Skip to content

A modern terminal with a built-in multiplexer: close it and your sessions keep running. Run shells and tons of AI agents across your machines, then pick them up from desktop, browser, TUI and more. Written in Rust with no UI framework: interface drawn straight on the GPU. No Electron, GPUI or webview.

docs: drop commit hashes from the benchmark reports

RoversX0d637bd1a265
diff --git a/docs/thinkterm/ascii-run-allocation-fix-results.json b/docs/thinkterm/ascii-run-allocation-fix-results.json
index 57a0aeb5..c461ae71 100644
--- a/docs/thinkterm/ascii-run-allocation-fix-results.json
+++ b/docs/thinkterm/ascii-run-allocation-fix-results.json
@@ -1,6 +1,4 @@
{
- "baseline_commit": "e791e31",
- "before_commit": "23b052a",
"production_patch_sha256": "d880e7ba41cc32c4ea8cc3c9e2dfd15972d4437d273a93cb67ddad3c7391e242",
"binary_sha256": {
"baseline": "654e2917ad8e23951c7bacd0b66ea0bd23f9f203d7f78393f2fdb465aff05ac2",
@@ -2068,6 +2066,5 @@
"server_exit": 0
}
}
- },
- "fixed_commit": "96dc67cf7f27cce34c6a0fe9c00316ff7b981355"
+ }
}
diff --git a/docs/thinkterm/ascii-run-allocation-fix.md b/docs/thinkterm/ascii-run-allocation-fix.md
index 91b25b44..b0b8e242 100644
--- a/docs/thinkterm/ascii-run-allocation-fix.md
+++ b/docs/thinkterm/ascii-run-allocation-fix.md
@@ -10,7 +10,7 @@ The existing scalar, owned-run, and first-action parser APIs keep their observab
## Comparison setup
-Three release binaries were tested: `e791e31` (before the three throughput optimizations), `23b052a` (owned runs, before this fix), and the borrowed-run fix (`96dc67c`; measured source matches its production patch). All runs used macOS arm64, a 120x32 PTY, 3,500 scrollback lines, private sockets, and a sandbox protecting existing user settings/sessions. No GUI client was attached. Builds and tests were finished before timing. Each suite used three rounds in the orders baseline/before/fixed, fixed/before/baseline, before/baseline/fixed.
+Three release binaries were tested: one before the three throughput optimizations, one with owned runs before this fix, and one with the borrowed-run fix (its measured source matches the production patch). All runs used macOS arm64, a 120x32 PTY, 3,500 scrollback lines, private sockets, and a sandbox protecting existing user settings/sessions. No GUI client was attached. Builds and tests were finished before timing. Each suite used three rounds in the orders baseline/before/fixed, fixed/before/baseline, before/baseline/fixed.
The short-run fixtures are deterministic: `unicode-N` repeats N ASCII x characters followed by 中; `sgr-N` alternates ANSI red and green text of N characters each. Each case sends approximately 32 MiB in approximately 64 KiB writes after a 256 KiB warm-up. A terminal status query waits for preceding output to be applied. Kitten 0.49.0 used `__benchmark__ --repetitions 100` for all six workloads; kitten randomizes its inputs.
diff --git a/docs/thinkterm/kitty-base64-results.json b/docs/thinkterm/kitty-base64-results.json
index 85694514..36d963ab 100644
--- a/docs/thinkterm/kitty-base64-results.json
+++ b/docs/thinkterm/kitty-base64-results.json
@@ -2,7 +2,6 @@
"date": "2026-09-23",
"platform": "macOS, Apple M1 Pro, 16 GiB RAM",
"kitten": "0.49.0",
- "source_base": "1f3b385",
"baseline_binary": "pre-fix parser-optimized binary recorded in parser-throughput-results.json",
"baseline_sha256": "98d5209cd4a41b4a1f9cf78f48d9b88592811f35c49ddaeaf0f0307e8f61d186",
"candidate_mux_sha256": "73a5230c893e32d1266a5fa6ddafb0c59440ae02c867656ec7edc81fbe5bede7",
diff --git a/docs/thinkterm/parser-throughput-followup-results.json b/docs/thinkterm/parser-throughput-followup-results.json
index 43e20443..0c82f85d 100644
--- a/docs/thinkterm/parser-throughput-followup-results.json
+++ b/docs/thinkterm/parser-throughput-followup-results.json
@@ -1,5 +1,4 @@
{
- "baseline_commit": "e791e310a72915ed7f9a03208437ec34166263cb",
"candidate_code_patch_sha256": "098e81d761c98d597d4813e10c4d9dd6ab3bb6fecf8c534124ee39a4313238a7",
"binary_sha256": {
"baseline": "654e2917ad8e23951c7bacd0b66ea0bd23f9f203d7f78393f2fdb465aff05ac2",
diff --git a/docs/thinkterm/parser-throughput-followup.md b/docs/thinkterm/parser-throughput-followup.md
index 6d438abe..b1f37392 100644
--- a/docs/thinkterm/parser-throughput-followup.md
+++ b/docs/thinkterm/parser-throughput-followup.md
@@ -4,7 +4,7 @@ A later [borrowed-run correction](ascii-run-allocation-fix.md) removes the
short ASCII/Unicode mixture allocation regression discovered after this series.
The measurements below describe the original owned-run implementation.
-This follow-up starts at `e791e31`, after the earlier screen-write batching and
+This follow-up starts after the earlier screen-write batching and
Kitty Base64 compatibility fixes. It evaluates the parser boundary, CSI row
updates, and repeated extended-attribute allocation separately from rendering.
diff --git a/docs/thinkterm/parser-throughput-results.json b/docs/thinkterm/parser-throughput-results.json
index 7a1fb2d1..6c04ae4e 100644
--- a/docs/thinkterm/parser-throughput-results.json
+++ b/docs/thinkterm/parser-throughput-results.json
@@ -1,5 +1,4 @@
{
- "baseline_commit": "06bdaa4",
"platform": "macOS, Apple M1 Pro, 16 GiB RAM",
"kitten": "0.49.0",
"size": [
diff --git a/docs/thinkterm/parser-throughput.md b/docs/thinkterm/parser-throughput.md
index 0fefe87e..533ef029 100644
--- a/docs/thinkterm/parser-throughput.md
+++ b/docs/thinkterm/parser-throughput.md
@@ -5,7 +5,7 @@ measures the next optimizations against a newer baseline. The results below
describe the earlier screen-write batching change.
Measured on 2026-09-23 with an Apple M1 Pro (16 GiB RAM), macOS, and kitten
-0.49.0. The baseline is commit `06bdaa4`, before the parser optimizations. Both
+0.49.0. The baseline is the build before the parser optimizations. Both
binaries were built locally with the same Cargo release profile. The measured
binary hashes and individual runs are in [the results JSON](parser-throughput-results.json).
diff --git a/docs/thinkterm/web-baseline.md b/docs/thinkterm/web-baseline.md
index a2fc0153..c19d2766 100644
--- a/docs/thinkterm/web-baseline.md
+++ b/docs/thinkterm/web-baseline.md
@@ -1,9 +1,8 @@
# Browser client: performance baseline
-Measured 2026-09-09 on `web-client` (one-pane numbers after 6473846, the
-four-pane run after the mirror landed), on an Apple-silicon Mac,
-headless Chrome (`--headless=new --enable-unsafe-webgpu`), release server,
-release bundle through `ci/build-web.sh` with `wasm-opt -Oz` (binaryen 132).
+Measured 2026-09-09 (the four-pane run after the mirror landed), on an
+Apple-silicon Mac, headless Chrome (`--headless=new --enable-unsafe-webgpu`),
+release server, release bundle through `ci/build-web.sh` with `wasm-opt -Oz` (binaryen 132).
Three runs of `thinkterm-web/smoke/bench.sh`; the spread between runs is
given where it matters. Nothing here is a target: it is the number to beat,
and the number a change must not make worse.