This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Two costs made the choice look hard: the slow-responding test craft at 22 m/s (73 against 45 crashes) and the wing at 16 m/s (40 against 26). Both came from layouts already looked at.
What was tested
Two fresh layouts per case, full extra room against the new default. The scoring was fixed and recorded before the results came in, and only the fresh layouts counted.
What was found then
Slow-responding test craft at 22 m/s: on the fresh pair the full room had 40 crashes against 53, the reverse of before. Wing at 16 m/s: 13 against 15, also reversed. Pooled over all layouts the full room is still slightly behind (113 against 98, and 53 against 41). The close-approach measurements meant to explain the difference were mixed both times. Neither cost repeated; the researcher's record calls both chance variation between layouts.
What changed
nothing
Corrections retained in the record
Earlier wording: the six fresh files' checksums are written into the registration before it is committed … registration and the read share the same second
Record correction: Reviewers noted that git order shows registration before the committed result, not before inspection; the records cannot rule out earlier inspection.
research/reviews/auditor-c179.md (F-FASTROW-TIMING)
Numbers and source references
- 73 vs 45; 40 vs 26
- earlier quoted costs (full bubble vs adopted crashes)research/data/oos/lagged_lateral_2x2.md; research/data/oos/wing_16_ext.md
- Two
- fresh seeds per caseresearch/data/oos/fastrow_lagged_lateral_22.md; research/data/oos/fastrow_wing_16.md
- 40 vs 53
- delayed-sideways 22 m/s crashes, fresh pair, full bubble vs adoptedresearch/data/oos/fastrow_lagged_lateral_22.md
- 13 vs 15
- wing 16 m/s crashes, fresh pairresearch/data/oos/fastrow_wing_16.md
- 113 vs 98; 53 vs 41
- all-seed pools (five / seven seeds)research/data/oos/fastrow_lagged_lateral_22.md; research/data/oos/fastrow_wing_16.md