41m3m35.1x
4[engine subsystem] [content generation] quality: fleet bank audit (10 checks/1.56M items) + unconditional write-time gates + prompt/judge fixes, 91 tests (P1 Opus agent)28.0h49m2m34.5x
5Infrastructure14.0h25m2m33.6x
6Deployment30.0h60m3m30.0x
7Deployment16.0h34m3m28.2x
8Deployment32.0h69m3m27.8x
9Deployment13.0h33m4m23.6x
10Deployment40.0h103m2m23.3x
11Coding7.0h19m2m21.8x
12Coding7.0h25m3m16.8x
13Diagnose + fix prod admin OAuth login failure (fleet_auth missing redirect_uri); seed fix + regression tests; branch sync + deploy avian-admin6.0h22m4m16.4x
14Documentation64.0h239m8m16.1x
15Deployment40.0h150m3m16.0x
16Engine boot halved + blue/green made permanent + rollback proven: root-caused TWO independent causes of a duplicate 7GB snapshot download (an unused gRPC server building a second engine context, and a 'cheap…24.0h98m2m14.7x
17[engine subsystem] actions+observability: regen fail-loud, regen_items concurrency, telemetry, knobs, env fingerprint, standalone-readability live-wired, 9 cards closed (P2 Opus agent)23.0h100m2m13.8x
18Debugging4.5h22m2m12.3x
19[engine subsystem] robustness: benchmarks fail-closed + real Luna rerun ([cost]) + leases/workers/SpecIndex/resolver/reclamation + 2 forensic investigations + auto-publish armed w/ 64,930-row proof (V2 Opus…37.0h194m2m11.4x
20Engine boot halved again + embedding model baked into the base image (zero HuggingFace calls at boot) + boot ceiling to 180 + investigated and closed a real data-staleness gap in blue/green: shutdown flush…20.0h111m2m10.8x
21Client quick strikes: plan-slug subscription gate + prod web deploy verified + help-center canon/i18n + electron parity (F1 Opus agent)16.0h91m2m10.5x
22DMCA takedown response: identify + [restricted] entire Meta Blueprint catalog, purge 60+ objects across 4 S3 buckets, rebuild/redeploy 3 sites + web app, fix provider [restricted] gate with regression tests,…14.0h126m4m6.7x
23Engine content correctness: embeddings id-binding fix + fleet npz audit (410/410 clean) + non-MCQ inventory (E1 Opus agent)12.0h110m2m6.5x
24Audit and review5.0h47m3m6.4x
25deploy.sh digest assertions across all three container handlers (fargate/lambda/ec2-container) with task-def pinning and immutable deploy tags, verified live on avian-api; Valkey-vs-NATS replication transport…9.0h104m2m5.2x
26Coding1.5h21m2m4.3x
27Deployment5.0h72m3m4.2x
28Deployment4.0h58m4m4.1x
29Fixed deploy.sh silently shipping stale code (task-def pinned an immutable tag while deploy pushed :latest; force-new-deployment re-pulled the old image and verify only curled health) - added task-def…14.0h205m3m4.1x
30Grant permanent all-access comps (karl/thomas/charles) after finding none existed; fix revenue charts plotting every day one day early (UTC-midnight parse) + 8 tests; trace prod revenue to an E2E test payment3.0h45m3m4.0x
31Fix avian-admin Health page crash (snake_case API mapping + regression tests) and engine-proxy: missing ENGINE_ADMIN_KEY SSM param plus missing security-group path to the engine ALB4.0h68m2m3.5x

Aggregate Statistics

MetricValue
Total tasks31
Total human-equivalent hours637.0
Total Claude minutes2,492
Total supervisory minutes86
Total tokens14,990,716
Weighted average leverage factor15.3x
Weighted average supervisory leverage factor444.4x
Human-equivalent weeks15.9

Analysis

The highest factor of the day came in at 48.8x and the lowest at 3.5x, a spread of 13.8 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 100.0 of the 637.0 human-equivalent hours, or 16 percent of the day. No single task dominated the total, so the weighted average is representative.

Supervisory time was 86 minutes against 2,492 minutes of execution, a ratio of about 1 to 29. Supervisory leverage of 444.4x 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.