Planning eval · Codex · all 25 builds
Every Codex build.
gpt-6-astra built each of the five plans five times. Each card shows the score (screenshot checklist judged by Claude; unseen checks are excluded), build time and code size, with the app's main screen. Click a screenshot to see it full size.
All bank data in the screenshots is invented.
Goal only mean 90%, builds 83% to 96%

Build 1
96%
14 of 25 checks graded
34 min · 5,589 lines

Build 2
87%
19 of 25 checks graded
34 min · 3,194 lines

Build 3
94%
16 of 25 checks graded
29 min · 2,510 lines

Build 4
83%
12 of 25 checks graded
41 min · 5,537 lines

Build 5
90%
15 of 25 checks graded
35 min · 5,524 lines
Epics mean 92%, builds 88% to 100%

Build 1
90%
21 of 25 checks graded
35 min · 1,087 lines

Build 2
100%
9 of 25 checks graded
44 min · 1,332 lines

Build 3
90%
15 of 25 checks graded
35 min · 1,114 lines

Build 4
88%
17 of 25 checks graded
20 min · 5,746 lines

Build 5
92%
19 of 25 checks graded
39 min · 6,055 lines
Stories mean 95%, builds 86% to 100%

Build 1
94%
17 of 25 checks graded
46 min · 6,531 lines

Build 2
86%
14 of 25 checks graded
41 min · 5,456 lines

Build 3
97%
15 of 25 checks graded
39 min · 1,319 lines

Build 4
100%
12 of 25 checks graded
23 min · 1,055 lines

Build 5
97%
17 of 25 checks graded
21 min · 1,078 lines
Epics, features, stories mean 94%, builds 89% to 100%

Build 1
89%
14 of 25 checks graded
52 min · 1,523 lines

Build 2
100%
20 of 25 checks graded
41 min · 5,763 lines

Build 3
93%
14 of 25 checks graded
37 min · 6,104 lines

Build 4
93%
15 of 25 checks graded
22 min · 5,866 lines

Build 5
97%
16 of 25 checks graded
47 min · 2,329 lines
Sprint mean 96%, builds 94% to 100%

Build 1
94%
16 of 25 checks graded
51 min · 1,526 lines

Build 2
94%
17 of 25 checks graded
20 min · 1,053 lines

Build 3
100%
13 of 25 checks graded
27 min · 6,566 lines

Build 4
95%
19 of 25 checks graded
50 min · 7,364 lines

Build 5
97%
15 of 25 checks graded
46 min · 6,091 lines