22 tasks. July 16, 2026 closed at 17.5x weighted leverage across 251.0 human-equivalent hours in 859 minutes of wall-clock time. Supervisory leverage came in at 485.8x.
That is 6.3 weeks of human-equivalent throughput in 14.3 hours. The ceiling was 36.9x; the floor was 6.5x. 22 of the 22 entries came from a single project.
Task Log
| # | Task | Human Est. | Claude | Sup. | Factor |
|---|---|---|---|---|---|
| 1 | WP-1.1 course assembler v1: 10 modules + CLI + 99 tests, CHEM1B dry run verified (determinism, integrity, weights) | 24.0h | 39m | 1m | 36.9x |
| 2 | WP-0.2 hand-authored CHEM1B course spec (units 11-17, 69-leaf exact coverage, drift reconciliation, chem-nuc verification) | 3.0h | 5m | 1m | 36.0x |
| 3 | beta→candidate fleet migration: 69 specs+registry+staging coherently, ~407 reconciliation (dir-count myth), 2 production bugs found ([quality threshold] NULL, funnel-apply promotion integrity) | 16.0h | 31m | 1m | 31.0x |
| 4 | WP-0.6+0.7: checker blindspot fix (4-layout validation, 3 real defects surfaced), hardlink-safe npz migration script (guardrail-halted), atoms_v2 de-flag, images glob recursion, waiver lifecycle resolution | 14.0h | 31m | 2m | 27.1x |
| 5 | WP-0.5 engine simulation->lab rename + assign-simulation-to-goal path (false-premise correction, 3 latent catalog regressions caught, 29 tests) | 22.0h | 50m | 1m | 26.4x |
| 6 | [engine subsystem] WP-7.1+7.2: units-aware loaders, StudentProfile mid-course fields, graded-fidelity unit_placement bootstrap, seed-from-mastery hook (+104 tests) | 14.0h | 35m | 1m | 24.0x |
| 7 | Scenario B/C/D kit: [engine subsystem] fixes (lessons-root, exam-label null), goal-level mid-course fallback, 10 scenario profiles, readiness comparator validated on live sweep data (+80 tests) | 20.0h | 52m | 1m | 23.1x |
| 8 | WP-0.1 course-spec schema/parser/validator across 2 repos + WP-0.4 status investigation (refused stale reclassification via ADR archaeology) | 6.0h | 16m | 1m | 22.5x |
| 9 | Fleet omniscient sweep prep: 406 profiles, Temporal runner readiness + gotchas (flat-dir, --clouds AND), harvest script (found Temporal path drops verification rows), pilot config | 7.0h | 21m | 1m | 20.0x |
| 10 | Audit and review | 40.0h | 125m | 3m | 19.2x |
| 11 | WP-1.2+1.3: CHEM1B registered via real [engine subsystem] ingest/materialization + golden E2E harness (correctly blocked on corpus embedding gap; status-model finding) | 8.0h | 31m | 1m | 15.5x |
| 12 | Content production | 12.0h | 50m | 2m | 14.4x |
| 13 | WP-1.4: embeddings backfill (identity-gated e5 backfill of 40 nodes/40 pairs), corpus v41 + course package v2 via real ingest, golden E2E 8/8 PASS, amplify_pairs root cause | 12.0h | 52m | 1m | 13.8x |
| 14 | Engine fix batch: taxonomy topic-key alias (58 pkgs), /domains/load idempotency, exam passed:null default+provenance (+15 tests, fixture blast-radius audit, restart forensics) | 6.0h | 28m | 1m | 12.9x |
| 15 | Migration completion: principled gate fix (shared REGISTRY, ALWAYS_OPTIONAL), 74 domains to candidate (399 total), regen pilot (identity-gated, 54 stems/s, legacy-embedding non-reproducibility finding) | 10.0h | 48m | 2m | 12.5x |
| 16 | [engine subsystem]-service: fix stale MCQ mock fixtures dropped by new structural shape gate (root-caused to 58d2ce3, verified not a production regression) | 2.0h | 11m | 1m | 10.9x |
| 17 | WP-0.5b client lab/simulation split (web+electron+catalog, latent settings bug caught) + mathjs/jsxgraph dependency fix + rescue of 4 stranded lib commits | 6.0h | 35m | 1m | 10.3x |
| 18 | FOCPRO re-embed (broken node/pair embeddings) + 83 non-serving finding waivers + fixture soft-delete; waive-logic caught 5 held real gaps | 5.0h | 30m | 2m | 10.0x |
| 19 | WP-0.3 engine units-awareness: authored manifest units + fallback, unit filter fix, per-unit rollups, snapshot v10 (patch-verified handoff applied, 6933 tests green) | 7.0h | 42m | 1m | 10.0x |
| 20 | PMP 550 malformed-MCQ deterministic repair + structural MCQ shape generation gate (module+wiring+12 tests, committed 58d2ce3) | 7.0h | 45m | 3m | 9.3x |
| 21 | Fleet-engine swap + pilot omniscient sweep: 408 domains loaded (156s bulk reload, no wedge), pilot 3/3 at 100% (834/834), 2 engine bugs root-caused (exam-label null for 142 pkgs, taxonomy 500 for 58… | 4.0h | 27m | 1m | 8.9x |
| 22 | Atoms production-bug fix (d61123a) + generated/promoted 3 atoms domains (FOCE/Linux+/Network+); caught+fixed own finalizer premature-promotion race + coverage/criticals/dup-option fixes | 6.0h | 55m | 2m | 6.5x |
Aggregate Statistics
| Metric | Value |
|---|---|
| Total tasks | 22 |
| Total human-equivalent hours | 251.0 |
| Total Claude minutes | 859 |
| Total supervisory minutes | 31 |
| Total tokens | 9,987,649 |
| Weighted average leverage factor | 17.5x |
| Weighted average supervisory leverage factor | 485.8x |
| Human-equivalent weeks | 6.3 |
Analysis
The highest factor of the day came in at 36.9x and the lowest at 6.5x, a spread of 5.6 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 24.0 of the 251.0 human-equivalent hours, or 10 percent of the day. No single task dominated the total, so the weighted average is representative.
Supervisory time was 31 minutes against 859 minutes of execution, a ratio of about 1 to 28. Supervisory leverage of 485.8x 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.