Bidirectional type inference: one engine for rules, hover and inlay hints

.gitignore

@@ -186,3 +186,6 @@ book/dist/
 # without this the residue gets committed and the next run type-checks a file
 # the suite believes it created fresh.
 vscode-extension/test-fixtures/workspace/ofo_no_scan_target.py
+
+
+.deslop/

.vscode/settings.json

@@ -41,5 +41,6 @@
     "basilisk.enabled": true,
     "basilisk.uv.enabled": true,
     "deslop.topOffenders.splitByLanguage": true,
-    "deslop.topOffenders.groupBy": "file"
+    "deslop.topOffenders.groupBy": "type",
+    "deslop.topOffenders.sortBy": "path"
 }

Makefile

@@ -5,7 +5,7 @@
 # Exactly 7 standard targets: build, test, lint, fmt, clean, ci, setup
 # =============================================================================
 
-.PHONY: build test lint fmt clean ci setup book mutation-test conformance bench bench-basilisk reinstall-vsix reinstall-vsix-macos reinstall-vsix-prerelease
+.PHONY: build test lint fmt clean ci setup book mutation-test conformance bench bench-basilisk reinstall-vsix reinstall-vsix-macos reinstall-vsix-prerelease package-zed
 
 # ---------------------------------------------------------------------------
 # OS Detection
@@ -39,6 +39,12 @@ PKG                        ?= basilisk-checker
 # new tests just for mutation. Slow/E2E-ish binaries are deliberately omitted so
 # the per-mutant test run stays cheap.
 #
+# The one class of NEW binary that belongs here is a `#[mutation_safe]` suite
+# WIDENING the examined scope ([CHKARCH-TESTING-MUTATION-RATCHET]): those tests
+# assert real rule behaviour first and would earn their place with the ratchet
+# switched off — they are listed so the functions they newly bring in-scope are
+# actually exercised, not scored as missed.
+#
 # Order matters, but only a little. `cargo test` stops at the first failing
 # binary, so a mutant dies as soon as a binary that kills it runs;
 # `mutation_kill_tests` exists to kill these mutants, so it runs first. Measured
@@ -47,6 +53,7 @@ PKG                        ?= basilisk-checker
 # what actually did ([CHKARCH-TESTING-MUTATION-RATCHET]).
 _CHECKER_MUTATION_TESTS := \
 	--test mutation_kill_tests \
+	--test mutation_kill_constructors_tests \
 	--test coverage_boost_tests \
 	--test coverage_boost_32_tests \
 	--test coverage_boost_33_tests \
@@ -239,6 +246,9 @@ conformance:
 	@python3 conformance/run_conformance.py --bin target/debug/basilisk
 
 ## bench: Benchmark Basilisk vs pyright/mypy/ty/pyrefly/zuban on the fixture suite.
+## INDICATIVE ONLY — this runs on a developer workstation under whatever else it
+## is doing, so nothing passes or fails on the result. Compare tools within one
+## run; do not compare across machines or across time.
 ## Requires hyperfine; competitor tools are skipped if not installed.
 ## run.sh does the CLEAN release rebuild itself (fresh binary under test) before
 ## timing, so the guarantee holds even when run.sh is invoked directly — this
@@ -247,14 +257,22 @@ bench:
 	@bash benchmarks/run.sh
 
 ## bench-basilisk: Re-time ONLY basilisk (local iteration on a perf fix).
-## Same clean release rebuild, same stability policy, same zero-tolerance gate
-## against the committed baseline — it just skips the five competitors, which
-## add minutes per iteration and say nothing about a change to this tree. Their
-## CSV cells and versions carry forward verbatim and the header records that
-## they were not re-timed. Refused in CI, which always runs the full sweep.
+## Same clean release rebuild and same stability policy — it just skips the five
+## competitors, which add minutes per iteration and say nothing about a change to
+## this tree. Their CSV cells and versions carry forward verbatim and the header
+## records that they were not re-timed. Refused in CI, which runs the full sweep.
 bench-basilisk:
 	@BENCH_ONLY_BASILISK=1 bash benchmarks/run.sh
 
+## torture: Type-torture scoreboard — hard, spec-grounded typing problems
+## scored conformance-style (`# E` lines) against pyright/mypy/ty/pyrefly/zuban,
+## every tool in its out-of-the-box defaults, with hang detection as a
+## correctness axis. WRITE-ALWAYS to benchmarks/torture/status/torture.csv,
+## read-only regression gate against the committed baseline (exit 3).
+## Needs target/release/basilisk (or BASILISK_BIN); build it first.
+torture:
+	@python3 benchmarks/torture/run_torture.py
+
 ## smoke-micropython: Real-world smoke test for typeshed-path
 ## [STUBRES-CUSTOM-TYPESHED] — points the checker at a pinned, unmodified
 ## micropython-stdlib-stubs release and asserts MicroPython stdlib resolves
@@ -500,8 +518,16 @@ _test_nvim:
 _test_zed:
 	@bash scripts/test-zed.sh
 
-_package_zed:
-	@echo -e '\033[1m\033[0;36m▶ Building basilisk CLI for Zed\033[0m' && \
+## package-zed: Build the local Zed dev loop — compile the extension to WASM,
+## install the basilisk CLI, then print the `zed: install dev extension` steps.
+## Point the dev extension at the locally built binary with
+## `BASILISK_PATH=$$(which basilisk)` or `lsp.basilisk.binary.path`
+## ([ZED-DIST]); with neither, it downloads the release binary.
+package-zed:
+	@echo -e '\033[1m\033[0;36m▶ Building Zed extension (wasm32-wasip2)\033[0m' && \
+	rustup target add wasm32-wasip2 && \
+	cargo build --release --target wasm32-wasip2 --manifest-path $(_ZED_DIR)/Cargo.toml && \
+	echo -e '\033[1m\033[0;36m▶ Building basilisk CLI for Zed\033[0m' && \
 	cargo install --path crates/basilisk-cli --force && \
 	echo "$$(which basilisk) installed" && \
 	echo "" && \

README-pypi.md

@@ -29,8 +29,8 @@
 
 <p align="center">
   <a href="https://www.basilisk-python.dev/docs/conformance/"><strong><!--g:score-->100.0%<!--/g:score--> PEP conformance</strong></a> &mdash; <!--g:pass-->141<!--/g:pass--> of <!--g:total-->141<!--/g:total--> tests in the official
-  <a href="https://github.com/python/typing/tree/0dc9b5d23b368713af33ac25338eeb08b80f6360/conformance"><code>python/typing</code></a>
-  conformance suite (commit <code><!--g:short-->0dc9b5d<!--/g:short--></code>), scored on the wheel-installed CLI in its default config by the real upstream harness.
+  <a href="https://github.com/python/typing/tree/a4906624f170c169cf667f962080c56d5a5ba6ff/conformance"><code>python/typing</code></a>
+  conformance suite (commit <code><!--g:short-->a490662<!--/g:short--></code>), scored on the wheel-installed CLI in its default config by the real upstream harness.
   We target <code>python/typing@main</code> and ratchet the score up only.
 </p>
 
@@ -49,14 +49,14 @@ And it is the **fastest checker we&rsquo;ve measured** &mdash; median cold full-
 
 | Type checker | Median cold check |
 | --- | --- |
-| ⚡ **Basilisk** | **<!--g:benchBasilisk-->10<!--/g:benchBasilisk--> ms** |
+| ⚡ **Basilisk** | **<!--g:benchBasilisk-->12<!--/g:benchBasilisk--> ms** |
 | zuban | <!--g:benchZuban-->28<!--/g:benchZuban--> ms |
 | ty | <!--g:benchTy-->39<!--/g:benchTy--> ms |
-| Pyrefly | <!--g:benchPyrefly-->110<!--/g:benchPyrefly--> ms |
-| Pyright | <!--g:benchPyright-->563<!--/g:benchPyright--> ms |
-| mypy | <!--g:benchMypy-->583<!--/g:benchMypy--> ms |
+| Pyrefly | <!--g:benchPyrefly-->111<!--/g:benchPyrefly--> ms |
+| Pyright | <!--g:benchPyright-->582<!--/g:benchPyright--> ms |
+| mypy | <!--g:benchMypy-->605<!--/g:benchMypy--> ms |
 
-Median cold full-file check across <!--g:benchCount-->26<!--/g:benchCount--> single-construct typing-spec stress fixtures on an <!--g:benchMachine-->Apple M4 Max<!--/g:benchMachine--> &mdash; lower is better. Basilisk&rsquo;s warm re-check drops to ~<!--g:benchWarm-->4<!--/g:benchWarm--> ms. Every figure is produced by [`hyperfine`](https://github.com/sharkdp/hyperfine) and committed per machine, so nothing here is hand-typed. **Clone the repo, run `make bench` on your own hardware, and send us the CSV &mdash; independent audits are welcome.** [Full benchmarks &amp; methodology &rarr;](https://www.basilisk-python.dev/docs/benchmarks/)
+Median cold full-file check across <!--g:benchCount-->26<!--/g:benchCount--> single-construct typing-spec stress fixtures on an <!--g:benchMachine-->Apple M4 Max<!--/g:benchMachine--> &mdash; lower is better. Basilisk&rsquo;s warm re-check drops to ~<!--g:benchWarm-->5<!--/g:benchWarm--> ms. Every figure is produced by [`hyperfine`](https://github.com/sharkdp/hyperfine) and committed per machine, so nothing here is hand-typed. **Clone the repo, run `make bench` on your own hardware, and send us the CSV &mdash; independent audits are welcome.** [Full benchmarks &amp; methodology &rarr;](https://www.basilisk-python.dev/docs/benchmarks/)
 
 ## Everything in one extension
 

README.md

@@ -29,8 +29,8 @@
 
 <p align="center">
   <a href="https://www.basilisk-python.dev/docs/conformance/"><strong><!--g:score-->100.0%<!--/g:score--> PEP conformance</strong></a> &mdash; <!--g:pass-->141<!--/g:pass--> of <!--g:total-->141<!--/g:total--> tests in the official
-  <a href="https://github.com/python/typing/tree/0dc9b5d23b368713af33ac25338eeb08b80f6360/conformance"><code>python/typing</code></a>
-  conformance suite (commit <code><!--g:short-->0dc9b5d<!--/g:short--></code>), scored on the wheel-installed CLI in its default config by the real upstream harness.
+  <a href="https://github.com/python/typing/tree/a4906624f170c169cf667f962080c56d5a5ba6ff/conformance"><code>python/typing</code></a>
+  conformance suite (commit <code><!--g:short-->a490662<!--/g:short--></code>), scored on the wheel-installed CLI in its default config by the real upstream harness.
   We target <code>python/typing@main</code> and ratchet the score up only.
 </p>
 
@@ -49,14 +49,14 @@ And it is the **fastest checker we&rsquo;ve measured** &mdash; median cold full-
 
 | Type checker | Median cold check |
 | --- | --- |
-| ⚡ **Basilisk** | **<!--g:benchBasilisk-->10<!--/g:benchBasilisk--> ms** |
+| ⚡ **Basilisk** | **<!--g:benchBasilisk-->12<!--/g:benchBasilisk--> ms** |
 | zuban | <!--g:benchZuban-->28<!--/g:benchZuban--> ms |
 | ty | <!--g:benchTy-->39<!--/g:benchTy--> ms |
-| Pyrefly | <!--g:benchPyrefly-->110<!--/g:benchPyrefly--> ms |
-| Pyright | <!--g:benchPyright-->563<!--/g:benchPyright--> ms |
-| mypy | <!--g:benchMypy-->583<!--/g:benchMypy--> ms |
+| Pyrefly | <!--g:benchPyrefly-->111<!--/g:benchPyrefly--> ms |
+| Pyright | <!--g:benchPyright-->582<!--/g:benchPyright--> ms |
+| mypy | <!--g:benchMypy-->605<!--/g:benchMypy--> ms |
 
-Median cold full-file check across <!--g:benchCount-->26<!--/g:benchCount--> single-construct typing-spec stress fixtures on an <!--g:benchMachine-->Apple M4 Max<!--/g:benchMachine--> &mdash; lower is better. Basilisk&rsquo;s warm re-check drops to ~<!--g:benchWarm-->4<!--/g:benchWarm--> ms. Every figure is produced by [`hyperfine`](https://github.com/sharkdp/hyperfine) and committed per machine, so nothing here is hand-typed. **Clone the repo, run `make bench` on your own hardware, and send us the CSV &mdash; independent audits are welcome.** [Full benchmarks &amp; methodology &rarr;](https://www.basilisk-python.dev/docs/benchmarks/)
+Median cold full-file check across <!--g:benchCount-->26<!--/g:benchCount--> single-construct typing-spec stress fixtures on an <!--g:benchMachine-->Apple M4 Max<!--/g:benchMachine--> &mdash; lower is better. Basilisk&rsquo;s warm re-check drops to ~<!--g:benchWarm-->5<!--/g:benchWarm--> ms. Every figure is produced by [`hyperfine`](https://github.com/sharkdp/hyperfine) and committed per machine, so nothing here is hand-typed. **Clone the repo, run `make bench` on your own hardware, and send us the CSV &mdash; independent audits are welcome.** [Full benchmarks &amp; methodology &rarr;](https://www.basilisk-python.dev/docs/benchmarks/)
 
 ## Everything in one extension
 

README.zh.md

@@ -28,8 +28,8 @@
 
 <p align="center">
   <a href="https://www.basilisk-python.dev/zh/docs/conformance/"><strong>PEP 一致性 <!--g:score-->100.0%<!--/g:score--></strong></a> &mdash; 官方
-  <a href="https://github.com/python/typing/tree/0dc9b5d23b368713af33ac25338eeb08b80f6360/conformance"><code>python/typing</code></a>
-  一致性套件（提交 <code><!--g:short-->0dc9b5d<!--/g:short--></code>）<!--g:total-->141<!--/g:total--> 项测试中通过 <!--g:pass-->141<!--/g:pass--> 项，
+  <a href="https://github.com/python/typing/tree/a4906624f170c169cf667f962080c56d5a5ba6ff/conformance"><code>python/typing</code></a>
+  一致性套件（提交 <code><!--g:short-->a490662<!--/g:short--></code>）<!--g:total-->141<!--/g:total--> 项测试中通过 <!--g:pass-->141<!--/g:pass--> 项，
   由真实的上游评分器在默认配置下对 wheel 安装的 CLI 评出。
   我们以 <code>python/typing@main</code> 为目标，且分数只升不降。
 </p>
@@ -50,14 +50,14 @@ Basilisk 是**唯一**在官方
 
 | 类型检查器 | 冷检查中位数 |
 | --- | --- |
-| ⚡ **Basilisk** | **<!--g:benchBasilisk-->10<!--/g:benchBasilisk--> ms** |
+| ⚡ **Basilisk** | **<!--g:benchBasilisk-->12<!--/g:benchBasilisk--> ms** |
 | zuban | <!--g:benchZuban-->28<!--/g:benchZuban--> ms |
 | ty | <!--g:benchTy-->39<!--/g:benchTy--> ms |
-| Pyrefly | <!--g:benchPyrefly-->110<!--/g:benchPyrefly--> ms |
-| Pyright | <!--g:benchPyright-->563<!--/g:benchPyright--> ms |
-| mypy | <!--g:benchMypy-->583<!--/g:benchMypy--> ms |
+| Pyrefly | <!--g:benchPyrefly-->111<!--/g:benchPyrefly--> ms |
+| Pyright | <!--g:benchPyright-->582<!--/g:benchPyright--> ms |
+| mypy | <!--g:benchMypy-->605<!--/g:benchMypy--> ms |
 
-在 <!--g:benchMachine-->Apple M4 Max<!--/g:benchMachine--> 上对 <!--g:benchCount-->26<!--/g:benchCount--> 个单一构造的类型规范压力用例测得的整文件冷检查中位数 &mdash; 越低越好。Basilisk 的热重检查可降至约 <!--g:benchWarm-->4<!--/g:benchWarm--> ms。每个数字都由 [`hyperfine`](https://github.com/sharkdp/hyperfine) 产生并按机器提交，没有一个是手写的。**克隆仓库，在你自己的硬件上运行 `make bench`，并把 CSV 发给我们 &mdash; 欢迎独立复核。** [完整基准与方法论 &rarr;](https://www.basilisk-python.dev/zh/docs/benchmarks/)
+在 <!--g:benchMachine-->Apple M4 Max<!--/g:benchMachine--> 上对 <!--g:benchCount-->26<!--/g:benchCount--> 个单一构造的类型规范压力用例测得的整文件冷检查中位数 &mdash; 越低越好。Basilisk 的热重检查可降至约 <!--g:benchWarm-->5<!--/g:benchWarm--> ms。每个数字都由 [`hyperfine`](https://github.com/sharkdp/hyperfine) 产生并按机器提交，没有一个是手写的。**克隆仓库，在你自己的硬件上运行 `make bench`，并把 CSV 发给我们 &mdash; 欢迎独立复核。** [完整基准与方法论 &rarr;](https://www.basilisk-python.dev/zh/docs/benchmarks/)
 
 ## 一个扩展，覆盖全部
 

basilisk-zed/README.md

@@ -10,35 +10,49 @@ Basilisk is the only Python type checker scoring 100% on the [official `python/t
   <img src="https://raw.githubusercontent.com/Nimblesite/Basilisk/main/website/src/assets/images/zed-screenshot.png" alt="Basilisk in the Zed editor — Python type checking and diagnostics inline" width="900">
 </p>
 
+## Install
+
+Command palette (`Cmd+Shift+P` / `Ctrl+Shift+P`) → **zed: install dev extension** → select this directory (clone [`Nimblesite/basilisk-zed`](https://github.com/Nimblesite/basilisk-zed) first if you do not have the monorepo). Zed compiles the extension to WASM itself — you never pre-build or copy a `.wasm` file.
+
+**You do not install the Basilisk binary separately.** On first activation the extension downloads the matching binary for your platform from the [GitHub release](https://github.com/Nimblesite/Basilisk/releases), caches it inside Zed's extension directory, and reuses it until a newer release appears. Override it only for development or a system install, via `lsp.basilisk.binary.path` in `settings.json` or the `BASILISK_PATH` environment variable.
+
+> The extension is not yet listed in the [Zed extension registry](https://github.com/zed-industries/extensions); until that listing lands, the dev-extension flow above is the install path.
+
+Full instructions, settings, debugging, and the slash-command reference: [basilisk-python.dev/docs/install-zed](https://www.basilisk-python.dev/docs/install-zed/).
+
 ## Role in Basilisk
 
 This is the **Zed editor integration**. It is a native Zed extension compiled to WASM that connects the Basilisk language server to Zed, providing real-time diagnostics, hover, go-to-definition, code actions, and debugging via DAP.
 
 ## Key concepts
 
-- **WASM extension** — compiled as a `cdylib` crate targeting `wasm32-wasip1`, loaded natively by Zed.
+- **WASM extension** — compiled as a `cdylib` crate targeting `wasm32-wasip2`, loaded natively by Zed.
 - **`zed_extension_api`** — uses Zed's official extension API for language server lifecycle management.
 - **`basilisk-common`** — shares diagnostic codes and constants with the rest of the Basilisk workspace (also WASM-compatible).
-- **Tree-sitter grammars** — provides Python syntax highlighting via tree-sitter.
+- **Built-in Python, untouched** — binds to Zed's own Python language by name. The extension ships no `languages/` directory and no grammar, so Zed compiles nothing from source and your highlighting, brackets, indent rules, and runnables stay exactly as Zed ships them.
 - **DAP debugging** — supports the Debug Adapter Protocol for integrated Python debugging.
 
 ## Building
 
+From a monorepo checkout, build the extension and set up the local dev loop:
+
 ```sh
 make package-zed
 ```
 
+Standalone (this repository on its own), the build is exactly the one the release pipeline gates the publish on:
+
+```sh
+cargo build --release --target wasm32-wasip2
+```
+
 ## Dependencies
 
 | Crate | Purpose |
 |-------|---------|
 | `zed_extension_api` | Zed extension API |
 | `basilisk-common` | Shared constants and types |
 
-## Status
-
-Phase 2 — extension structure complete, connecting to the Basilisk LSP.
-
 ## License
 
 MIT.