docs: indicative Node vs Bun speed (packaging); trim benchmark cross-ref

Author: SutuSebastian

docs/benchmark.md

@@ -53,7 +53,7 @@ Each scenario runs both approaches back-to-back on the same machine, same data.
 This document measures **indexed SQL vs traditional glob/read** on an existing database — **not** process startup time or **Node vs Bun** as runtimes.
 
 - **Lazy CLI:** **`dist/index.mjs`** stays small; **`codemap --help`** / **`version`** load only bootstrap + the matching **dynamic import** chunk ([architecture.md § Layering](./architecture.md#layering)).
-- **Node vs Bun:** expect **lower latency** for short commands under **Bun** (typical in local checks); **full index** wall time varies by machine and cold cache. **`console.table`** output can differ slightly between runtimes for the same query rows. Same SQL semantics: [packaging.md § Node vs Bun](./packaging.md#node-vs-bun).
+- **Node vs Bun:** **`console.table`** output can differ slightly; SQL semantics match ([packaging.md § Node vs Bun](./packaging.md#node-vs-bun)). Indicative startup / small-rebuild speed ratios are documented there — this benchmark file does not measure Node vs Bun.
 
 **CI** runs **`node dist/index.mjs query "SELECT 1"`** after build to smoke-test the **Node + better-sqlite3** path ([`ci.yml`](../.github/workflows/ci.yml)).
 

docs/packaging.md

@@ -40,6 +40,8 @@ One schema and SQL surface; backend is chosen in **`src/sqlite-db.ts`**: **`bett
 
 **Checking both runtimes** after **`bun run build`**: run the same commands with **`node dist/index.mjs …`** and **`bun dist/index.mjs …`** (e.g. **`--version`**, **`query "SELECT 1"`**, **`CODEMAP_ROOT=fixtures/minimal`** **`--full`**). Startup and **`console.table`** formatting may differ; row counts and index stats should match.
 
+**Speed (indicative, same machine — not a spec):** Bun is often **~1.5×** faster than Node for **`--version`** and short **`query`** (process startup + loading **`dist/`**). A **`fixtures/minimal`** **`--full`** rebuild is often **~2×** faster on Bun by median wall time; first runs can be noisier. Large-app full rebuilds follow the same idea but scale with tree size and I/O.
+
 | Track        | Where                                                                                                  |
 | ------------ | ------------------------------------------------------------------------------------------------------ |
 | **CI**       | `node dist/index.mjs query "SELECT 1"` after build ([`ci.yml`](../.github/workflows/ci.yml)).          |