02 — Runs · v2 · jettygrowthteam · auto-evolved runbook (weakest-axis targeting)
claude-code
anthropic/claude-sonnet-5
claude-code driving anthropic/claude-sonnet-5 through OpenRouter, 5 hill-climb rounds from the v1 runbook.
Self-score vs judge, by axis
Best round v4, 0–10 per axis
Judge: The pelican reads well, with a white body, a long neck, an orange throat pouch and orange legs reaching toward the pedals, though the gray arm-like wing looks odd. The bicycle frame is structurally confused: the front wheel has no fork connecting it to the frame, the handlebar area is a tangle, and the body floats above an unclear seat. The clean gradient sky and speed lines are nice, but these joins in the frame and wing look glitchy.
Score by round
Out of 40. Each round starts from the previous round's best SVG. Dashed line: the 28/40 pass threshold.
5 drawings, oldest first.
Click a drawing to open it in the head-to-head viewer. The outlined card is the agent's best round. Verdicts are derived (code checks on the stored SVG and judge ≥ 28/40); see Runs.
Scores per round
| Round | Pelican | Bicycle | Composition | Polish | Self total | Judge total | Verdict | Judge spread | Cost | Minutes | Trajectory |
|---|---|---|---|---|---|---|---|---|---|---|---|
| v1 | 7self 7 | 6.2self 8 | 6.5self 7 | 6.8self 6 | 28 | 26.5 | fail | 1 | $0.80 | 5.7 | ea416761 |
| v2 | 7.2self 8 | 6self 8 | 7self 9 | 7.2self 9 | 34 | 27.3 | fail | 1 | $0.96 | 5.9 | b77ea720 |
| v3 | 7.3self 9 | 6self 8 | 6.8self 9 | 6.8self 8 | 34 | 27 | fail | 1.5 | $0.74 | 5.6 | 6e98127a |
| v4 | 7.5self 9 | 5self 9 | 6.5self 8 | 6.3self 10 | 36 | 25.3 | fail | 1 | $0.81 | 6.4 | 240351f2 |
| v5 | 7.3self 9 | 5.3self 9 | 6.8self 9 | 6.7self 9 | 36 | 26.2 | fail | 0.5 | $1.26 | 11.8 | 1925c665 |