[jnsdls]

Summary

The docs had fallen behind #39 (1.11.3), #40 (interning) and #43 (cached mapped ASTs). This brings them level and adds current benchmark ratios.

 README.md
+  "Queries are shared": interned queries, mapped AST and JSON reused; treat as immutable
-  dangling "wire and SQL tiers run against every adapter" line
+  **Adapters.** bullet in the compatibility list
+  build/map in the CPU case list
+  Py/TS ratio table at ec0768a, 1.11.3, Apple Silicon laptop
 docs/compat.md
+  known difference: logging (stdlib `logging`, not upstream's LogSink)
   re-wrap the interned-queries paragraph
 docs/plan.md
-  "small bun runner"            -> Node runner in harness/
-  "first pin is 1.10.0"         -> was 1.10.0; current pin in ZERO_VERSION
+  status: phases 1-7 done; phase 8 publishes on tag, npm watch not built
 CONTRIBUTING.md "Bumping Zero"
+  README version, docstring tag/commit, compat_scope line remap,
+  `python -m bench.capture`, record new differences
 bench/__init__.py
+  build and map groups

Evidence

Nothing on screen changes; docs only.

• Every API name the README names is exported from zero_server. The version is 1.11.3 everywhere it's pinned (ZERO_VERSION, both package.json files, the compose image, the docstrings).
• The ratio table comes from a full local uv run python -m bench (exit 0) at ec0768a. Postgres cases: Python within 10% of npm or faster. Query endpoint with JSON: 1.3–1.7x. build/map: 6–14x, but microseconds per call.
• ruff check and ruff format --check pass on bench/.

Merge Danger

Door: two-way

Blast Radius: docs

README is the PyPI long description, so the next release's page shows the new table. The ratios are from one laptop and will drift; the table names the commit and the machine.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[codspeed]

Merging this PR will not alter performance

✅ 46 untouched benchmarks

---

_{Comparing docs/refresh-1.11.3 (35851a3) with main (1a2fa10)}