Publish the resource benchmark harness

.github/workflows/benchmark.yml

@@ -0,0 +1,73 @@
+name: benchmark
+
+# CI guardrail for benchmark/ -- the one directory of runnable
+# source in this otherwise source-free, closed-source repository. Runs on
+# macOS because the harness is macOS-only by design: it depends on
+# lsappinfo, footprint, vm_stat, notifyutil, and pmset, all Darwin-only
+# tools.
+#
+# This workflow builds, lints, and tests the harness only. It deliberately
+# never runs `resource-benchmark run` (or `doctor` against real installed
+# subjects): a real sweep needs an exclusive, quiesced physical machine for
+# hours, and its numbers would be meaningless -- and misleading -- if
+# produced on a shared, noisy CI runner. Do not add that here.
+
+on:
+  push:
+    paths:
+      - 'benchmark/**'
+      - '.github/workflows/benchmark.yml'
+  pull_request:
+    paths:
+      - 'benchmark/**'
+      - '.github/workflows/benchmark.yml'
+
+concurrency:
+  group: benchmark-${{ github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+defaults:
+  run:
+    working-directory: benchmark
+
+jobs:
+  check:
+    runs-on: macos-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install Rust stable
+        uses: dtolnay/rust-toolchain@stable
+        with:
+          components: clippy
+
+      # benchmark/rustfmt.toml sets nightly-only options (e.g.
+      # unstable_features), so the format check needs a nightly rustfmt --
+      # stable `cargo fmt` reads a different subset of the config and
+      # produces different output.
+      # Installing nightly second makes it rustup's default, so every step
+      # below names its toolchain explicitly rather than relying on that
+      # ordering -- `cargo clippy` under nightly fails outright, since the
+      # nightly toolchain installed here carries rustfmt and nothing else.
+      - name: Install Rust nightly (rustfmt only)
+        uses: dtolnay/rust-toolchain@nightly
+        with:
+          components: rustfmt
+
+      - name: Rust cache
+        uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: benchmark -> target
+
+      - name: Format check (nightly rustfmt)
+        run: cargo +nightly fmt -- --check
+
+      - name: Clippy
+        run: cargo +stable clippy --all-targets -- -D warnings
+
+      - name: Test
+        run: cargo +stable test

.gitignore

@@ -1,2 +1,7 @@
 .DS_Store
 .dn/
+
+# benchmark/ is a standalone Rust crate; its own build output is scratch,
+# but benchmark/Cargo.lock is committed deliberately (see benchmark/README.md)
+# so a third party builds the exact dependency versions behind published numbers.
+benchmark/target/

LICENSE.md

@@ -1,7 +1,7 @@
 # Licence
 
-**TermTree is proprietary software. This repository contains documentation and release metadata
-only — it contains no TermTree source code.**
+**TermTree is proprietary software. This repository contains documentation, release metadata, and
+one standalone benchmark harness — it contains no TermTree source code.**
 
 ## This repository
 
@@ -10,9 +10,20 @@ accompanying images — are published by Document Node Pty Ltd so that people ca
 follow its releases, and report problems. They are made available for that purpose. All rights are
 reserved; no licence to the TermTree application is granted by anything in this repository.
 
-This repository is deliberately **not** published under an open source licence. TermTree's source
-code is not distributed, so there is nothing here for an open source licence to cover, and applying
-one would misrepresent what is on offer.
+Outside of the `benchmark/` directory described below, this repository is deliberately **not**
+published under an open source licence. TermTree's source code is not distributed here, so there
+is nothing else in this repository for an open source licence to cover, and applying one would
+misrepresent what is on offer.
+
+## The benchmark harness
+
+`benchmark/` is a deliberate, narrowly scoped exception. It is a standalone measurement tool, not
+part of the TermTree application, and TermTree's published memory, CPU, and cold-start figures
+depend on it — a benchmark nobody can run is an assertion, not evidence. For that reason `benchmark/`
+is published under the Apache License, Version 2.0, reproduced in
+[`benchmark/LICENSE`](benchmark/LICENSE). That licence covers the contents of the `benchmark/`
+directory only; it does not extend to any other part of this repository or to the TermTree
+application itself.
 
 ## The TermTree application
 

README.md

@@ -61,7 +61,7 @@ Full walkthrough: **[Getting Started](https://termtree.com/guide/getting-started
 
 ## Issues and support
 
-This repository is TermTree's public issue tracker. The application itself is closed source, so there is no code here — but bug reports, feature requests, and questions are all welcome and are read.
+This repository is TermTree's public issue tracker. The application itself is closed source — the only exception is the `benchmark/` harness described below — but bug reports, feature requests, and questions are all welcome and are read.
 
 - **Found a bug?** [Open a bug report](https://github.com/documentnode/termtree/issues/new?template=bug_report.yml). Include your OS, TermTree version, and steps to reproduce.
 - **Want a feature?** [Open a feature request](https://github.com/documentnode/termtree/issues/new?template=feature_request.yml).
@@ -72,6 +72,15 @@ You can also send feedback from inside the app using the feedback button in the
 
 Response times vary, but every issue gets triaged.
 
+## Resource benchmark
+
+TermTree's memory, CPU, and cold-start claims rest on a harness published in this repository at
+[`benchmark/`](benchmark/), the one deliberate exception to this repo's otherwise source-free,
+closed-source posture: numbers are only credible if strangers can measure them independently.
+It is released under the Apache License 2.0, scoped to that directory — see
+[`benchmark/LICENSE`](benchmark/LICENSE). For what it measures and how to run it, see
+[`benchmark/README.md`](benchmark/README.md).
+
 ## Links
 
 - [Website](https://termtree.com)

SECURITY.md

@@ -29,6 +29,11 @@ security fixes. Update from within the app, or download the current version at
 In scope: the TermTree desktop application, its update mechanism, its optional cloud sync, and the
 install script served at <https://termtree.com/install.sh>.
 
+Out of scope: the `benchmark/` harness published in this repository. It is a local developer tool
+you build and run yourself — it launches other applications on your own machine under your own
+account, and writes only to its own disposable scratch directory. Report bugs in it as an issue,
+not as a security report.
+
 Out of scope: the security of programs you choose to run inside TermTree's terminals. Those are
 ordinary terminal sessions — Claude Code, Codex, shells, servers, and other CLIs run with your own
 credentials under their own security models, exactly as they would in any other terminal emulator.

benchmark/.gitignore

@@ -0,0 +1,6 @@
+# Build output for this standalone crate.
+target/
+# Run results are attached to the published artifact deliberately, not
+# committed as run debris (design §4.1).
+results/*.json
+results/*.md