[astral-automations-bot]

Fresh checkouts can change build-script input timestamps and invalidate otherwise reusable workspace artifacts. Enable the shared cache action's timestamp restoration for the Linux test job and the Linux x86_64 development-binary job.

Uses the timestamp restoration from astral-sh/github-actions#40.

Twenty paired exact warm-cache hits per row. Times are means in seconds, excluding queue and pre-step runner waiting. In the table, parentheses give the 90% confidence interval for mean time saved.

Job	Before (s)	After (s)	Speedup	Saved (s; 90% CI)
Linux tests	256.5	212.6	1.21×	43.9 (27.7 to 60.1)
Linux development binary	41.6	32.4	1.29×	9.2 (7.0 to 11.5)

Timing scope and runs

uv source is pinned to 0eac269. The baseline uses the merged cache action at 3c9fe5d; the candidate enables timestamp restoration at 374b3d5. Each pair uses the same runner class and workload.

Times span the first job step through post steps, including setup, cache restoration, and the development binary upload. Speedup is the ratio of mean times. The paired t intervals estimate uncertainty in the mean saving under these conditions. Values are rounded independently. Cache creation is outside these warm-hit timings.

The Linux breakdown below uses Cargo’s reported build time and nextest’s execution timer. Other job steps are outside this breakdown. Negative savings mean slower execution.

Job	Before (s)	After (s)	Speedup	Saved (s; 90% CI)
Linux test compilation	45.4	11.9	3.80×	33.4 (30.0 to 36.9)
Linux nextest execution	136.1	126.9	1.07×	9.2 (-3.3 to 21.6)

The nextest interval includes zero, so these runs do not establish a change in test execution time.

Linux pairs: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20.

Development binary pairs: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20.