M5 Pro, macOS arm64
Tool: hyperfine 2.0.0, --warmup 3 --runs 10 --shell=none
Input: 2,182,862,808-byte temporary file made from a read-only 272,857,851-byte CCH Claude transcript repeated 8 times.
All comparisons read this same copy. Warmup runs primed the page cache. Output is suppressed.
Raw command output: docs/bench/m5-hyperfine.txt
Raw measurements: docs/bench/m5-hyperfine.json

Mean wall time (10 runs):
- jlogtail --lines 1000: 5.049 ms (σ 0.290 ms; min 4.573 ms; max 5.477 ms)
- tail -n 1000 | jq -c .: 66.516 ms (σ 3.233 ms; min 61.381 ms; max 71.081 ms)
- Python full read + splitlines + last 1000: 1395.601 ms (σ 188.841 ms; min 1259.268 ms; max 1877.208 ms)

The Rust reader seeks backward from EOF and reads the selected suffix. The two comparison commands scan the complete 2.18 GB file.
