PERFORMANCE
Benchmarks.
Every tebako release runs a benchmark suite on shared CI runners and publishes the full matrix, red cells included. The method note at the bottom of this page describes how the numbers are measured.
Results land on thebench-history branchof the tebako repository after each release; this page rebuilds nightly and on every benchmark update.
metanorma: tebako v1 vs v2
A real document-compilation workload: the metanorma CLI packaged the v1 way against the same CLI running on the v2 runtime and shim.
No results yet.
Results for this suite are published with each tebako release; the first run is still pending.
runtime overhead: on-system vs tebako
The same interpreter installed on the system against the tebako runtime pair — what mounting the environment image costs.
No results yet.
Results for this suite are published with each tebako release; the first run is still pending.
How these numbers are measured
Each figure is the median of repeated runs. The minimum is noted alongside as the figure comparable across runs, because shared CI runners inflate noise and never deflate it.
A cold run starts with the tebako store wiped, so it carries the one-time cost of downloading, verifying, installing, and first mounting the runtime. A warm run starts with the store present. On-system arms have no cold cell: package-manager install time is not tebako's comparison.
On-system and tebako arms run the same interpreter version — parity is asserted at measure time — and interleave A/B/A/B on the same runner. Ratios on this page are tebako median ÷ on-system median throughout: 1.00× is parity, above 1.00× is overhead, below 1.00× means tebako is faster.
Cells that were unavailable or failed are listed with their reasons, never hidden. The matrix is published complete.