🚀 Release 0.8.1

test:ci now runs c8 with --all --include "src/**/*.ts" --100, so the
build job fails when any runtime file under src/ is untested. --all is
what makes the gate non-vacuous: without it c8 counts only the files the
suite happened to load, and a new untested module stays invisible.

Add src/index.test.ts to load the public barrel, which was previously
never imported at runtime and so read as 0% under --all. matcher-shared.ts
is types-only (an empty runtime image) and carries a file-level c8 ignore
with the reason.

Why 100% and the rejected alternatives: development/ci.md § Coverage
threshold.

.gitea/workflows/ci.yml

@@ -77,6 +77,9 @@ jobs:
             - run: npm ci
             - run: npm run build
             - run: npm run check
+            # Fails the build below 100% coverage on `src/` (`c8 --all --100`);
+            # the same run produces the report published below. See
+            # development/ci.md § Coverage threshold.
             - run: npm run test:ci
             # Publish this tag's coverage to the self-hosted pages server,
             # served read-only at

.gitignore

@@ -10,5 +10,6 @@ coverage
 !.vscode/extensions.json
 !.vscode/settings.json
 !.vscode/tasks.json
+!.vscode/launch.json
 .idea
-.DS_Store
+.DS_Store

.oxlintrc.json

@@ -13,6 +13,8 @@
         "eslint/no-undefined": "off",
         "eslint/sort-keys": "off",
         "eslint/id-length": "off",
+        "eslint/capitalized-comments": "off",
+        "eslint/no-ternary": "off",
         "import/no-named-export": "off",
         "eslint/one-var": "off",
         "import/group-exports": "off",
@@ -20,7 +22,8 @@
         "eslint/sort-imports": "off",
         "import/consistent-type-specifier-style": "off",
         "unicorn/prefer-export-from": "off",
-        "typescript/method-signature-style": "off"
+        "typescript/method-signature-style": "off",
+        "typescript/promise-function-async": "off"
     },
     "options": { "typeAware": true },
     "env": { "builtin": true, "es2024": true, "node": true },
@@ -31,7 +34,9 @@
                 "no-unused-expressions": "off",
                 "no-empty-file": "off",
                 "import/no-nodejs-modules": "off",
-                "eslint/no-magic-numbers": "off"
+                "eslint/no-magic-numbers": "off",
+                "unicorn/no-null": "off",
+                "typescript/no-floating-promises": "off"
             }
         },
         {
@@ -39,6 +44,12 @@
             "rules": {
                 "import/no-nodejs-modules": "off"
             }
+        },
+        {
+            "files": ["src/util/__tests__/**"],
+            "rules": {
+                "import/no-nodejs-modules": "off"
+            }
         }
     ],
     "ignorePatterns": ["dist", "node_modules", "coverage"]

.vscode/extensions.json

@@ -1,5 +1,6 @@
 {
     "recommendations": [
+        "connor4312.nodejs-testing",
         "oxc.oxc-vscode",
         "streetsidesoftware.code-spell-checker",
         "typescriptteam.native-preview",

.vscode/launch.json

@@ -0,0 +1,15 @@
+{
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "type": "node",
+            "request": "launch",
+            "name": "Debug current test file",
+            "runtimeExecutable": "node",
+            "runtimeArgs": ["--test", "--strip-types"],
+            "args": ["${file}"],
+            "cwd": "${workspaceFolder}",
+            "console": "integratedTerminal"
+        }
+    ]
+}

.vscode/settings.json

@@ -1,4 +1,14 @@
 {
+    "nodejs-testing.extensions": [
+        {
+            "extensions": ["mjs", "cjs", "js"],
+            "parameters": []
+        },
+        {
+            "extensions": ["ts"],
+            "parameters": ["--strip-types"]
+        }
+    ],
     "[typescript]": {
         "editor.defaultFormatter": "oxc.oxc-vscode",
         "editor.formatOnSave": true

AGENTS.md

@@ -9,9 +9,10 @@ first-action facts. Do not restate evolving prose here — it will drift.
 
 - Project: F#-style pattern matching for TypeScript/ESM. Node `>=26` (pinned via `.node-version`), ESM-only (no CommonJS shim).
 - **While iterating:** `npm run test` (`check:tsc` + the unit suite) for fast feedback on the files you changed.
-- **Optional code intelligence:** this repo installs `@spences10/pi-lsp` (pinned in `.pi/settings.json`) as a project-local pi extension. It talks to the repo's own TypeScript 7 via `tsc --lsp --stdio` and exposes **read-only** tools — `lsp_hover`, `lsp_definition`, `lsp_references`, `lsp_find_symbol`, `lsp_document_symbols`, `lsp_diagnostics(_many)`. Prefer `lsp_references` over `grep -w` for widely-colliding identifiers (`matches`, `type`, …); use `lsp_hover` to read inferred types on generic-heavy code. It has no rename / code-action / apply-edit surface — the write side is pi's `edit` tool + `check:tsc`. Treat empty LSP output as _inconclusive_, not success: **`npm run test` / `npm run verify` remain the sole authoritative gate** (see the next bullet). The server keeps running across that gate with a ~5 min idle timeout and registers no file watchers, so if you change `tsconfig.json` / `package.json` mid-session its diagnostics can be stale — when LSP output disagrees with `check:tsc`, trust `check:tsc` and restart pi (or wait out the idle timeout) before concluding the LSP is wrong.
+- **Optional code intelligence:** this repo installs `@spences10/pi-lsp` (pinned in `.pi/settings.json`) as a project-local pi extension. It talks to the repo's own TypeScript 7 via `tsc --lsp --stdio` and exposes **read-only** tools — `lsp_hover`, `lsp_definition`, `lsp_references`, `lsp_find_symbol`, `lsp_document_symbols`, `lsp_diagnostics(_many)`. **When you are looking for a symbol, reach for the LSP before `rg`/`grep`** — `lsp_references` / `lsp_find_symbol` / `lsp_definition` / `lsp_document_symbols` are semantic and cross-file, so they see shadowing, imports and overloads that a text search cannot; use `lsp_hover` to read inferred types on generic-heavy code. Use `rg` for what the LSP cannot see — doc prose, string literals, config, task lists, file discovery — and reconcile the two sets before editing (symbols from the LSP, strings and prose from `rg`). It has no rename / code-action / apply-edit surface — the write side is pi's `edit` tool + `check:tsc`. Treat empty LSP output as _inconclusive_, not success: **`npm run test` / `npm run verify` remain the sole authoritative gate** (see the next bullet). The server keeps running across that gate with a ~5 min idle timeout and registers no file watchers, so if you change `tsconfig.json` / `package.json` mid-session its diagnostics can be stale — when LSP output disagrees with `check:tsc`, trust `check:tsc` and restart pi (or wait out the idle timeout) before concluding the LSP is wrong.
 - **Definition of done — run this before you call the work finished:** `npm run verify`. If all green, commit. If red, look at the output, fix the root cause, and re-run.
 - **On commit:** write a good message (see [CONTRIBUTING.md § Commit messages](./CONTRIBUTING.md#commit-messages)). Lefthook's pre-commit hook already runs the fast, offline, staged-file checks — don't run them by hand. If the hook fails on style, `npm run fix`, restage, recommit.
+- **Document decisions where the next maintainer will look:** rationale, rejected alternatives and known issues go in `development/<category>.md` (see [development/README.md](./development/README.md)); the actionable rule stays in [CONTRIBUTING.md](./CONTRIBUTING.md) and links to it. Write each fact once — never copy the rule into `development/` or the reason into `CONTRIBUTING.md` — and change both in the same commit when a rule changes.
 - **`npm run maintain` is NOT part of the feature loop.** Its scans are advisory, never a gate; run them only on an explicit maintenance / update-deps branch.
 
     ```sh
@@ -25,9 +26,10 @@ Don't silence the type system to force a green run. As an agent these are forbid
 
 - `// @ts-nocheck`, `// @ts-ignore`, `// @ts-expect-error`
 - `// oxlint-disable` / `// oxlint-disable-next-line`
+- editing `.oxlintrc.json` to silence a finding (e.g. turning `typescript/no-floating-promises` off)
 - `as` casts used to push an expression through (type-aware oxlint already flags unsafe assertions)
 
-Fix the root cause with the type system instead — narrowing, generics, `satisfies`, conditional / mapped types, utility types (`NonNullable`, `Exclude`, …). TypeScript can express it; that's the intended tool. The `oxlint-disable`-location rule in [CONTRIBUTING.md § Rules the tools don't enforce](./CONTRIBUTING.md#rules-the-tools-dont-enforce) is a **human** last-resort convention (so a reviewer can spot a deliberate suppression) — it is not permission for you to add one. If the types genuinely cannot express something, stop and surface the conflict (commit message / handover) rather than suppress it.
+Fix the root cause with the type system instead — narrowing, generics, `satisfies`, conditional / mapped types, utility types (`NonNullable`, `Exclude`, …). TypeScript can express it; that's the intended tool. Suppressions — a source `oxlint-disable` **or** a `.oxlintrc.json` entry — are a **human** last resort, not a tool for you. If the types genuinely cannot express something, stop and surface the conflict (commit message / handover) rather than suppress it.
 
 The same applies to the checks themselves: **never `git commit --no-verify`** (or otherwise skip a pre-commit / pre-push hook). The checks are fast and offline, so a redundant run is fine — bypassing a hook to get green is the identical anti-pattern. If a commit already skipped a hook, redo it through one: `git reset --soft HEAD~1 && git commit -C <skipped-sha>`.
 
@@ -41,7 +43,9 @@ Never start a long-lived / blocking process such as `npm run watch`. It runs unt
 
 ### Working on tasks
 
-- **Task with subtasks** (a task that has indented children): create the branch with `npm run create:branch -- <prefix>/<desc>`, inferring the prefix from the task content (`feature/…` / `fix/…` / `chore/…`) — do not hand-write `git switch -c`, the command enforces the clean-tree / current-`main` / green-baseline precondition. Work on each subtask with commits, then present a concise handover for the user to review. Use this fixed shape:
+Every task — with or without subtasks — goes through the complete branching model:
+
+- Create the branch with `npm run create:branch -- <prefix>/<desc>`, inferring the prefix from the task content (`feature/…` / `fix/…` / `chore/…`) — do not hand-write `git switch -c`, the command enforces the clean-tree / current-`main` / green-baseline precondition. Work with commits (each subtask gets one or more), then present a concise handover for the user to review. Use this fixed shape:
 
     ```md
     ## Handover — <branch>
@@ -53,9 +57,7 @@ Never start a long-lived / blocking process such as `npm run watch`. It runs unt
 
     Once the user has no further objections, merge back: `npm run create:finish` (on the branch — it merges `--no-ff`, runs `npm run verify`, and deletes the branch). The branching model is documented in [CONTRIBUTING.md § Branching model](./CONTRIBUTING.md#branching-model).
 
-- **Leaf task** (no indented children): implement on the current branch and commit.
+Follow [CONTRIBUTING.md § Testing discipline (type-driven)](./CONTRIBUTING.md#testing-discipline-type-driven) throughout.
-
-In both cases, follow [CONTRIBUTING.md § Testing discipline (type-driven)](./CONTRIBUTING.md#testing-discipline-type-driven). Each subtask gets one or more commits.
 
 ## Read these
 
@@ -64,5 +66,6 @@ In both cases, follow [CONTRIBUTING.md § Testing discipline (type-driven)](./CO
 - [CONTRIBUTING.md § Script prefix convention](./CONTRIBUTING.md#script-prefix-convention) — adding an `npm run` script? reuse an existing prefix or it doesn't belong.
 - [CONTRIBUTING.md § Commit messages](./CONTRIBUTING.md#commit-messages) — gitmoji + imperative + 50/72.
 - [CONTRIBUTING.md § Feedback tiers](./CONTRIBUTING.md#feedback-tiers) — what runs when and at what cost (`watch` / pre-commit / pre-push / `check` / `verify` / `fix` / `maintain` / CI).
-- [README.md § Tooling decisions](./README.md#tooling-decisions) — the rationale behind each tool choice; read before changing tooling.
+- [development/](./development/README.md) — the decisions, rejected alternatives and known issues behind the rules; the “why” that CONTRIBUTING.md links to. Read the relevant file before changing an area.
+- [development/tooling.md](./development/tooling.md) — the rationale behind each tool choice; read before changing tooling.
 - [package.json `#scripts`](./package.json) — the source of truth for every command (the `LEFTHOOK_FILES` convention scopes them to staged files vs. the whole project).

CHANGELOG.md

@@ -7,6 +7,75 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.8.1] - 2026-09-23
+
+- gate CI at 100% coverage: `test:ci` runs c8 with `--all --100` over `src/`,
+  and a test loads the `index.ts` barrel so it is measured
+- pin the duplicate-tag union collapse (members sharing a tag dispatch through
+  one handler) with tests
+
+## [0.8.0] - 2026-09-23
+
+- reject a universe that mixes a literal with a broad type (e.g.
+  `"a" | \`x-${number}\``), which the finite-literal gate let through
+- narrow the tagged-union matcher's handler parameters for a union-valued or
+  optional discriminant, and its fallback to the unhandled tags, instead of
+  passing `never`
+
+## [0.7.1] - 2026-09-23
+
+- upgrade dependencies
+
+## [0.7.0] - 2026-09-23
+
+- reject broad universes (`string`, `number`, template literals) and
+  value/stringification collisions (`true | "true"`, `1 | "1"`) at the factory
+- type handler parameters by the matched member, so a
+  standalone `"true"` universe is typed `"true"` rather than `true`
+
+## [0.6.0] - 2026-09-22
+
+- add `getTaggedUnionMatcher` / `getTaggedUnionMatcherW` for discriminated unions
+- rename the primitive matcher to primitive-union: `getMatcher` /
+  `getMatcherW` → `getPrimitiveUnionMatcher` / `getPrimitiveUnionMatcherW`, and
+  `src/primitive.ts` / `src/primitive.test.ts` → `src/primitive-union.*`
+
+## [0.5.0] - 2026-09-21
+
+- reject a fallback when the handler map already covers the universe
+- allow boolean, null and undefined in the matcher universe
+
+## [0.4.0] - 2026-09-20
+
+- move the `_` fallback out of handlers
+
+## [0.3.0] - 2026-09-18
+
+- condensed primitive matcher factories down to 2 from formerly 4
+- house shared test helpers under `src/util/__tests__/`
+
+## [0.2.0] - 2026-09-16
+
+- allow ternaries and lowercase comments in oxlint
+- switch `src/primitive.ts` prose from block comments to line comments
+- remove the template's `match`/`P` example modules and their documentation, and point `src/index.ts` at the primitive matchers
+
+## [0.1.8] - 2026-09-16
+
+- upgrade dependencies
+- ignore `@types/node` in `maintain:outdated` (misleading `latest` dist-tag)
+- require extremely concise prose in `development/`
+
+## [0.1.7] - 2026-09-16
+
+- require a short summary under `[Unreleased]` in the changelog before a branch is finished
+- let `create:branch` start from a `main` that is ahead of its upstream, so a finished merge no longer blocks the next branch until it is pushed
+
+## [0.1.6] - 2026-09-15
+
+- restructure the documentation: README.md for users, CONTRIBUTING.md for contributors, and development/ for the decisions, rejected alternatives and known issues
+- document the decisions and known issues for CI, tooling, testing, publishing and the workflow
+
 ## [0.1.5] - 2026-09-15
 
 - improve CI configuration
@@ -32,7 +101,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - basic setup
 
-[Unreleased]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.1.5...main
+[Unreleased]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.8.1...main
+[0.8.1]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.8.0...0.8.1
+[0.8.0]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.7.1...0.8.0
+[0.7.1]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.7.0...0.7.1
+[0.7.0]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.6.0...0.7.0
+[0.6.0]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.5.0...0.6.0
+[0.5.0]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.4.0...0.5.0
+[0.4.0]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.3.0...0.4.0
+[0.3.0]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.2.0...0.3.0
+[0.2.0]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.1.8...0.2.0
+[0.1.8]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.1.7...0.1.8
+[0.1.7]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.1.6...0.1.7
+[0.1.6]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.1.5...0.1.6
 [0.1.5]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.1.4...0.1.5
 [0.1.4]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.1.3...0.1.4
 [0.1.3]: https://gitea.e1nsnull.de/tmu/tiny-pattern-ts/compare/0.1.2...0.1.3

CONTRIBUTING.md

@@ -1,49 +1,36 @@
 # Contributing
 
-This document is for maintainers and contributors working on the project itself. End-user documentation is in [README.md](./README.md). The machine entry point for AI coding agents is [AGENTS.md](./AGENTS.md); keep this file as the prose home for the rules below so agents and humans don't diverge.
+This document is for maintainers and contributors working on the project
+itself. End-user documentation is in [README.md](./README.md). The reasons
+behind the rules here — the decisions, rejected alternatives, and known issues —
+live in [development/](./development/README.md). The machine entry point for AI
+coding agents is [AGENTS.md](./AGENTS.md); keep this file as the prose home for
+the rules so agents and humans don't diverge.
 
-## Rules the tools don't enforce
+## Setup
 
-CI and review will bounce these even though `npm run check` and the linters don't catch them. They're the high-frequency things a contributor (or an agent) reaches for by default:
+1. Clone the repository.
+2. Install Node.js >= 26 — see [.node-version](./.node-version); the exact pinned
+   version is what CI and the runner image use.
+3. `npm ci`.
+4. `npm run setup` — the one-time clone configuration (currently registers the
+   commit-message template).
 
-- **Source imports use `.ts` extensions, never `.js`.** `node --strip-types` only resolves the `.ts` form at test time; "Pre-fixing" an import to `.js` breaks the inner loop. (rationale: README § Tooling decisions)
+## Development commands
-- **A new `npm run` script must reuse an existing prefix** (`create:` / `check:` / `fix:` / `test:` / `watch:` / `maintain:` / `publish:` / `setup:`). If none fits, that's a signal the script doesn't belong in the pipeline — not a reason to invent a new prefix. If it genuinely does belong, add the prefix to both lists here in the same commit; an undocumented prefix becomes invisible and quietly accrues members. (see [Script prefix convention](#script-prefix-convention))
-- **`oxlint-disable` directives live in source, not `.oxlintrc.json`.** The trade-off must sit next to the code it silences. This is a _human_ last-resort convention; agents must not add these — see [AGENTS.md § Never do](./AGENTS.md#never-do). (rationale: README § Tooling decisions)
-- **Don't put slow / network / whole-project scans in `check` or pre-commit.** Advisory scans are not correctness gates; they belong under `maintain:`. (see [Feedback tiers](#feedback-tiers) and [Script prefix convention](#script-prefix-convention))
-- **New work starts with `npm run create:branch`, never a hand-written `git switch -c`/`git checkout -b`.** The command carries the branch precondition; branching around it skips the clean-tree, current-`main` and green-baseline checks, and the skip is invisible until a failure can no longer be attributed. (see [Branching model](#branching-model))
-- **Work is merged back with `npm run create:finish`, never a hand-written `git merge`.** The command carries the merge-side preconditions (clean tree, current `main`, a `feature/`/`fix/`/`chore/` branch) and runs `npm run verify` after the merge, so a merge cannot land unverified. (see [Branching model](#branching-model))
-- **There is no local `npm run publish`, and `publish:publint` / `publish:attw` don't go in `check`.** (see [Publishing workflow](#publishing-workflow))
 
-## Editor configuration
+- **Build:** `npm run build`
-
+- **Test:** `npm run test`, `npm run test:ci`
-`.editorconfig` is for editor compatibility, not a gate — it is a sane fallback for the files oxfmt does not format (shell scripts, dotfiles, `LICENSE`, the commit-message template, and git's `COMMIT_EDITMSG` buffer). Where both apply, `.oxfmtrc.json` is authoritative: oxfmt is the formatter, and the overlapping `.editorconfig` keys only keep non-oxfmt editors close to the formatted result.
+- **Watch:** `npm run watch` - re-runs tests on file save, humans only
-
+- **Checks:** `npm run check`, `npm run fix`
-## Commit messages
+- **Verify:** `npm run verify` — the definition of done
-
+- **Maintenance:** `npm run maintain` — advisory only
-Gitmoji subject, imperative mood, 50/72 wrapping. The template is `commit-message-template`; run `npm run setup:git-commit-message` once after cloning to register it as git's `commit.template` (or `npm run setup` to run every one-time clone step).
+- **Individual fixes:** `npm run fix:oxfmt`, `npm run fix:oxlint`
-
-Examples from history: `:sparkles: Add watch tier with watch:test child`, `:recycle: Move type-aware config to .oxlintrc.json; use source-level disable directives`, `:memo: Restore unique maintainer content as CONTRIBUTING.md`. The body explains _what and why_, not _how_; link issues with `Resolves #...`.
-
-## Script prefix convention
-
-Script names in `package.json` use a prefix that signals _when_ the script is intended to run. A `<prefix>:<name>` script is implicitly aggregated by a `<prefix>` script (if one exists) and run by the corresponding lefthook hook or CI step. Picking the right prefix documents the script's intended lifecycle:
-
-- `create:*` — front doors of the repo's own workflow; these mutate git state rather than the source. `create:branch` opens a unit of work (asserts a clean tree, a current `main` and a green baseline before it branches), `create:finish` closes the branch half (merges the current unit of work into `main` and verifies the result), `create:release` closes the release half (maintainer-only). No bare `create` aggregator on purpose — see `publish:*` for the precedent.
-- `check:*` — read-only verification; never modifies files. Aggregated by `npm run check`.
-- `fix:*` — mutating counterpart of a `check:*` script. Aggregated by `npm run fix`; the diff is the review surface.
-- `test:*` — test scripts. `test` is the canonical entry point (`check:tsc` + unit tests); `test:unit` skips the typecheck for fast local iteration; `test:ci` adds c8 coverage.
-- `watch:*` — long-running watchers for the manual inner dev loop. Aggregated by `watch`; currently a single child (`watch:test`), and a future `watch:oxlint` / `watch:tsc` would run concurrently under that umbrella.
-- `maintain:*` — advisory repo-maintenance scans: read-only, but whole-project and/or network-bound, so never a correctness gate. Aggregated by `npm run maintain`.
-- `publish:*` — validates the _publishable artifact_ (e.g. `dist/`) rather than the source, so it needs a fresh build. (see [Rules the tools don't enforce](#rules-the-tools-dont-enforce) and [Publishing workflow](#publishing-workflow))
-- `setup:*` — one-time configuration of a fresh clone; mutates the local environment (git config, editor settings) rather than the repo source, so it is never part of a hook or CI step. Aggregated by `npm run setup` (the umbrella), run once after cloning.
-
-A new script should pick the prefix that matches its lifecycle, not invent a new one. If no existing prefix fits, that's a signal the script doesn't belong in the standard pipeline. When it genuinely does belong, a new prefix is allowed — but it enters both lists in this file in the same commit as its first member, otherwise the rule "reuse an existing prefix" silently develops an exception (the old `use:` prefix was exactly that; it is now retired into `setup:`).
-
-Separately, some top-level scripts are **bare** (no prefix): the entry points that either run a single tool (`build`, `clean`) or aggregate a `prefix:*` family (`check`, `fix`, `test`, `watch`, `maintain`, `setup`), plus one convenience that composes across tiers: `verify` — composing `check` + `test:unit` into one whole-project correctness gate (it deliberately uses `test:unit` rather than `test` because `check` already runs `check:tsc`, so the type checker runs exactly once). Bare commands are how you invoke a tier; the `prefix:*` scripts are what those tiers are made of.
 
 ## Feedback tiers
 
-The tools are organized into a feedback ladder. Each tier catches different things at different costs; the rule of thumb is "earlier tiers fire more often, faster tiers catch less, slower tiers are more thorough":
+The tools are organized into a feedback ladder. Each tier catches different
+things at different costs; the rule of thumb is "earlier tiers fire more often,
+faster tiers catch less, slower tiers are more thorough":
 
 | Tier                             | When                    | What it runs                                                                                                                                            | Time  |
 | -------------------------------- | ----------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
@@ -54,66 +41,203 @@ The tools are organized into a feedback ladder. Each tier catches different thin
 | `npm run verify`                 | manual                  | Definition of done: `npm run check` + unit tests, one shot                                                                                              | ~6s   |
 | `npm run fix`                    | manual                  | Auto-resolve fixable issues (lint, format)                                                                                                              | ~3s   |
 | `npm run maintain`               | manual / CI (advisory)  | `maintain:knip` + `maintain:outdated` (whole-project + network scans)                                                                                   | ~10s  |
-| CI build (auto)                  | on push to `main` / tag | `build` job (build + correctness + packaging) — see [.gitea/workflows/ci.yml](./.gitea/workflows/ci.yml)                                                | ~30s+ |
+| CI build (auto)                  | on push to `main` / tag | `build` job (build + correctness + coverage + packaging) — see [.gitea/workflows/ci.yml](./.gitea/workflows/ci.yml)                                     | ~30s+ |
 | CI maintain (auto, non-blocking) | on push to `main`       | `npm run maintain` — reports, never fails the build                                                                                                     | ~10s  |
 | CI publish (auto)                | on tag                  | packaging checks + `publish:publint` / `publish:attw`, then the Gitea release page and `npm publish` (skipped, and the job failed, without `NPM_TOKEN`) | ~15s  |
 
-### Why these splits?
+Before pushing, run `npm run verify` — the one-shot correctness gate. Run
-
+`npm run maintain` only on a maintenance / update-deps branch. Why the splits
-- **`watch:*` is a manual tier, not a hook.** The developer starts it on demand (it has to be killed with Ctrl-C) and it runs in a dedicated terminal pane. It sits as the earliest tier in the feedback ladder, catching failures the moment a file is saved — before staging, before commit.
+are where they are: [development/workflow.md § Feedback tiers](./development/workflow.md#feedback-tiers).
-- **`check:tsc`, `check:oxlint`, `check:oxfmt`, `check:cspell`** are in pre-commit because they are fast (~0.2–0.5s each), fully offline, and naturally scope to staged files via the `LEFTHOOK_FILES` env var convention. They give instant feedback on what you typed.
-- **`test` (and the `tsc` it includes) is in pre-push** because it runs the whole test suite across the whole project. The pre-commit `LEFTHOOK_FILES` convention doesn't apply to the test runner, so pre-commit isn't the right home. Pre-push runs after all commits are made but before the push leaves the machine, catching regressions that span multiple commits.
-
-### Before pushing
-
-Run `npm run verify` — the one-shot correctness gate in the table above. Run `npm run maintain` only on a maintenance / update-deps branch.
 
 ## Testing discipline (type-driven)
 
-For this library the types _are_ the feature — narrowing, `exhaustive()` returns, the `Matcher<T>` contract — so a runtime-only test loop would verify the wrong thing. New behavior follows **type-driven development** (in Edwin Brady's sense): _treat the type as the plan for a program, and use the compiler and type checker as your assistant, guiding you to a complete program that satisfies the type_ ([idris-lang.org](https://www.idris-lang.org/)). Here that plan is the `expectTypeOf` assertion, written first. The loop is **type → red → green → refactor**:
+For this library the types _are_ the feature, so development is **type-driven**:
+the compile-time expectation is written before the runtime assertion, and both
+before the implementation. The loop is **type → red → green → refactor**:
 
-1. **Type** — write the compile-time expectation first (`expectTypeOf(...).toEqualTypeOf<…>()`) and let `npm run check:tsc` fail on the _type_. The type error is the spec you want to hit before the runtime logic exists.
+1. **Type** — write the compile-time expectation first
-2. **Red** — add the matching runtime assertion (`assert.*`) so `npm run test:unit` now fails on behavior.
+   (`expectTypeOf(...).toEqualTypeOf<…>()`) and let `npm run check:tsc` fail on
-3. **Green** — implement in `src/*.ts` until both the type check and the test pass.
+   the _type_. The type error is the spec you want to hit before the runtime
-4. **Refactor** — with the type system and the tests as the safety net, then `npm run verify` as the definition-of-done gate.
+   logic exists.
+2. **Red** — add the matching runtime assertion (`assert.*`) so
+   `npm run test:unit` now fails on behavior.
+3. **Green** — implement in `src/*.ts` until both the type check and the test
+   pass.
+4. **Refactor** — with the type system and the tests as the safety net, then
+   `npm run verify` as the definition-of-done gate.
 
-This is why every test in the suite pairs an `expectTypeOf(...)` with an `assert.*` — keep them together. Type-first is also enforced structurally: `npm test` runs `check:tsc` before the test runner, so a wrong type can never be papered over by a passing assertion. Per [AGENTS.md § Never do](./AGENTS.md#never-do), reach green honestly — fix the types so both the type check and the runtime assertion pass, never suppress the ones you can't make pass.
+Every test pairs an `expectTypeOf(...)` with an `assert.*`; keep them together
+— including the expectations inside handler bodies, which pair with an
+assertion on the value dispatch passed, not only the type it inferred (see
+[development/testing.md § Handler arguments](./development/testing.md#handler-arguments)).
+The autocomplete tests (`src/util/__tests__/lsp-completion.test.ts` for the
+helper, `src/primitive-union.test.ts` for the matcher's popup) are the
+exception —
+the language server, not the type system, is the oracle (see
+[development/testing.md § Autocomplete](./development/testing.md#autocomplete)).
+Each test body follows **AAA (Arrange–Act–Assert)** with labeled blocks
+separated by a blank line: `// Arrange` sets up the inputs (e.g. the matcher
+factory), `// Act` exercises the subject once from them (not a second
+throwaway call), `// Assert` holds every check — type expectations first,
+runtime assertions last; an empty block drops its label (see
+[development/testing.md § AAA ordering](./development/testing.md#aaa-ordering)).
+Type-first is enforced structurally: `npm test` runs `check:tsc` before the
+test runner, so a wrong type can never be papered over by a passing assertion.
+Per [AGENTS.md § Never do](./AGENTS.md#never-do), reach green honestly — fix the
+types, never suppress the checks you can't make pass. Full rationale:
+[development/testing.md](./development/testing.md).
+
+## Code style and formatting
+
+`oxfmt` is the formatter and `oxlint` is the linter (with type-aware rules).
+`npm run fix` resolves the fixable issues; `npm run check` verifies without
+writing. Suppressions must be fixed at the root — do not add `oxlint-disable`
+directives or `as` casts to force a green run (see
+[AGENTS.md § Never do](./AGENTS.md#never-do)).
+
+Suggested VSCode extensions are in
+[.vscode/extensions.json](./.vscode/extensions.json); the project's formatter
+and linter are wired up there. Toolchain decisions:
+[development/tooling.md](./development/tooling.md).
+
+## Commit messages
+
+Gitmoji subject, imperative mood, 50/72 wrapping. The template is
+[commit-message-template](./commit-message-template); `npm run setup`
+(or `npm run setup:git-commit-message`) registers it as git's
+`commit.template`. Examples and rationale:
+[development/workflow.md § Commit messages](./development/workflow.md#commit-messages).
+
+## Script prefix convention
+
+Script names in `package.json` use a prefix that signals _when_ the script is
+intended to run. A `<prefix>:<name>` script is implicitly aggregated by a
+`<prefix>` script (if one exists) and run by the corresponding lefthook hook or
+CI step. Pick the prefix that matches the script's lifecycle:
+
+- `create:*` — front doors of the repo's own workflow; these mutate git state
+  rather than the source. `create:branch` opens a unit of work, `create:finish`
+  closes the branch half, `create:release` closes the release half
+  (maintainer-only). No bare `create` aggregator on purpose.
+- `check:*` — read-only verification; never modifies files. Aggregated by
+  `npm run check`.
+- `fix:*` — mutating counterpart of a `check:*` script. Aggregated by
+  `npm run fix`; the diff is the review surface.
+- `test:*` — test scripts. `test` is the canonical entry point (`check:tsc` +
+  unit tests); `test:unit` skips the typecheck for fast local iteration;
+  `test:ci` runs the suite under c8 and fails below 100% coverage on `src/`
+  (CI-only; `verify` stays coverage-free).
+- `watch:*` — long-running watchers for the manual inner dev loop. Aggregated by
+  `watch`.
+- `maintain:*` — advisory repo-maintenance scans: read-only, but whole-project
+  and/or network-bound, so never a correctness gate. Aggregated by
+  `npm run maintain`.
+- `publish:*` — validates the _publishable artifact_ (e.g. `dist/`) rather than
+  the source, so it needs a fresh build.
+- `setup:*` — one-time configuration of a fresh clone; mutates the local
+  environment rather than the repo source, so it is never part of a hook or CI
+  step. Aggregated by `npm run setup`, run once after cloning.
+
+A new script must reuse an existing prefix. If none fits, that's a signal the
+script doesn't belong in the pipeline — not a reason to invent a new prefix. If
+it genuinely does belong, add the prefix to this list in the same commit as its
+first member; an undocumented prefix becomes invisible and quietly accrues
+members. Why `create:` exists, the rejected names, and the design of the bare
+scripts: [development/workflow.md § Script prefix convention](./development/workflow.md#script-prefix-convention).
+
+## Rules the tools don't enforce
+
+CI and review will bounce these even though `npm run check` and the linters
+don't catch them. They're the high-frequency things a contributor (or an agent)
+reaches for by default:
+
+- **Source imports use `.ts` extensions, never `.js`.** `node --strip-types`
+  only resolves the `.ts` form at test time; "pre-fixing" an import to `.js`
+  breaks the inner loop. (why:
+  [development/tooling.md](./development/tooling.md#source-imports-use-ts-extensions))
+- **A new `npm run` script must reuse an existing prefix.** See
+  [Script prefix convention](#script-prefix-convention).
+- **`oxlint-disable` directives live in source, not `.oxlintrc.json`.** The
+  trade-off must sit next to the code it silences. This is a _human_ last-resort
+  convention; agents must not add one, nor edit `.oxlintrc.json` to silence a
+  finding (e.g. `typescript/no-floating-promises`) — see
+  [AGENTS.md § Never do](./AGENTS.md#never-do). (why:
+  [development/tooling.md](./development/tooling.md#oxlint-disable-directives-live-next-to-the-code))
+- **Don't put slow / network / whole-project scans in `check` or pre-commit.**
+  Advisory scans are not correctness gates; they belong under `maintain:`. (why:
+  [development/workflow.md](./development/workflow.md#feedback-tiers))
+- **New work starts with `npm run create:branch`, never a hand-written
+  `git switch -c` / `git checkout -b`.** The command carries the branch
+  precondition; branching around it skips the clean-tree, current-`main` and
+  green-baseline checks, and the skip is invisible until a failure can no longer
+  be attributed. (why:
+  [development/workflow.md](./development/workflow.md#branching-model))
+- **Work is merged back with `npm run create:finish`, never a hand-written
+  `git merge`.** The command carries the merge-side preconditions (clean tree,
+  current `main`, a `feature/`/`fix/`/`chore/` branch) and runs `npm run verify`
+  after the merge, so a merge cannot land unverified. (why:
+  [development/workflow.md](./development/workflow.md#branching-model))
+- **There is no local `npm run publish`, and `publish:publint` / `publish:attw`
+  don't go in `check`.** (why:
+  [development/publishing.md](./development/publishing.md#ci-only-publishing))
+- **A branch ends with a changelog note:** before `npm run create:finish`,
+  summarize the work under `[Unreleased]` in [CHANGELOG.md](./CHANGELOG.md).
+  (why:
+  [development/workflow.md](./development/workflow.md#changelog-notes))
+- **A decision or its rationale belongs in `development/`, not here.** This file
+  holds the actionable rule; `development/<category>.md` holds why, the rejected
+  alternatives and the known issues. When you change a rule, update its category
+  file in the same commit and cross-link the two. (why:
+  [development/README.md](./development/README.md))
+- **Prose in `development/` is extremely concise.** When adding or changing a
+  decision, write fragments if needed — sacrifice grammar for concision. (why:
+  [development/README.md § Decision blocks](./development/README.md#decision-blocks))
 
 ## Branching model
 
-**GitHub Flow (single-developer).** Every change — feature, fix, refactor — branches off `main` and is merged back via a local commit (no PR workflow on Gitea yet). Collaborative review via Gitea UI is not in place — Gitea is the lab; when something is tested and ready for production it will be promoted to GitHub.
+**GitHub Flow (single-developer).** Every change — feature, fix, refactor —
+branches off `main` and is merged back via a local commit. There is no pull
+request workflow on Gitea yet.
 
 - **Base branch:** `main`
 - **Branch naming:** `feature/<desc>` / `fix/<desc>` / `chore/<desc>`
-- **Starting work:** `npm run create:branch -- <prefix>/<desc>`. It refuses, without changing anything, unless the working tree is clean (untracked files included), no merge/rebase/cherry-pick is in progress, `main` matches its upstream, and `npm run test` is green on `main` — so a later failure is always attributable to your edits. The prefix is still _your_ call, inferred from the task; the script validates it rather than guessing it.
+- **Starting work:** `npm run create:branch -- <prefix>/<desc>`. It refuses,
-- **Merging:** `npm run create:finish` (on the branch). It asserts the same clean-tree / no-operation / current-`main` preconditions, fast-forwards a stale `main` (a true divergence is refused), merges the branch `--no-ff`, runs `npm run verify`, and deletes the branch only after the merge is green. The push is deliberately left to `create:release`, so the merge stays local and reviewable — read the diff yourself before finishing.
+  without changing anything, unless the working tree is clean, no
-- CI runs `npm run check` + `npm run test:ci` on every push to `main` — this is the authoritative gate. The one exception: a push headed by a release commit (`:rocket: Release x.y.z`) skips the full `build`/`maintain` jobs, because `create:release` pushes the tag for that exact commit right after and the tag run is the authoritative one (see `release-gate` in [.gitea/workflows/ci.yml](./.gitea/workflows/ci.yml)).
+  merge/rebase/cherry-pick is in progress, `main` is not behind its upstream
-- **Releases are NOT triggered by pushes.** Only the maintainer triggers a release (see [Publishing workflow](#publishing-workflow)).
+  (a local merge not yet pushed is fine — the push belongs to `create:release`),
+  and `npm run test` is green on `main`. The prefix is _your_ call, inferred from
+  the task; the script validates it rather than guessing it.
+- **Merging:** `npm run create:finish` (on the branch). It re-asserts the same
+  preconditions, merges `--no-ff`, runs `npm run verify`, and deletes the branch
+  only after the merge is green. The push is left to `create:release`, so the
+  merge stays local and reviewable.
+- CI runs on every push to `main` — see [Feedback tiers](#feedback-tiers) and
+  [.gitea/workflows/ci.yml](./.gitea/workflows/ci.yml).
+- **Releases are NOT triggered by pushes.** Only the maintainer triggers a
+  release; see [Publishing](#publishing).
 
-## CI runner image
+Full rationale, including the front-door decisions:
+[development/workflow.md § Branching model](./development/workflow.md#branching-model).
 
-The `build` / `maintain` / `publish` jobs run in `gitea.e1nsnull.de/tmu/act-ci:<version>` ([docker/Dockerfile](./docker/Dockerfile)) — the runner's default act image with the Node distribution overlaid at the exact `/opt/hostedtoolcache` layout `actions/setup-node` probes before downloading, so no job pays the ~50 MB fetch. The image tag MUST equal the exact version pinned in `.node-version`, and the image is rebuilt only as part of a Node bump — there is no other trigger. `release-gate` uses no Node and stays on the default image. The script is deliberately NOT an `npm run` script: building requires a docker daemon and registry credentials, so it belongs to no feedback tier — per [Script prefix convention](#script-prefix-convention), no existing prefix fits and that is the signal.
+## Submitting changes
 
-Bumping Node is one coordinated change, committed as a unit:
+There is no pull request workflow on Gitea yet, so a contribution is submitted
+as a branch that is merged locally:
 
-1. Edit `.node-version` to the new exact `x.y.z` — floats like `26` resolve to the latest patch at runtime and silently bust the baked entry; `scripts/runner-image.sh` refuses them.
+1. `npm run create:branch -- <prefix>/<desc>`.
-2. `docker login gitea.e1nsnull.de` (user + package/access token), then `./scripts/runner-image.sh --push` — it reads the version from `.node-version` and builds/pushes `<IMAGE_REPO>:<version>`.
+2. Commit your work (one or more commits, per the tests and style rules above).
-3. Repoint the three `container.image` tags in [.gitea/workflows/ci.yml](./.gitea/workflows/ci.yml) to the same version.
+3. `npm run verify` — the definition of done.
+4. Add a changelog note under `[Unreleased]` (see
+   [Rules the tools don't enforce](#rules-the-tools-dont-enforce)).
+5. `npm run create:finish` to merge the branch into `main` and verify the
+   result.
+6. Present a handover for review. Once there are no further objections, the
+   maintainer pushes.
 
-Skipping step 2 fails CI at image pull; skipping step 3 silently reverts to the per-job download.
+When the project is promoted to GitHub, this step becomes a normal pull request
+against `main`.
 
-Two invariants the image must satisfy for the probe to hit, both easy to break:
+## Publishing
 
-- **The `x64.complete` marker.** `actions/tool-cache` accepts a cached tool only when `<version>/<arch>.complete` exists next to the directory (`tc.find()` checks it); a plausible-looking `node/<version>/x64/` alone is ignored and the download happens anyway. See the comment in [docker/Dockerfile](./docker/Dockerfile).
+Publishing is maintainer-only and CI-only. See
-- **Tag freshness (force-pull deliberately off).** The tag encodes only the Node version, and the image is rebuilt only when that version changes — so the normal flow always yields a new tag, and `act_runner` pulls it. `force_pull` stays disabled (the job log shows `forcePull=false`): it is acceptable to miss a runner-side image change, and forcing a pull would re-pull the image on every job for no benefit. Known issue: a Dockerfile-only change (like the marker above) re-pushed under an unchanged tag is invisible to the runner — it keeps the old image while the registry shows the new digest. If that ever matters, remove the stale tag on the runner host (`docker rmi gitea.e1nsnull.de/tmu/act-ci:<version>`); do not reach for force-pull.
+[development/publishing.md](./development/publishing.md).
-
-## Publishing workflow
-
-Publishing is CI-only by policy. Local `npm publish` is not supported. The maintainer triggers releases from `main`:
-
-1. All intended changes are merged to `main` and passing CI.
-2. The maintainer runs `npm run create:release`. VS Code opens `CHANGELOG.md` to finalize the `[Unreleased]` notes; because pubv refuses a dirty tree, any edit is committed first (then folded into the release commit), and pubv's interactive prompt suggests a version from those notes — the maintainer confirms or edits it.
-3. `scripts/release.sh` creates a single release commit (graduated changelog + package.json bump, amended into one commit), tags it, and pushes everything to Gitea.
-4. CI fires on both pushes: the `publish` job runs on the tag (`build` + publish-tier checks + release page + `npm publish`), while the branch run's `release-gate` job recognizes the release commit and skips `build`/`maintain` — the tag verifies the identical SHA, so no work is duplicated. The job graph lives in [.gitea/workflows/ci.yml](./.gitea/workflows/ci.yml) — keep that file, not this list, as the source of truth. The publish-tier checks must pass before the artifact is published. The `publish` job also creates the Gitea release page from the matching Keep-a-Changelog section (`scripts/release-notes.sh`); it runs _before_ `npm publish` so a broken page fails CI without consuming a version, and `npm publish` stays the last step.
-
-The Gitea release page uses the run's automatic token (`github.token`), so it only needs `contents: write`. `npm publish` is gated on `NPM_TOKEN`, lifted into job-level `env` because `secrets` is not an allowed context in a step `if`: an unset secret skips the publish instead of attempting an unauthenticated one. A tag is all-or-nothing, though — a final `always()` step fails the job unless both the release page and `npm publish` reported `success`, so a skipped or failed npm half turns the job red rather than silently green. Set `NPM_TOKEN` (npm publish rights) under Settings → Actions → Secrets.

README.md

@@ -2,68 +2,64 @@
 
 Pattern matching for TypeScript/ESM environments (F#-style, not regex).
 
-## Development
+## Description
 
-- **Build:** `npm run build`
+`tiny-pattern-ts` brings F#-style pattern matching to TypeScript. Patterns are
-- **Test:** `npm run test`, `npm run test:ci`
+ordinary objects whose `matches` method is a TypeScript type guard, so narrowing
-- **Watch:** `npm run watch`
+composes the way any other guard does. It is deliberately not a regex engine and
-- **Checks:** `npm run check`, `npm run fix`
+not a macro: there is no transpiler and no DSL to learn, and the type-level
-- **Verify:** `npm run verify` — the definition of done
+contract is the feature — see [development/library.md](./development/library.md)
-- **Maintenance:** `npm run maintain` — advisory only
+for the design decisions and [Caveats](#caveats) for the limits.
-- **Individual fixes:** `npm run fix:oxfmt`, `npm run fix:oxlint`
 
-What each tier runs, when it fires and what it costs:
+## Requirements
-[CONTRIBUTING.md § Feedback tiers](./CONTRIBUTING.md#feedback-tiers). How the
-`prefix:` in a script name is chosen:
-[§ Script prefix convention](./CONTRIBUTING.md#script-prefix-convention).
 
-### Tooling
+- **Node.js >= 26** (`engines` field; pinned via `.node-version`).
+- **TypeScript >= 5.0** to consume the published declarations. The emitted `.d.ts`
+  use `const` type parameters (TS 5.0) and keep their relative `.ts` specifiers;
+  both resolve on TS >= 5.0 in `node10` / `node16` / `nodenext` / `bundler`.
+- The package is **ESM-only** (no CommonJS shim).
 
-- **TypeScript 7** — type checker and build (`tsc`).
+## API
-- **node --test** + `--strip-types` — test runner.
-- **c8** — code coverage for `test:ci`.
-- **oxlint** — Rust-based linter, with type-aware rules powered by **oxlint-tsgolint** (typescript-go).
-- **oxfmt** — Rust-based formatter (Prettier-compatible). Formats JS/TS, JSON/JSONC, YAML, Markdown, MDX, and more; built-in `package.json` key sorting replaces `sort-package-json`.
-- **cspell** — spell checking.
-- **knip** — finds unused dependencies, exports, and files.
-- **check-outdated** — reports dependencies behind the registry; it exits non-zero whenever _any_ dependency is outdated.
-- **publint** — validates `package.json` for ESM publishing correctness.
-- **@arethetypeswrong/cli** (`attw`) — validates `.d.ts` declarations against multiple module-resolution scenarios.
-- **lefthook** — git hooks.
-- **@spences10/pi-lsp** — read-only LSP code intelligence for AI coding agents (project-local `.pi/settings.json`). Talks to this repo's TypeScript 7 via `tsc --lsp --stdio`.
 
-Each tool's configuration trade-off is recorded in [Tooling decisions](#tooling-decisions); when it runs is in [CONTRIBUTING.md § Feedback tiers](./CONTRIBUTING.md#feedback-tiers).
+Yet to be implemented
 
-### Tooling decisions
+## Caveats
 
-The choice and configuration of each tool above is the result of deliberate trade-offs, not defaults. The non-obvious ones:
+- **Only finite universes are supported.** The factory must be given a finite
+  union of literals; `string`, `number` and template literals are rejected. This
+  is what lets the exhaustive overload be proven, so the runtime `dispatch`
+  throw stays unreachable through the typed API.
+- **A value and its stringification must not both be present.** Object keys
+  stringify, so a universe containing both a member and the string it
+  stringifies to — `1 | "1"`, `true | "true"`, `null | "null"` — is rejected at
+  the factory. Either form alone is fine, and one value's string form may
+  coexist with a _different_ value's bare form (`"true" | false`).
+- **`symbol` and `bigint` are not supported.** A `symbol` brand is a
+  compile-time phantom with nothing to match at runtime, and a `bigint` is not a
+  valid property key; neither satisfies the matcher's universe constraint.
+- **`NaN` and `-0` cannot be matched specifically.** They have no literal type,
+  so both stay part of `number`.
 
-- **`tsconfig.json` extends `@tsconfig/strictest` + `@tsconfig/node26`**; `tsconfig.build.json` extends it to add the emit-only options (`declaration`, `sourceMap`, `inlineSources`, `outDir`, `target: es2024`, `rewriteRelativeImportExtensions: true`) and to exclude test files. `inlineSources` embeds the original TypeScript in `dist/*.js.map`, so debuggers can map into `src/` without it being shipped; `declarationMap` is intentionally off because a `.d.ts.map` cannot embed source and would dangle. This separation lets the editor and CI type-check from one config while the build emits from the other.
+### Why open universes are rejected
-- **`npm run build` first runs a `prebuild` hook that empties `dist/`.** `tsc` does not prune orphaned emit output — dropping `declarationMap`, for example, left stale `*.d.ts.map` files behind — so the build must start from an empty `dist/` to be reproducible. `prebuild` removes only `dist`; the manual `clean` still resets `dist` + `coverage`, so a local coverage report survives a build.
-- **Source imports use `.ts` extensions** so `node --strip-types` resolves them at test time. `rewriteRelativeImportExtensions: true` in `tsconfig.build.json` rewrites them to `.js` in the emitted JavaScript; the emitted `.d.ts` keep the `.ts` specifier, which TypeScript >= 5.0 resolves (see [Requirements](#requirements)), so no post-processing step is needed.
-- **Type-aware oxlint is enabled declaratively** via `options.typeAware: true` in `.oxlintrc.json` (powered by `oxlint-tsgolint`). The script commands stay clean — no CLI flag — and type-aware mode is a property of the config, not the invocation.
-- **Source-level `oxlint-disable` directives** are used for known type-aware false positives (see `src/pattern.ts`, `src/match.ts`, `src/index.test.ts`). The disable lives next to the code it silences, not in `.oxlintrc.json`, so the trade-off is visible to anyone reading the source.
-- **`knip --include dependencies,exports,files`** intentionally omits the `types` category, which produces systematic false positives for libraries whose exported types are part of the public API. The targeted scope keeps the signal high without config-file boilerplate.
-- **`attw --profile esm-only`** is semantically correct: this package is intentionally ESM-only (no CommonJS shim), so CJS resolution scenarios are out of scope by design, not a bug.
-- **`check:tsc` runs first** in the `npm run check` chain so a type error short-circuits the rest (faster feedback than letting oxlint/oxfmt run and then failing on tsc at the end).
-- **The pre-commit hook sets the `LEFTHOOK_FILES` env var** to the staged-files list, and the affected scripts use `${LEFTHOOK_FILES:-<default>}` to default to the whole project when invoked manually. This keeps `package.json#scripts` as the single source of truth for the underlying commands — `lefthook.yml` only describes _what to run on which files_.
-- **`tslib` and `type-fest` are deliberately not used.** `tslib` is a runtime helper for old ES3/ES5 targets (the project targets ES2024); `type-fest` was never imported. knip caught both.
-- **`@spences10/pi-lsp` is pinned to `0.0.46` and is read-only by design.** The package inspects `node_modules/typescript`, sees major ≥ 7 with no `lib/tsserver.js` (true of the `typescript-go` / `tsgo` port), and spawns the repo's own `tsc --lsp --stdio` binary — no `typescript-language-server` dependency is required. Earlier releases (`≤ 0.0.10`) hard-wire to `typescript-language-server --stdio` and are TS6-only. The tool is _intermediate_ agent feedback (hover, references, definition, symbols, diagnostics); it has no rename / code-action / apply-edit surface, and never a correctness gate — `npm run check` / `verify` remain that. `.pi/settings.json` is the shared, committed declaration; `.pi/npm/` is a gitignored install cache that pi recreates automatically on a trusted startup (it runs `npm install` for any missing project package), so the cache is deliberately not tracked.
 
-### Requirements
+An open universe — one carrying a broad member, as in
+`type Units = "s" | "ms" | "min" | (string & {})` — is not a dispatch concern.
+If the values arrive from outside the program, parse them at the boundary down
+to a finite union and match the narrowed result; the openness never reaches the
+matcher. If the domain is genuinely extensible, the right shape is a runtime
+`Map` of handlers, where "no handler" is a lookup, not a pattern. Either way an
+open matcher would abandon the one guarantee this library exists to give —
+provable exhaustiveness — to automate what a `switch` and a default arm already
+cover. The type-level cost of supporting open universes is recorded in
+[development/library.md](./development/library.md#supported-universes).
 
-- Node.js >= 26 (engines field; pinned via `.node-version`).
+## License
-- TypeScript >= 5.0 to consume the published declarations. The emitted `.d.ts` use `const` type parameters (TS 5.0) and keep their relative `.ts` specifiers; both resolve on TS >= 5.0 in `node10`/`node16`/`nodenext`/`bundler`.
 
-## VSCode integration
+MIT © 2025 tmu. See [LICENSE](./LICENSE).
-
-- Recommended extensions: see `.vscode/extensions.json` (oxc, cspell, TypeScript native-preview, EditorConfig, todo-tasks).
-- TypeScript 7 is used via the `typescriptteam.native-preview` extension.
-- oxc extension provides oxlint squiggles and oxfmt format-on-save; `.vscode/settings.json` pins it per language so a user's local `[language]` formatter settings cannot override the project's choice.
 
 ## Contributing
 
-For maintainer and contributor docs — the script prefix convention, the feedback-tier system, the rules the tools don't enforce, and the publishing workflow — see [CONTRIBUTING.md](./CONTRIBUTING.md). AI coding agents: your entry point is [AGENTS.md](./AGENTS.md), which points back to CONTRIBUTING.md.
+Contributions are documented in [CONTRIBUTING.md](./CONTRIBUTING.md); the
-
+reasons behind the project's decisions, rejected alternatives, and known issues
-- Commit signing (GPG).
+live in [development/](./development/README.md). AI coding agents start at
-- Type-only tests use `expect-type`'s `expectTypeOf(...)` inside `node --test` cases.
+[AGENTS.md](./AGENTS.md).

backlog.tasks

@@ -5,38 +5,36 @@ Backlog and tracking for tiny-pattern-ts. Managed in vscode-todotasks format.
 ---
 
 Setup:
-✔ Add gitea release page in CI @high @done
+☐ Split off template into separate package => pi --session 01a07dde-7050-7054-bb36-1606d7eb2bc3 @low
-✔ Manually verify the Gitea release page on a real tag push (needs main) @high @done (9/15/2026, 1:18:15 PM)
-☐ Split off template into separate package => pi --session 01a07dde-7050-7054-bb36-1606d7eb2bc3 @high
 
 v1.0:
 ☐ API surface is stable and fully typed
     ☐ Finalize public exports in `src/index.ts`
     ☐ Document all exported types and functions
     ☐ Add JSDoc for public APIs
-☐ Test coverage meets threshold
+✔ Test coverage meets threshold @done
-    ☐ Achieve 100% branch coverage on `src/pattern.ts`
+    ✔ Achieve 100% branch coverage on `src/primitive-union.ts` @done
-    ☐ Achieve 100% branch coverage on `src/match.ts`
+    ✔ Achieve 100% branch coverage on `src/index.ts` @done
-    ☐ Achieve 100% branch coverage on `src/index.ts`
 
-Bugs:
+Matcher:
-
+✔ when using a union type as a property, the current behavior of tagged union matcher is @done
-Enhancements:
+    to pass never to handler parameters
+    → new matcher function needed or can be fixed in tagged union matcher
+    ✔ optional discriminant (`{ type?: "x" }`) is the same hole: the boolean/nullish change now admits the `undefined` tag, so the factory accepts the key, but `Extract<T, Record<K, V>>` still passes `never` to both the `x` and `undefined` handlers @done
 
 Documentation:
-☐ Clean up CONTRIBUTING.md and README.md, create docs
+☐ Bring README.md back to its previous form — synopsis and examples restored, in the correct place
-☐ Add usage examples to README.md
+    → previous section order: title, tagline, Synopsis, Description, Requirements, Examples, API, License, Contributing
+    → previous Examples order: literal/exhaustive, typeof, structural/discriminated unions, when, any
 ☐ Create `examples/` directory with runnable snippets
-☐ Add comparison section vs. other TS pattern-matching libs
+☐ Add comparison section vs. other TS pattern-matching libs in Readme.md
 ☐ Write migration guide for users coming from discriminated unions
 ☐ Create backlog tasks for implementation
+☐ Validate code fences in Markdown (start with README.md) — compile the TypeScript examples against `src/` so the docs cannot drift from the API
 
 Maintenance:
 ☐ Serve CI coverage over a tiny self-hosted webserver (replace the zip artifact) @low
-    ✔ Add a minimal dir-listing webserver to the gitea docker setup (e.g. caddy `file_server browse` reusing the existing reverse proxy, or any single-binary static server, lipanski/docker-static-website) @done (9/13/2026, 9:02:37 PM)
-    ✔ drop the `actions/upload-artifact` coverage step in favour of the shared-dir layout @done (9/13/2026, 10:37:22 PM)
     ☐ Explore serving coverage for non-tag pushes (e.g. `main/coverage`, PR previews) @low
-    ✔ Manually verify the coverage was created on a real tag push (needs main) @low @done (9/14/2026, 1:55:03 PM)
     → design: no deploy step in CI; the webserver just exposes the shared directory (decided over Gitea Pages / Codecov — neither confirmed available/ wanted)
 ☐ serve docs over self hosted server @low
     ☐ Add a minimal dir-listing webserver to the gitea docker setup for serving docs (reuse existing reverse proxy)
@@ -46,19 +44,4 @@ Maintenance:
     ☐ Add a minimal dir-listing webserver to the gitea docker setup for serving landing page (reuse existing reverse proxy)
     ☐ CI writes landing page to a shared volume keyed by project + tag (e.g. `/landing/tiny-pattern-ts/<tag>/`)
     ☐ Browse to `…/tiny-pattern-ts/index.html` in the browser
-✔ Stop Gitea CI re-downloading Node on every job @done
+☐ Add testing with TypeScript 5.0 baseline in CI
-    ✔ Share the warm npm cache with the publish job @done
-    ✔ Bake Node into the CI job image (docker/Dockerfile, container.image in ci.yml) @done
-    ✔ Build/push gitea.e1nsnull.de/tmu/act-ci:26.8.2 and confirm setup-node skips the download @done
-    ✔ Write the `<version>/x64.complete` marker — actions/tool-cache ignores a bare directory, so the probe missed and the download continued @done
-    ✔ Log tool-cache state from the job container to find it (temporary, removed once understood) @done
-    ✔ Guard the invariant in CI (`Assert the baked tool cache is present`) @done
-✘ Enable force-pull for the runner so a changed act-ci image is never missed @low @cancelled
-    → decided against: it is acceptable to miss a runner-side image change, and the image is only rebuilt on a Node bump, which changes the tag anyway. A Dockerfile-only change re-pushed under an unchanged tag is a known issue with a manual `docker rmi` workaround (see CONTRIBUTING § CI runner image).
-✔ Improve CI publish @done
-    ✔ Check whether publish job is only run on tags, if not, guard it @done
-    ✔ Gate only single steps @done
-        ✔ Do not publish to npm, if NPM_TOKEN is not set (e.g. PRs from forks) @done
-        ✔ Do not publish to Gitea — uses the run's automatic `github.token`, so no secret gate is needed @done
-        ✔ Otherwise run the steps @done
-        ✔ Fail the job unless both the Gitea release and npm publish succeeded @done

cspell.json

@@ -42,7 +42,10 @@
         "bestikk",
         "silverwind",
         "idris",
-        "todotasks"
+        "todotasks",
+        "connor",
+        "injective",
+        "injectivity"
     ],
     "ignorePaths": ["dist", "node_modules", "coverage", "*.svg", ".gitignore"]
 }

development/README.md

@@ -0,0 +1,81 @@
+# Development documentation
+
+Why this project works the way it does: the decisions, what was rejected, and
+the shortcomings and known issues we carry. Written for maintainers and
+contributors.
+
+The actionable rules — setup, running, testing, submitting — live in
+[CONTRIBUTING.md](../CONTRIBUTING.md). **Each fact is written once**: the rule
+there, the reason here; neither restates the other, and where a fact is useful
+in both they link. Read the relevant file before changing an area, and when a
+rule changes update its rationale here in the same commit.
+
+User-facing documentation is [README.md](../README.md). `docs/` is deliberately
+unused: that name is reserved for the future user documentation site, and
+deploying it is out of scope. These files are not part of that site.
+
+## Layout
+
+One file per category:
+
+| File                             | Covers                                                                  |
+| -------------------------------- | ----------------------------------------------------------------------- |
+| [library.md](./library.md)       | Public API design, the type-level contract, and its limitations         |
+| [workflow.md](./workflow.md)     | Branching and merging, script prefixes, feedback tiers, commit messages |
+| [tooling.md](./tooling.md)       | Toolchain choices and configuration, editor setup                       |
+| [testing.md](./testing.md)       | Test strategy and type-driven development                               |
+| [ci.md](./ci.md)                 | CI pipeline, runner image, coverage serving                             |
+| [publishing.md](./publishing.md) | Release and npm publishing                                              |
+
+We start with one file per category so each area stays small enough to hold in
+mind; a category that outgrows it becomes a folder with an index, and the links
+in CONTRIBUTING.md and README.md point at the category, not a single decision.
+
+## Decision blocks
+
+Record every non-obvious choice as a block in the relevant category file:
+
+```md
+## Runner image
+
+#### Decision (2026-09)
+
+Bake Node into the CI job image at the setup-node tool-cache layout instead
+of downloading per job.
+
+#### Why
+
+- ...
+
+#### Rejected
+
+- Gitea Pages / per-job download
+- force-pull
+
+#### Known issue
+
+- a Dockerfile-only change re-pushed under an unchanged tag is invisible to the runner
+    - recover with `docker rmi <image>`
+```
+
+- The date is the month the decision was made, not when the file was edited —
+  the anchor for "current" versus "was current once".
+- `Rejected` stops the project re-litigating the same alternatives; an empty one
+  usually means they were never written down.
+- `Known issue` is where shortcomings live. A caveat not tied to one decision
+  goes under a `## Known issues` section at the end of the file.
+- Replace a superseded decision in place rather than archiving it; git history
+  is the archive.
+- Terse is the point: humans skim and agents imitate the style already in the
+  file, so verbosity compounds edit over edit. Grammar loses to density here on
+  purpose.
+
+## Adding to these docs
+
+1. Pick the category: `library`, `workflow`, `tooling`, `testing`, `ci`,
+   `publishing`.
+2. Add or update a decision block; keep existing text unless the decision
+   changed.
+3. If an actionable rule changes, update
+   [CONTRIBUTING.md](../CONTRIBUTING.md) in the same commit and cross-link.
+   Never change a rule there without updating its rationale here.

development/ci.md

@@ -0,0 +1,169 @@
+# CI
+
+[.gitea/workflows/ci.yml](../.gitea/workflows/ci.yml) is the source of truth for
+the job graph; this file records why it is shaped the way it is.
+
+## Pipeline
+
+- **`build`** (push to `main` / tag) — build + correctness + packaging.
+- **`maintain`** (push to `main`, non-blocking) — `npm run maintain`; reports,
+  never fails the build.
+- **`publish`** (tag) — packaging checks + `publish:publint` / `publish:attw`,
+  then the Gitea release page and `npm publish` (see
+  [publishing.md](./publishing.md)).
+- **`release-gate`** — on a `:rocket: Release x.y.z` commit it skips
+  `build`/`maintain`, because `create:release` pushes the tag for the same commit
+  right after and the tag run is authoritative. It uses no Node and stays on the
+  default image.
+
+## Runner image
+
+#### Decision (2026-09)
+
+`build` / `maintain` / `publish` run in `gitea.e1nsnull.de/tmu/act-ci:<version>`
+([docker/Dockerfile](../docker/Dockerfile)): the default act image with Node
+overlaid at the exact `/opt/hostedtoolcache` layout `actions/setup-node` probes.
+
+#### Why
+
+- No job pays the ~50 MB Node fetch, because the probe hits the baked entry.
+- The tag must equal the exact [.node-version](../.node-version) pin, and the
+  image is rebuilt only as part of a Node bump — there is no other trigger.
+- Building needs a docker daemon and registry credentials, so it belongs to no
+  feedback tier. That is why it is **not** an `npm run` script: no
+  [prefix](./workflow.md#script-prefix-convention) fits, and that is the signal.
+
+#### Rejected
+
+- Downloading Node in every job — the ~50 MB fetch was the original problem.
+- Caching Proxy (Squid or similar) — adds complexity to global setup
+- Mounting the tool cache - No invalidation will fill the cache with stale versions
+
+## Bumping Node
+
+Bumping Node is one coordinated change, committed as a unit:
+
+1. Edit [.node-version](../.node-version) to the exact `x.y.z` — floats like `26`
+   resolve to the latest patch at runtime and bust the baked entry, so
+   [scripts/runner-image.sh](../scripts/runner-image.sh) refuses them.
+2. `docker login gitea.e1nsnull.de` (user + package/access token), then
+   `./scripts/runner-image.sh --push`, which reads the version and pushes
+   `<IMAGE_REPO>:<version>`.
+3. Repoint the three `container.image` tags in
+   [.gitea/workflows/ci.yml](../.gitea/workflows/ci.yml) to that version.
+
+Skipping step 2 fails CI at image pull; skipping step 3 silently reverts to the
+per-job download.
+
+## Image invariants
+
+For the `setup-node` probe to hit, two things must hold — both easy to break:
+
+### The `x64.complete` marker
+
+#### Decision (2026-09)
+
+Bake a `<version>/<arch>.complete` marker next to the Node directory.
+
+#### Why
+
+- `actions/tool-cache` accepts a cached tool only when
+  `<version>/<arch>.complete` exists beside it (`tc.find()` checks). A bare
+  `node/<version>/x64/` is ignored and the download happens anyway. See the
+  comment in [docker/Dockerfile](../docker/Dockerfile).
+
+### Tag freshness, with force-pull deliberately off
+
+#### Decision (2026-09)
+
+Leave `act_runner`'s `force_pull` disabled.
+
+#### Why
+
+- The tag encodes only the Node version, and the image is rebuilt only when that
+  changes — so the normal flow always yields a new tag and the runner pulls it.
+- Forcing a pull re-pulls the image on every job for no benefit.
+
+#### Rejected
+
+- Enabling `force_pull`: it is acceptable to miss a runner-side image change,
+  and a `Dockerfile`-only change is not worth a per-job pull.
+
+#### Known issue
+
+- A `Dockerfile`-only change (like the marker above) re-pushed under an
+  unchanged tag is invisible to the runner, which keeps the old image while the
+  registry shows the new digest. Remove the stale tag on the runner host
+  (`docker rmi gitea.e1nsnull.de/tmu/act-ci:<version>`); do not reach for
+  force-pull.
+
+## Coverage threshold
+
+#### Decision (2026-09)
+
+`npm run test:ci` fails below 100% statements / branches / functions / lines
+across `src/**/*.ts` (`c8 --all --include "src/**/*.ts" --100`). The gate rides
+the `build` job; `npm run verify` stays coverage-free.
+
+#### Why
+
+- The types are the feature, so an untested branch is a hole in the contract,
+  not a metric to trade off; 100% is the only threshold that means "no hole".
+- `--all` counts a `src/` file no test imports. Without it c8 reports only the
+  files the suite happened to load, so a new untested module is invisible and
+  the threshold passes vacuously.
+- The gate rides `test:ci`, which `build` already runs — no new job or step.
+- `verify` stays fast and local; the slower coverage run is a CI-only tier (see
+  [workflow.md § Feedback tiers](./workflow.md#feedback-tiers)).
+
+#### Rejected
+
+- Per-file thresholds: a global 100% already forces every counted file to 100%.
+- A `check:coverage` script: it would re-run the suite or read c8's temp dir,
+  and no `check:*` script runs tests.
+- `--all` without `--include`: it would also sweep `scripts/`, which is not the
+  shipped surface.
+
+#### Known issue
+
+- `src/matcher-shared.ts` is types only, so its runtime image is empty; c8 still
+  lists it under `--all`. It carries a file-level `/* c8 ignore start */` with
+  the reason. Adding runtime code there means removing that directive.
+
+## Coverage serving
+
+#### Decision (2026-09)
+
+Serve CI coverage from a shared directory on the runner, with no deploy step in
+CI.
+
+#### Why
+
+- The webserver exposes the shared directory and the Gitea docker setup reuses
+  the existing reverse proxy — no upload artifact, no external service.
+- Coverage is written to a shared volume keyed by project and tag (for example
+  `/docs/tiny-pattern-ts/<tag>/`).
+
+#### Rejected
+
+- Gitea Pages and Codecov: neither was confirmed available or wanted.
+
+#### Known issue
+
+- Coverage is served for tag pushes only; non-tag pushes (for example
+  `main/coverage`) are tracked separately.
+
+[scripts/precompress.ts](../scripts/precompress.ts) emits `.br` / `.gz` / `.zst`
+sidecars next to text assets. The Gitea pages service (`static-web-server` with
+`SERVER_COMPRESSION_STATIC=true`) serves the sidecar matching `Accept-Encoding`
+and falls back to the original.
+
+#### Decision (2026-09)
+
+Precompress into sidecars rather than per request.
+
+#### Why
+
+- The assets are static and change only on deploy, so the work is paid once.
+- Images, fonts and archives are already compressed; a sidecar would only grow
+  them, so only text extensions are emitted.

development/library.md

@@ -0,0 +1,280 @@
+# Library design
+
+The type-level design of the public API and the limitations it carries. The
+user-facing reference is [README § API](../README.md#api).
+
+The matchers are implemented in `src/primitive-union.ts` (`getPrimitiveUnionMatcher` /
+`getPrimitiveUnionMatcherW`) and `src/tagged-union.ts` (`getTaggedUnionMatcher` /
+`getTaggedUnionMatcherW`), re-exported from `src/index.ts`; the rest of the
+library is placeholder code.
+
+## Matcher shape
+
+#### Decision (2026-09)
+
+A factory takes the universe and returns a builder; the builder takes a handler
+map and an optional fallback:
+
+```ts
+const matcher = getPrimitiveUnionMatcher<"a" | "b">()({ a: (s) => …, b: (s) => … });
+const fallback = getPrimitiveUnionMatcher<"a" | "b" | "c">()({ a: (s) => … }, (s) => …);
+```
+
+Exhaustive or fallback is decided **at the call site**, by whether the second
+argument is present. The fallback's parameter is the remainder
+`Exclude<T, keyof Handled>`. Only the return-strictness axis remains, so there
+are two factories:
+
+- `getPrimitiveUnionMatcher` — one common `R`; the fallback must fit it;
+- `getPrimitiveUnionMatcherW` — the union `PatternReturns<Handled> | R`.
+
+Each factory is two overloads whose order is load-bearing:
+
+1. `Handlers<T, R>` — the exhaustive form, and the contextual type of the
+   handler-map popup;
+2. `Handled extends Exact<Partial<Handlers<T, R>>, Handled>` intersected with
+   `MustBePartial<T, Handled>`, plus `Fallback<T, Handled, R>` — a partial
+   handler map plus the fallback, rejected when the map already covers `T`.
+
+#### Why
+
+- **The fallback is an argument, not a property.** TypeScript fixes a property's
+  contextual type before it infers its sibling keys, so `_: (s) => …` in the
+  handler map can only see all of `T`, never `Exclude<T, keyof Handled>`. A later
+  argument is contextually typed from inference on an earlier one, so the split
+  is what makes the remainder expressible.
+- **The redundant-fallback guard is an F-bounded constraint.** A map that
+  already covers `T` plus a fallback is rejected by folding
+  `MustBePartial<T, Handled>` into `Handled`'s own constraint. The guard is
+  checked _after_ `Handled` is inferred, so the contextual pass that types the
+  handler callbacks survives. The obvious conditional
+  `Exclude<T, keyof Handled> extends never ? …` in the fallback's parameter
+  type is evaluated while `Handled` is still its constraint and rejects every
+  partial map whose callbacks are context-sensitive.
+- **Overload order keeps both messages.** #1 supplies the contextual type
+  (`a, b, c`); #2 accepts a partial map once a fallback is present, so its popup
+  is optional (`a?, b?, c?`). A gap without a fallback is reported against #1.
+- **`R` needs an inference site.** `R` inside the `Exact<…>` constraint is not
+  one, so `handlers: Handled & Partial<Handlers<T, R>>` re-adds it; without that
+  `R` collapses to `unknown` when the handler params are inferred.
+- **`Exact` restores the excess-property check.** TypeScript skips it for a
+  generic constraint, so without `Exact` the handler map accepts keys outside
+  `T`.
+- Two factories, not four: the fallback is an argument, not a separate API.
+
+#### Rejected
+
+- **Single-object `_`** (the former shape). `_` sees only all of `T`; the
+  remainder is not expressible there, and an exhaustive map plus `_` was
+  accepted.
+- **Curried handlers-first** — `(handlers)(fallback)`. Rejected: two calls for
+  the common case. It is not needed for the redundant-fallback guard, which the
+  F-bounded constraint already provides (see Why).
+- **`this` / HKT self-reference.** `this` is post-construction (method bodies,
+  return positions); a parameter's contextual type is pre-construction.
+  `keyof this` in an interface method is the interface, not the literal.
+- **Variance / `const` type parameters / `NoInfer` / `unique symbol` brands /
+  defaulted type-param guards.** None change inference or evaluation order;
+  `in`/`out` on the handler map broke contextual typing outright. `NoInfer`
+  specifically leaks into the emitted `.d.ts`, raising the consumer floor to
+  TypeScript 5.4 (README promises `>= 5.0`).
+- **Union merge**, **overload merge with only the exhaustive arm last**,
+  **inferred universe**, **conditional `RequireKeys`**, **cases-first curried** —
+  decided against while the API was single-object; their reasons (reported
+  near-miss member, no `_` in the exhaustive popup, `NoInfer`/floor, `keyof P`
+  counts optional keys, not pipe-friendly) hold where they still apply.
+
+#### Known issue
+
+- `PatternReturns` must be
+  `ReturnType<Extract<ValueOf<P>, (...args: never[]) => unknown>>` so it survives
+  the closed, partly-optional `P` constraints.
+- `Parameters<typeof factory>[0]` resolves only the **last** overload, so it is
+  not a sound "rejected" oracle for a factory. Factory-negative tests use
+  `@ts-expect-error` call sites (the test file only — the general ban stands).
+
+## Shared internals
+
+#### Decision (2026-09)
+
+`src/matcher-shared.ts` holds the universe-agnostic pieces both matchers use:
+`UnaryFn`, `PatternReturns`, `RedundantFallback`, `HandlerMap`, the shared
+`Matchable` universe, the `PatternKey` key projection and its `Member` inverse,
+and the `Stringified` / `Collisions` / `UnsupportedReason` / `UnsupportedUniverse`
+/ `UniverseGate` universe gate.
+
+#### Why
+
+- `RedundantFallback`'s property name is the diagnostic, so one definition
+  keeps the two matchers' message from drifting; the other pieces appear
+  verbatim in both public signatures or are the same projection over each
+  matcher's universe.
+- **`Matchable` is one definition, not two.** The primitive-union matcher's
+  universe and the tagged-union matcher's allowed `Tag` values are the same set,
+  so aliasing them keeps the two matchers from drifting apart on what they
+  accept (`symbol`/`bigint` rejected once).
+
+#### Rejected
+
+- **A generic `Matcher<Universe>` over the interface pair, `Handlers`,
+  `Fallback` and `MustBePartial`.** Each is built from its own universe
+  (`Tags`/`MapTaggedUnion` vs the primitive values); abstracting over the
+  F-bounded `Handled` constraint that makes the remainder work risks the
+  contextual typing it exists to preserve. `Matchable`, the `PatternKey` /
+  `Member` projection and the `UniverseGate` are the pieces both universes
+  genuinely share.
+
+## Supported universes
+
+#### Decision (2026-09)
+
+A universe must be a **finite union of literals** with **no
+value/stringification collision**. Broad types (`string`, `number`, a template
+literal) and `"true" | true` / `1 | "1"` are rejected; the factory intersects
+`UniverseGate<T>` (the `UnsupportedUniverse<Reason>` diagnostic) into the
+handler and fallback parameters. `Member<T, K>` replaces the `PatternParam<K>`
+inversion: the handler parameter is the member(s) of `T` whose `PatternKey` is
+`K`, so a standalone `"true"` is `"true"`, not `true`.
+
+#### Why
+
+- **`PatternKey` is not injective.** `"true"` and `true` (and `1` / `"1"`)
+  share a runtime key, so `PatternParam<K>` cannot recover the member.
+  `Member<T, K>` inverts against `T`, which is exact.
+- **Broad types cannot be proven exhaustive.** An index-like map lets a partial
+  object satisfy the exhaustive overload and reaches the `dispatch` throw.
+  Rejecting at the boundary avoids threading an open/closed branch through
+  `Handlers`, `Fallback` and `MustBePartial`.
+- **The finite-literal predicate is `IsLiteral<PatternKey<T>> extends true`.**
+  `IsLiteral` is `boolean` for a union that mixes a literal with a broad type
+  (`"a" | \`x-${number}\``), so `extends false`would treat the mix as
+supported;`extends true` is the check that rejects it.
+- **Collisions are rejected, not merged.** `Member<T, K>` would be sound (the
+  handler gets the union), but the API is one handler per member; rejecting
+  keeps `Member` a singleton and the remainder exact.
+- **The collision predicate is type-checkable.** `Collisions<T> =
+Extract<T, Stringified<T>>` catches numeric collisions too.
+- **The gate is an intersection, not a branch,** so `R` inference and the popup
+  survive; a conditional parameter type would not.
+
+#### Rejected
+
+- **Open universes with a required fallback** (`fix/open-universe-*`): sound,
+  but left the collision hole and added an `IsLiteral` /
+  `OpenUniverseNeedsFallback` branch through every handler type. Findings, kept
+  so they are not re-run: `{}` satisfies an index signature (and `Exact` misses
+  it); an index signature dominates contextual typing; `R` infers only from a
+  non-self-referential parameter type (`{ [K in keyof H]: … R … }` gives
+  `unknown`); an F-bounded guard referencing `keyof Handled` in `Handled`'s own
+  constraint sees the constraint, not the map; all handlers share one `R` (only
+  the fallback widens it); `IsLiteral` is the finite-literal predicate. The
+  user-facing consequence — open universes are a parsing or registry concern,
+  not a dispatch one — is guidance in README § Why open universes are rejected.
+- **`Member<T, K>` without the gate:** sound, but a colliding handler gets a
+  union and `1 | "1"` stays one runtime key.
+- **A round-trip injectivity gate** (`IsEqual<T, PatternParam<PatternKey<T>>>`):
+  over-rejects standalone `"true"` / `"false"` / `"null"` / `"undefined"`.
+- **A case-list / ts-pattern builder:** removes the collision class but drops
+  the object map (footprint, popup) and reimplements an existing library.
+- **Normalize numeric keys to strings:** makes `PatternKey` injective but
+  changes "numeric keys stay numbers" and defeats the numeric dispatch fast
+  path.
+
+#### Known issue
+
+- A multi-collision universe lists every collision in the diagnostic.
+- The `dispatch` throw is unreachable through the typed API; the throw tests
+  widen the factory to `Function` to reach it.
+
+## Tagged-union matcher
+
+#### Decision (2026-09)
+
+`getTaggedUnionMatcher` / `getTaggedUnionMatcherW` mirror the primitive-union pair
+with one extra curried step for the discriminant key:
+
+```ts
+type Shape =
+    | { kind: "circle"; radius: number }
+    | { kind: "square"; side: number };
+
+const area = getTaggedUnionMatcher<Shape>()("kind")({
+    circle: (s) => Math.PI * s.radius ** 2,
+    square: (s) => s.side ** 2,
+});
+const fallback = getTaggedUnionMatcher<Shape>()("kind")(
+    { circle: (s) => … },
+    (s) => …, // s: { kind: "square"; side: number }
+);
+```
+
+The key is a separate call because `K` is inferred from its literal argument and
+`T` is fixed by the first factory; one call could not infer both.
+`Discriminated<T>` restricts the key to properties whose values are tags.
+
+#### Why
+
+- **Same fallback/remainder machinery as the primitive-union matcher.** `HandledTags`
+  recovers the tag values the map handled (`Member<Tags, keyof Handled>`) and
+  `Narrowed<T, K, Exclude<Tags, …>>` is the fallback's parameter; the
+  redundant-fallback guard is the same F-bounded constraint. Only the "universe"
+  changes — `T`'s members instead of primitive values.
+- **`T extends object`, not `Record<PropertyKey, unknown>`.** An `interface` has
+  no implicit index signature, so the `Record` constraint would reject
+  interface-based unions. The runtime reads the tag off `object` with one
+  assertion, the tagged twin of the primitive-union dispatch's `shape as string | number`.
+- **`Narrowed` distributes over `T`, narrowing `K` to the tag.** A member whose
+  `K` cannot take the tag drops out; a duplicated tag yields a union of members
+  instead of dropping one. `T[K] extends V` returns the exact member (a
+  discriminated union's declared interface) untouched, so the mapped form only
+  handles a property that is itself a union.
+- **A union-valued or optional discriminant is supported.** A single shape whose
+  property is a union (`{ color: "red" | "green" | "blue" }`) is narrowed per
+  handler instead of being passed `never`. Matching a defined tag on an optional
+  property (`{ type?: "x" }`) proves the key is present, so it becomes required
+  (`{ type: "x" }`); the `undefined` tag narrows it to `{ type?: never }` under
+  `exactOptionalPropertyTypes` (absence) or `{ type?: undefined }` when the
+  property explicitly admits `undefined`.
+- **A `boolean` / `null` / `undefined` tag goes through the shared
+  `PatternKey` / `Member` projection.** `Discriminated` admits those tags
+  (they are in `Tag`), but they cannot key a mapped type, so the handler map is
+  keyed by the stringified form (`true` → `"true"`) and `Member` inverts it
+  against the tag set to recover the member. This is the same projection the
+  primitive-union matcher uses over its universe, which is why it lives in
+  `matcher-shared.ts`.
+
+#### Known issue
+
+- A tag need not be unique across the union. Two members sharing one is not a
+  soundness hole: they select a single runtime key, so one handler receiving
+  their union is the only correct behavior — the key is simply not a
+  discriminant. The gate rejects only the distinct-value collision
+  (`true | "true"`), where two values share a key and `Member` can no longer
+  invert it; see § Supported universes. Pinned by the duplicate-tag tests in
+  `src/tagged-union.test.ts`.
+
+## Primitive universe
+
+#### Decision (2026-09)
+
+The universe (`Matchable`) is `string | number | boolean | null | undefined`,
+with `boolean` admitted as `true | false`.
+
+`boolean`/`null`/`undefined` are not property keys, so handler-map keys are a
+projection (`PatternKey`: each member stringified) and `Member` inverts it
+against the universe, so callbacks receive the real member (`true`, not
+`"true"`; the standalone string `"true"` stays `"true"`). The popup offers
+`true`, `false`, `null`, `undefined` by name (verified over LSP). The same
+projection is shared with the tagged-union matcher; see § Tagged-union matcher.
+The supported universes are constrained as described in § Supported universes.
+
+#### Why
+
+- Runtime dispatch indexes with the raw `shape`; `handlers[true]` coerces to
+  `"true"` at runtime exactly as `String` would. The `shape as string | number`
+  assertion only placates `TS2538` and buys the number fast path (an explicit
+  `String()` defeats V8's numeric-key path: measured ~2× on number-keyed
+  dispatch).
+- `symbol`/`bigint`/`NaN`/`-0` are rejected, and a member colliding with its
+  stringification is rejected by the gate: user-facing, stated once in
+  [README § Caveats](../README.md#caveats).

development/publishing.md

@@ -0,0 +1,131 @@
+# Publishing
+
+Publishing is CI-only: local `npm publish` is not supported, and the maintainer
+triggers releases from `main`. The mechanics are in
+[scripts/release.sh](../scripts/release.sh) and
+[scripts/release-notes.sh](../scripts/release-notes.sh); the job graph is
+[.gitea/workflows/ci.yml](../.gitea/workflows/ci.yml).
+
+## Release steps
+
+1. All intended changes are merged to `main` and passing CI.
+2. The maintainer runs `npm run create:release`. VS Code opens
+   [CHANGELOG.md](../CHANGELOG.md) to finalize the `[Unreleased]` notes; because
+   pubv refuses a dirty tree, the edit is committed first (then folded into the
+   release commit), and pubv suggests a version from those notes to confirm or
+   edit.
+3. `scripts/release.sh` creates one release commit (graduated changelog +
+   `package.json` bump, amended together), tags it, and pushes.
+4. CI fires on both pushes. `publish` runs on the tag (build + publish checks +
+   release page + `npm publish`), while `release-gate` recognizes the release
+   commit and skips `build`/`maintain`: the tag verifies the identical SHA, so no
+   work is duplicated. The publish checks pass before the artifact is published,
+   and the release page is created from the matching Keep-a-Changelog section
+   _before_ `npm publish`, so a broken page fails CI without consuming a version
+   and `npm publish` stays the last step.
+
+## CI-only publishing
+
+#### Decision (2026-09)
+
+Releases are cut by CI from `main`; there is no local `npm publish` and no
+`publish:*` script in the `check` chain.
+
+#### Why
+
+- The tag is the artifact marker: CI verifies the exact commit it points at, so
+  a local publish could ship something the tag does not describe.
+- `publish:publint` / `publish:attw` validate the _publishable artifact_, which
+  needs a fresh build; they are not source-correctness checks and do not belong
+  in `check`.
+
+## The release commit is assembled from two tools
+
+#### Decision (2026-09)
+
+`create:release` uses `pubv` for the changelog graduation and bump heuristic,
+then `npm version` for the `package.json` + lockfile bump, amended into a single
+release commit.
+
+#### Why
+
+- We want hand-written Keep-a-Changelog notes, an `[Unreleased]` ->
+  `## [x.y.z] - DATE` graduation, and a tag on the exact commit that gets
+  published — and no single tool did both the graduation and the `package.json`
+  bump.
+- Split by strength: `pubv` (tiny, changelog-driven) owns preflight, the
+  interactive major/minor/patch heuristic, and graduating and committing
+  `CHANGELOG.md` (no tag, no push); `npm version` syncs `package.json` + the
+  lockfile; `--amend` folds them into one commit; the tag is created _after_ the
+  amend so it is never orphaned.
+- The notes are finalized _before_ pubv because its bump heuristic reads the
+  `[Unreleased]` body — editing afterwards would inform the changelog only, not
+  the version. The staging commit that satisfies pubv's clean-tree check is
+  folded back into the single release commit.
+
+#### Rejected
+
+- The conventional-commits family: the history is gitmoji, not Conventional, and
+  the notes are hand-written (see
+  [workflow.md § Commit messages](./workflow.md#commit-messages)).
+- `changesets` / `rtk`: config plus a heavier flow that fights the CI-only
+  publish.
+- `knope` / `kacl` / `bestikk`: changelog-only (no `package.json` bump) and
+  5-year / 2-year / brand-new maintenance.
+- `pubv` alone: it never writes `package.json`.
+- `versions` (silverwind): good Gitea support, but pairing it with a hand-rolled
+  promote became a ~180-line script, which this ~30-line version replaces.
+
+## The version has one source of truth
+
+#### Decision (2026-09)
+
+The version is derived from the graduated `## [x.y.z]` heading in
+[CHANGELOG.md](../CHANGELOG.md) and written to `package.json` +
+`package-lock.json` by `npm version`.
+
+#### Why
+
+- `release.sh` reads the version from the changelog, so the changelog is the
+  input and `package.json` the derived copy — one direction, no drift.
+
+#### Rejected
+
+- A `## Version` line in the README: it makes `release.sh` responsible for a
+  third file.
+- Linking `package.json` from the README: it invites a hand-maintained duplicate
+  the link does not keep in sync.
+
+## Release notes are extracted from the changelog
+
+#### Decision (2026-09)
+
+`scripts/release-notes.sh <tag>` prints the Keep-a-Changelog section for the tag
+and exits non-zero when it is missing.
+
+#### Why
+
+- CI reuses the release body from the same file that drove the version, so the
+  page and the changelog cannot disagree.
+- Failing on a missing section means a release can never publish an empty body.
+  A leading `v` is tolerated so both `v1.2.3` and `1.2.3` match `## [1.2.3]`.
+
+## Token gates
+
+#### Decision (2026-09)
+
+The Gitea release page uses the run's automatic token (`github.token`).
+`npm publish` is gated on `NPM_TOKEN`, lifted into job-level `env`. A final
+`always()` step fails the job unless both halves reported `success`.
+
+#### Why
+
+- The automatic token needs only `contents: write`, so the release page needs no
+  secret gate.
+- `secrets` is not allowed in a step `if`, so `NPM_TOKEN` must be lifted into
+  job-level `env`; an unset secret then skips the publish instead of attempting
+  an unauthenticated one.
+- A tag is all-or-nothing: without the `always()` guard a skipped or failed npm
+  half would leave the job silently green. The guard turns it red.
+
+Set `NPM_TOKEN` (npm publish rights) under Settings -> Actions -> Secrets.

development/testing.md

@@ -0,0 +1,214 @@
+# Testing
+
+For this library the types _are_ the feature, so a runtime-only test loop would
+verify the wrong thing. The commands are in
+[CONTRIBUTING.md](../CONTRIBUTING.md); this file records why the loop is shaped
+the way it is.
+
+## Type-driven development
+
+The rules — the loop and the pairing rule — are in
+[CONTRIBUTING.md § Testing discipline (type-driven)](../CONTRIBUTING.md#testing-discipline-type-driven).
+What follows is why and what was rejected.
+
+#### Decision (2026-09)
+
+The compile-time expectation is written before the runtime assertion, and both
+before the implementation.
+
+#### Why
+
+- A runtime-only test can pass while the type is wrong, so a type-level library
+  would ship a broken feature its tests bless.
+- The type error is a more precise spec than a failing assertion, because it
+  states the exact expected type before the logic exists.
+
+#### Rejected
+
+- Runtime-first (classic red/green): it verifies the value, not the contract,
+  and the contract is the product.
+- Testing the type only: it would not catch handler dispatch or the `_`
+  fallback (see `src/primitive-union.test.ts`).
+
+The runner is `node --test --strip-types "src/**/*.test.ts"` and the tiers are
+in
+[CONTRIBUTING.md § Development commands](../CONTRIBUTING.md#development-commands).
+`c8` uses V8 coverage, so the `--strip-types` source is instrumented without a
+build step, and the runner relies on the `.ts` import-extension convention (see
+[tooling.md](./tooling.md#source-imports-use-ts-extensions)). CI gates that
+coverage at 100% (see [ci.md § Coverage threshold](./ci.md#coverage-threshold)).
+
+## Handler arguments
+
+#### Decision (2026-09)
+
+A handler's `expectTypeOf(shape)` is always paired with an assertion on the
+argument `dispatch` actually passed: `assert.equal` where the handler runs for
+one shape, `assert.ok(s === … || s === …)` over the set a `_` fallback accepts
+(`assert` is imported as `strict`, so each comparison is `Object.is`). Where a
+test should also prove that the _exact_ value reached the handler unchanged,
+the fallback returns the shape verbatim and the call site asserts it.
+
+#### Why
+
+- The parameter's type is what the compiler inferred from the pattern; the
+  argument is what the runtime passed. Only the second can drift, and the keys
+  whose property name differs from their value (`true`, `null`, `1`) are
+  exactly where it can — see [library.md](./library.md).
+- `String(shape)` at the call site keeps every type expectation and every
+  return-value assertion green; the argument assertions fail (10 of the 33
+  tests). Without them the suite never looks at the passed argument.
+- Returning the shape verbatim costs a widening pattern nothing: the remainder
+  type joins the union of handler returns in place of a marker literal, so the
+  test still shows the widening it is named for.
+
+#### Rejected
+
+- A recorded `unknown[]` of every fallback call compared with `deepEqual`:
+  strong, but it couples the assertion to call order, and the sink sits three
+  blocks away from the value it observes.
+- `typeof` checks: they cannot separate `2` from its key text `"2"`, which is
+  the drift a fallback with a numeric remainder can hit.
+- One expected value asserted inline in a fallback: its argument is a _set_ of
+  shapes, so only the disjunction holds on every call.
+
+## AAA ordering
+
+The rule is in
+[CONTRIBUTING.md § Testing discipline (type-driven)](../CONTRIBUTING.md#testing-discipline-type-driven).
+
+#### Decision (2026-09)
+
+Test bodies read arrange → act → assert: inputs (the factory) set up first, the
+subject exercised once from them, all checks last — types then runtime. The
+blocks are labeled with `// Arrange` / `// Act` / `// Assert` comments and
+separated by a blank line; an empty block drops its label.
+
+#### Why
+
+- Interleaved setup/checks hide what runs vs. what is observed; the eye
+  re-reads the block to find the seams.
+- A factory built mid-test invites a second throwaway call of the subject;
+  arranging it once makes the positive construction and the negative
+  `Parameters<…>` check share one source of truth.
+- Labels make the seams explicit, not inferred — grep-able and reviewable
+  without reading the statements.
+
+#### Rejected
+
+- Unlabeled ordering (bare blank lines): the seams still have to be found by
+  reading; the labels cost nothing.
+
+## Test helpers
+
+The rule is enforced by `import/no-relative-parent-imports`; this section records
+why the mechanism is shaped like this.
+
+#### Decision (2026-09)
+
+A shared test helper — code that scattered `*.test.ts` files import to do their
+testing — lives under `src/util/__tests__/` and is addressed by the
+`#test-utils/…` self-reference (`package.json#imports`:
+`"#test-utils/*": "./src/util/__tests__/*"`), never by a relative path:
+
+```ts
+import { LspSession } from "#test-utils/lsp-completion.ts";
+```
+
+#### Why
+
+- The lint rule bans upward (`../`) imports, and a cross-cutting helper can
+  always be placed above _some_ scattered consumer, wherever it goes. Name
+  beats path: a `#test-utils/…` specifier has no direction, so the rule never
+  fires and file moves only touch the one mapping in `package.json`.
+- `#…` is Node's reserved prefix for _private_ subpath imports: publishing
+  `package.json` leaks nothing and resolves nothing for consumers.
+- `__tests__` as the folder name is not about tests living there; it is the
+  directory pattern `tsconfig.build.json` already excludes, so a helper can
+  never be emitted into `dist/` and shipped by accident.
+- Node's own resolver handles `#…` under `--strip-types`, and `tsc` resolves it
+  via the same `imports` field — one mechanism for runtime and type gate, no
+  loader needed.
+- The helper uses `node:` builtins (it drives a language server), so
+  `import/no-nodejs-modules` is off for `src/util/__tests__/**` in
+  `.oxlintrc.json` — the same exception the `scripts/**` scope carries.
+- Helpers carry no library coupling: their tests probe in-memory documents
+  whose contextual types are written inline, so only the helper's own contract
+  (marker handling, position, label extraction) is under test. A test that
+  asserts a _matcher's_ popup belongs with the matcher.
+
+#### Rejected
+
+- A helper beside the tests (`src/util/test.ts`, flat `src/`): composes only
+  while `src/` stays flat; the first nested test reaching it reintroduces the
+  banned upward import.
+- Bare `~/…` specifier: not valid in `imports` (keys must start with `#`) —
+  resolution fails at runtime with `ERR_MODULE_NOT_FOUND`. A `#~/…` “home”
+  shorthand was dropped in review; `#test-utils/…` states what it is.
+- tsconfig `paths` alias: resolves for `tsc` but not for plain
+  `node --test --strip-types` (no loader hook), breaking the fast tier.
+- Turning `import/no-relative-parent-imports` off for test files: the rule
+  still guards non-test helpers importing each other, and the exemption is
+  only needed for the one specifier the mapping already solves cleanly.
+
+## Autocomplete
+
+#### Decision (2026-09)
+
+Completion is verified by driving the repo's own language server
+(`tsc --lsp --stdio`, the same server pi's LSP extension talks to) through
+the test helper `#test-utils/lsp-completion.ts`
+(`src/util/__tests__/lsp-completion.ts`), not through the type system:
+
+```sh
+node --strip-types src/util/__tests__/lsp-completion.ts <file> [<marker>]
+```
+
+The script prints the labels the server offers at a `/*COMPLETE*/` marker inside
+`<file>` (the marker is stripped before the document is sent). Its `LspSession`
+is imported by `src/util/__tests__/lsp-completion.test.ts` — which tests the
+helper itself against inline documents, never the library's code — and by
+`src/primitive-union.test.ts`, where the same probe asserts the matcher's popup;
+the CLI is for manual inspection.
+
+#### Why
+
+- Completion is a contextual-type property: it depends on which overload
+  signature TypeScript picks for the object literal, and no type-level assertion
+  observes that.
+- `Parameters<typeof factory>[0]` resolves only the _last_ overload, so it is
+  not the popup's contextual type either — see
+  [library.md § Matcher shape](./library.md#matcher-shape).
+- The server is the only ground truth; the script reproduces what the editor
+  shows.
+
+#### Rejected
+
+- **`expect-type` would not work**: there is no operator for “the popup offers
+  these labels”. `toExtend` / `toEqualTypeOf` test assignability and cannot say
+  which overload supplied the contextual type.
+- **Checking by hand in the editor**: not reproducible in review or by an agent.
+- **`@ts-expect-error` at a completion position**: it asserts the absence of a
+  compile error, not the presence of specific labels.
+
+#### Known issue
+
+- Each test spawns its own `tsc` server so the tests share no state and pass in
+  any order; the file is an integration test (~1.6 s) that needs `node_modules`.
+  `didOpen` is handled in order before the completion request, so no settle
+  delay is needed.
+- The server answers some requests with a string id (`client/registerCapability`);
+  the client must tolerate `string | number` ids or the server stalls.
+- `LspSession.close()` sends `shutdown` and then closes stdin instead of
+  sending `exit`. The TS 7 Go server's `handleExit` returns `io.EOF`, cancelling
+  the background context while a watch update is still in flight, and logs a
+  bare `context canceled` before exiting 1; EOF on stdin exits 0 with no output.
+  The kill stays as a fallback for a server that does not exit.
+
+## Known issues
+
+- The type-aware linter misidentifies `expectTypeOf()` as a floating promise.
+  It is a known false positive, so `typescript/no-floating-promises` is off for
+  `**/*.test.ts` in the `.oxlintrc.json` override rather than repeated as a
+  file-level header (see
+  [tooling.md § oxlint-disable directives live next to the code](./tooling.md#oxlint-disable-directives-live-next-to-the-code)).

development/tooling.md

@@ -0,0 +1,345 @@
+# Tooling
+
+Every tool below was chosen and configured deliberately. The commands a
+contributor runs are in [CONTRIBUTING.md](../CONTRIBUTING.md) and the versions
+in [package.json](../package.json).
+
+## Tool inventory
+
+- **TypeScript 7** — type checker and build (`tsc`).
+- **node --test** + `--strip-types` — test runner.
+- **c8** — coverage for `test:ci`.
+- **oxlint** — Rust linter, type-aware via **oxlint-tsgolint** (typescript-go).
+- **oxfmt** — Rust formatter (Prettier-compatible) for JS/TS, JSON/JSONC, YAML,
+  Markdown, MDX, and more; its `package.json` key sorting replaces
+  `sort-package-json`.
+- **cspell** — spell checking.
+- **knip** — unused dependencies, exports, and files.
+- **check-outdated** — dependencies behind the registry; exits non-zero when any
+  is outdated.
+- **publint** — validates `package.json` for ESM publishing.
+- **@arethetypeswrong/cli** (`attw`) — validates `.d.ts` against module-resolution
+  scenarios.
+- **lefthook** — git hooks.
+- **@spences10/pi-lsp** — read-only LSP code intelligence for AI agents
+  (project-local `.pi/settings.json`); talks to this repo's TypeScript 7 via
+  `tsc --lsp --stdio`.
+- **vscode-languageserver-protocol** — LSP client and protocol types for the
+  autocomplete test helper (`src/util/__tests__/lsp-completion.ts`).
+
+When each runs is in
+[CONTRIBUTING.md § Feedback tiers](../CONTRIBUTING.md#feedback-tiers).
+
+## TypeScript and build
+
+### One type-check config, one emit config
+
+#### Decision (2026-09)
+
+`tsconfig.json` extends `@tsconfig/strictest` + `@tsconfig/node26`.
+`tsconfig.build.json` adds the emit-only options (`declaration`, `sourceMap`,
+`inlineSources`, `outDir`, `target: es2024`,
+`rewriteRelativeImportExtensions: true`) and excludes test files.
+
+#### Why
+
+- The editor and CI type-check from one config while the build emits from the
+  other, so a test file cannot leak into `dist/`.
+- `inlineSources` embeds the original TypeScript in `dist/*.js.map`, so
+  debuggers map into `src/` without it being shipped.
+- `declarationMap` stays off: a `.d.ts.map` cannot embed source and would
+  dangle.
+
+### The build starts from an empty `dist/`
+
+#### Decision (2026-09)
+
+`npm run build` runs a `prebuild` hook that empties `dist/`.
+
+#### Why
+
+- `tsc` does not prune orphaned emit output — dropping `declarationMap` left
+  stale `*.d.ts.map` files — so reproducibility needs an empty `dist/`.
+- `prebuild` removes only `dist`; the manual `clean` resets `dist` + `coverage`,
+  so a local coverage report survives a build.
+
+### Source imports use `.ts` extensions
+
+#### Decision (2026-09)
+
+Source imports use `.ts`, never `.js`.
+
+#### Why
+
+- `node --strip-types` resolves the `.ts` form at test time.
+- `rewriteRelativeImportExtensions` rewrites them to `.js` in the emitted
+  JavaScript.
+- The emitted `.d.ts` keep the `.ts` specifier, which TypeScript >= 5.0 resolves
+  (see [README § Requirements](../README.md#requirements)), so no
+  post-processing step is needed.
+
+#### Rejected
+
+- "Pre-fixing" an import to `.js`: it breaks the inner `node --strip-types`
+  loop.
+
+## Linting and formatting
+
+### Type-aware oxlint is a config property
+
+#### Decision (2026-09)
+
+Type-aware oxlint is enabled via `options.typeAware: true` in `.oxlintrc.json`
+(powered by `oxlint-tsgolint`).
+
+#### Why
+
+- The script commands stay clean — no CLI flag.
+- A config property cannot be forgotten on one call site.
+
+#### Rejected
+
+- A CLI flag in the `check:oxlint` / `fix:oxlint` scripts: it puts the mode in
+  two places and invites them to drift.
+
+### `oxlint-disable` directives live next to the code
+
+#### Decision (2026-09)
+
+A type-aware rule that false-positives **at one site** is silenced with a
+source-level `oxlint-disable` directive (see `src/primitive-union.ts`). A rule that is
+wrong for a whole **file class** is turned off in a `.oxlintrc.json` `overrides`
+entry instead — e.g. `typescript/no-floating-promises` (synchronous
+`expectTypeOf` reads as an unhandled promise) and `unicorn/no-null` (intentional
+`null` inputs) for `**/*.test.ts`. The same exemption is not repeated as a
+file-level header in every affected file.
+
+#### Why
+
+- A one-site disable sits next to the code it silences, visible to anyone
+  reading the source, and the rule stays on everywhere else.
+- A file-class rule is a property of the file class, not of one line; the
+  override states it once, where the rest of the file-class config lives.
+
+#### Rejected
+
+- A project-wide disable in `.oxlintrc.json` for a one-site false positive: it
+  hides the exemption from the reader of the affected code and switches the rule
+  off repo-wide for a one-site problem.
+- A repeated file-level `oxlint-disable` header for a file-class false positive:
+  the copies drift and scatter one config decision across the tree.
+
+#### Known issue
+
+- Both placements are _human_ last resorts. AI agents must neither add a source
+  disable nor edit `.oxlintrc.json`; they fix the type at its root (see
+  [AGENTS.md § Never do](../AGENTS.md#never-do)).
+
+### Unwanted stylistic rules are turned off in the config
+
+#### Decision (2026-09)
+
+A stylistic rule the project rejects is `"off"` in the `.oxlintrc.json` `rules`
+map, not silenced at a use site. Current entries: `eslint/capitalized-comments`
+(comments may start lowercase) and `eslint/no-ternary` (ternaries are allowed),
+joining the oxfmt-superseded rules already off.
+
+#### Why
+
+- The rule is wrong for the whole project, not mis-firing at one site, so there
+  is no line to annotate.
+- Keeping the two mechanisms separate keeps a source-level `oxlint-disable`
+  meaningful: it marks a lone exception.
+
+#### Rejected
+
+- A source-level `oxlint-disable` per use: the same exemption repeated at every
+  site, and oxfmt can move the site.
+
+### `check:tsc` runs first
+
+#### Decision (2026-09)
+
+`check:tsc` runs first in the `npm run check` chain.
+
+#### Why
+
+- A type error short-circuits the rest, which is faster than running
+  oxlint/oxfmt and failing on `tsc` at the end.
+
+### `.editorconfig` is a fallback, not a gate
+
+#### Decision (2026-09)
+
+`.editorconfig` exists for editor compatibility; where both apply,
+`.oxfmtrc.json` is authoritative.
+
+#### Why
+
+- `.editorconfig` covers the files oxfmt does not format: shell scripts,
+  dotfiles, `LICENSE`, the commit-message template, and git's `COMMIT_EDITMSG`
+  buffer.
+- oxfmt is the formatter; the overlapping keys only keep non-oxfmt editors close
+  to the formatted result, so they cannot disagree with the checker.
+
+## Static analysis and packaging
+
+### `knip` omits the `types` category
+
+#### Decision (2026-09)
+
+`knip --include dependencies,exports,files` omits the `types` category.
+
+#### Why
+
+- `types` produces systematic false positives for libraries whose exported types
+  are part of the public API.
+- The narrower scope keeps the signal high without config-file boilerplate.
+
+### `knip` lists `src/index.ts` as an entry
+
+#### Decision (2026-09)
+
+`knip.json` declares `"entry": ["src/index.ts", "scripts/*.ts"]`.
+
+#### Why
+
+- Supplying `entry` **replaces** knip's default entry detection, which otherwise
+  derives the public entry from `package.json` `exports`. Adding `scripts/*.ts`
+  there therefore dropped the library entry, so knip resolved the package through
+  its `dist/index.js` output and reported the unreferenced source entry file
+  `src/index.ts` as an unused file.
+- Naming the source entry restores the link between the public API and the
+  source graph without pointing knip at build output.
+
+#### Rejected
+
+- `paths` mapping `dist/index.*` back to `src/index.ts`: more config to model a
+  relation the explicit entry states directly, and it would break whenever the
+  build layout changes.
+
+### `maintain:outdated` ignores `@types/node`
+
+#### Decision (2026-09)
+
+Pass `--ignore-packages @types/node`.
+
+#### Why
+
+- DT pins `@types/node`'s `latest` dist-tag to LTS (22.x); current-line types
+  ride other tags. Scan sees latest < installed — permanent "reverted", exit
+  1, zero signal. `--ignore-pre-releases` no help: 22.20.3 is stable.
+
+#### Rejected
+
+- `--types major,minor,patch`: hides real reverted reports elsewhere.
+- `@types/node@26.*`: tag stays wrong across majors; un-pin per bump = ritual.
+
+#### Known issue
+
+- A genuinely behind `@types/node` goes unreported; match it to
+  `.node-version` by hand.
+
+### `attw` targets ESM-only
+
+#### Decision (2026-09)
+
+`attw --profile esm-only` is used.
+
+#### Why
+
+- The package is intentionally ESM-only (no CommonJS shim), so CJS resolution
+  scenarios are out of scope by design, not a bug.
+
+### `tslib` is deliberately not used
+
+#### Decision (2026-09)
+
+`tslib` is not a dependency.
+
+#### Why
+
+- `tslib` is a runtime helper for old ES3/ES5 targets; this project targets
+  ES2024.
+
+## Git hooks and script wiring
+
+### `LEFTHOOK_FILES` scopes commands to staged files
+
+#### Decision (2026-09)
+
+The pre-commit hook sets `LEFTHOOK_FILES` to the staged-files list, and the
+affected scripts use `${LEFTHOOK_FILES:-<default>}` to default to the whole
+project.
+
+#### Why
+
+- It keeps `package.json#scripts` the single source of truth; `lefthook.yml`
+  only says what to run on which files.
+- The same script works by hand (whole project) and staged (scoped), so there is
+  no second command to maintain.
+
+## Language server tooling
+
+### `vscode-languageserver-protocol` backs the autocomplete helper
+
+#### Decision (2026-09)
+
+The autocomplete helper (`src/util/__tests__/lsp-completion.ts`) drives
+`tsc --lsp --stdio` through `vscode-languageserver-protocol`'s
+`createMessageConnection` and its typed request / notification objects, instead
+of a hand-rolled JSON-RPC client.
+
+#### Why
+
+- Framing, `Content-Length` parsing, the pending-request map and server-request
+  dispatch are protocol plumbing the helper only reimplemented; the official
+  client owns them and tolerates the server's `string | number` ids.
+- `InitializeRequest`, `CompletionRequest`, `DidOpenTextDocumentNotification`,
+  … carry their parameter and result types, so `CompletionList` / `CompletionItem`
+  replace the helper's ad-hoc shape guards.
+- The `./node` entry re-exports `vscode-jsonrpc/node`, so one devDependency
+  supplies both the transport and the protocol types. It is test-only and never
+  ships (`files` publishes `dist/` only).
+
+#### Rejected
+
+- `vscode-languageclient`: the editor-side client with a full feature registry
+  — far more than a test helper needs.
+- Generic JSON-RPC (`jsonrpc-lite`, `jayson`): still no LSP types, so they
+  replace framing only and leave the typed protocol surface unimplemented.
+- Keeping the hand-rolled client: the low-level shape is the maintenance cost
+  the helper exists to remove, and it must be re-audited against the server.
+
+## Editor and agent tooling
+
+### VSCode integration
+
+- Recommended extensions are in
+  [.vscode/extensions.json](../.vscode/extensions.json) (oxc, cspell, TypeScript
+  native-preview, EditorConfig, todo-tasks).
+- TypeScript 7 runs via the `typescriptteam.native-preview` extension.
+- The oxc extension provides oxlint squiggles and oxfmt format-on-save;
+  `.vscode/settings.json` pins it per language so a local `[language]` formatter
+  setting cannot override the project's choice.
+
+### `@spences10/pi-lsp` is pinned and read-only
+
+#### Decision (2026-09)
+
+`@spences10/pi-lsp` is pinned to `0.0.46` and used read-only.
+
+#### Why
+
+- It inspects `node_modules/typescript`, sees major >= 7 with no
+  `lib/tsserver.js` (true of the `typescript-go` / `tsgo` port), and spawns the
+  repo's own `tsc --lsp --stdio` — no `typescript-language-server` dependency is
+  needed.
+- Earlier releases (`<= 0.0.10`) hard-wire to `typescript-language-server
+--stdio` and are TS6-only.
+- It is _intermediate_ agent feedback (hover, references, definition, symbols,
+  diagnostics), with no rename / code-action / apply-edit surface, and is never a
+  gate — `npm run check` / `verify` are.
+- `.pi/settings.json` is the committed declaration; `.pi/npm/` is a gitignored
+  install cache that pi recreates on a trusted startup (running `npm install`
+  for any missing project package), so it is deliberately not tracked.

development/workflow.md

@@ -0,0 +1,173 @@
+# Workflow
+
+How work moves through the repository. The rules are in
+[CONTRIBUTING.md](../CONTRIBUTING.md); this file records why they are shaped the
+way they are.
+
+## Branching model
+
+The model is GitHub Flow (single-developer); the steps are in
+[CONTRIBUTING.md § Branching model](../CONTRIBUTING.md#branching-model). Context
+behind it: Gitea has no collaborative review UI in use, so it is the lab, and
+the project moves to GitHub once it is tested and ready.
+
+#### Decision (2026-09)
+
+Work is opened and closed by `create:branch` / `create:finish`, not by prose
+plus hand-written `git`.
+
+#### Why
+
+- The preconditions were prose, and prose rots: a rule nobody checks is a
+  suggestion. A script asserts, then acts, so the branch or merge only exists if
+  the assertions passed.
+- Type-driven work is only trustworthy if the baseline was green before the
+  first edit. Cheap checks run first and `npm run test` last, so the expensive
+  gate is not paid on an ineligible tree.
+- The merge half owns the post-merge `npm run verify`, so a merge cannot land
+  unverified. The push stays with `create:release` so the merge is reviewed
+  locally first; `main` is therefore routinely ahead of its upstream between a
+  merge and the release that ships it. `create:branch` requires only that `main`
+  is not _behind_ — matching `create:finish`, which tolerates the local merge and
+  fast-forwards over a remote one — rather than an exact match.
+- Every failure is non-mutating except the baseline test, which runs on `main`
+  after switching there: a red `main` restores the branch you started on, and a
+  merge conflict aborts back to the feature branch rather than stranding a
+  half-merged `main`.
+
+#### Rejected
+
+- Hand-written `git switch -c` / `git merge`: same rules, no enforcement.
+- Reusing `pubv`'s preflight for `create:branch`: release-shaped, third-party,
+  and it would pay for a build and pack a new branch has no use for.
+- Leaving the merge to reviewer judgment: that judgment moved earlier, to the
+  handover review before `create:finish`, rather than living in a command anyone
+  can run from a dirty tree.
+- Fast-forward instead of `--no-ff`: `--no-ff` keeps each unit of work visible
+  in `git log`.
+- Pushing from `create:finish` to keep `main` level with its upstream: it would
+  trade the local review the push waits for for a network side effect, and a
+  failed push would leave the merge landed but unpublished.
+
+## Changelog notes
+
+The rule is in
+[CONTRIBUTING.md § Rules the tools don't enforce](../CONTRIBUTING.md#rules-the-tools-dont-enforce).
+
+#### Decision (2026-09)
+
+A merged branch carries its own summary under `[Unreleased]` in
+[CHANGELOG.md](../CHANGELOG.md), added before `create:finish`;
+`create:release` graduates it into the tagged section (see
+[publishing.md](./publishing.md)).
+
+#### Why
+
+- `create:release` derives the bump heuristic from the `[Unreleased]` body, so
+  the notes must exist before release day.
+- The contributor has fresh context; at release day the intent of a branch is
+  only its diff.
+- Gitmoji subjects are signposts, not semantic keys, so notes cannot be derived
+  from the history.
+
+#### Rejected
+
+- Generating notes from subjects at release time: subjects carry no parseable
+  type/scope (see § Commit messages).
+- The maintainer writing one summary during `create:release`: reconstruction
+  after the fact.
+- Enforcing it in `create:finish`: the front doors assert git state, not
+  content — and _notable_ is exactly the judgment a tool cannot make.
+
+## Script prefix convention
+
+The prefix taxonomy is the rule, and it lives in
+[CONTRIBUTING.md § Script prefix convention](../CONTRIBUTING.md#script-prefix-convention).
+The design behind it: bare scripts are the tier entry points — a single tool
+(`build`, `clean`) or an aggregator of a `prefix:*` family (`check`, `fix`,
+`test`, `watch`, `maintain`, `setup`) — while `verify` composes `check` +
+`test:unit` into the whole-project gate (it uses `test:unit`, not `test`,
+because `check` already runs `check:tsc`).
+
+#### Decision (2026-09)
+
+`create:` is the prefix for workflow front doors, with no bare `create`
+aggregator.
+
+#### Why
+
+- Both members create something real: a branch, a release.
+- It joined both lists in [CONTRIBUTING.md](../CONTRIBUTING.md) alongside its
+  first members, so it could not go invisible the way the retired `use:` prefix
+  did.
+- `publish:*` already set the precedent for a prefix without an aggregator.
+
+#### Rejected
+
+- `run:` / `perform:`: they mean only "do the named thing", so every script fits
+  and the taxonomy collapses.
+- `git:`: names the tool, not the lifecycle moment, and implies passthrough
+  aliases.
+- `start:`: describes the branch half, not the release.
+- `cut:`: idiomatic but needs VCS slang to decode.
+- `flow:`: overloaded in a type-level matching library.
+- The existing families: `check:*` is read-only (CI would run a state-mutating
+  command), `fix:*` reviews as a diff not a branch, `maintain:*` is advisory and
+  never a gate.
+
+## Feedback tiers
+
+The table and invocation rules are in
+[CONTRIBUTING.md § Feedback tiers](../CONTRIBUTING.md#feedback-tiers); this
+section explains the split.
+
+#### Decision (2026-09)
+
+Fast, offline, staged-file checks sit in pre-commit; whole-project test runs in
+pre-push and `verify`; slow or network-bound scans under `maintain`.
+
+#### Why
+
+- `watch:*` runs until killed, in its own pane, so it is the earliest tier,
+  firing on save before staging or commit.
+- `check:tsc` / `check:oxlint` / `check:oxfmt` / `check:cspell` are fast
+  (~0.2–0.5s each), offline, and scope to staged files via `LEFTHOOK_FILES`, so
+  pre-commit gives instant feedback on what you typed.
+- `test` (and its `tsc`) runs the whole suite over the whole project, and the
+  staged-file convention does not apply to the test runner, so it belongs in
+  pre-push, after the commits exist but before the push leaves the machine.
+
+#### Rejected
+
+- `maintain:*` in `check` or pre-commit: advisory, whole-project and
+  network-bound scans are not correctness gates and would slow the fast tier.
+- Treating a green pre-commit as the definition of done: it sees only staged
+  files, hence `npm run verify`.
+- A separate `git push` hook for `verify`: the pre-push test tier already covers
+  it.
+
+## Commit messages
+
+The convention is in
+[CONTRIBUTING.md § Commit messages](../CONTRIBUTING.md#commit-messages).
+Examples: `:sparkles: Add watch tier with watch:test child`,
+`:recycle: Move type-aware config to .oxlintrc.json; use source-level disable
+directives`, `:memo: Restore unique maintainer content as CONTRIBUTING.md`. The
+body explains what and why, not how; link issues with `Resolves #...`.
+
+#### Decision (2026-09)
+
+Gitmoji subjects, imperative mood, wrapped 50/72, not Conventional Commits.
+
+#### Why
+
+- The history is gitmoji and predates any commit-lint tooling; switching would
+  rewrite the convention for no gain.
+- The body carries the reasoning a reviewer needs; the subject is a signpost,
+  not a semantic key.
+
+#### Rejected
+
+- Conventional Commits: the release flow uses hand-written Keep-a-Changelog
+  notes, not generated ones, so the prefix has no automation value here (see
+  [publishing.md](./publishing.md)).

knip.json

@@ -1,5 +1,5 @@
 {
     "$schema": "./node_modules/knip/schema.json",
-    "entry": ["scripts/*.ts"],
+    "entry": ["src/index.ts", "scripts/*.ts"],
     "ignoreDependencies": ["@runwisp/pubv"]
 }

package.json

@@ -1,6 +1,6 @@
 {
     "name": "tiny-pattern-ts",
-    "version": "0.1.5",
+    "version": "0.8.1",
     "description": "Pattern matching for TypeScript/ESM environments (F#-style, not regex)",
     "keywords": [
         "adt",
@@ -27,6 +27,9 @@
     ],
     "type": "module",
     "sideEffects": false,
+    "imports": {
+        "#test-utils/*": "./src/util/__tests__/*"
+    },
     "exports": {
         ".": {
             "types": "./dist/index.d.ts",
@@ -53,9 +56,9 @@
         "create:release": "./scripts/release.sh",
         "maintain": "npm run maintain:knip; npm run maintain:outdated",
         "maintain:knip": "knip --include dependencies,exports,files",
-        "maintain:outdated": "check-outdated --ignore-pre-releases",
+        "maintain:outdated": "check-outdated --ignore-pre-releases --ignore-packages @types/node",
         "test": "npm run check:tsc && node --test --strip-types \"src/**/*.test.ts\"",
-        "test:ci": "c8 --reporter=text --reporter=lcov --reporter=html node --test --strip-types \"src/**/*.test.ts\"",
+        "test:ci": "c8 --all --include \"src/**/*.ts\" --reporter=text --reporter=lcov --reporter=html --100 node --test --strip-types \"src/**/*.test.ts\"",
         "test:unit": "node --test --strip-types \"src/**/*.test.ts\"",
         "verify": "npm run check && npm run test:unit",
         "watch": "npm run watch:test",
@@ -65,23 +68,27 @@
         "setup": "npm run setup:git-commit-message",
         "setup:git-commit-message": "git config commit.template commit-message-template"
     },
+    "dependencies": {
+        "type-fest": "^5.9.0"
+    },
     "devDependencies": {
         "@arethetypeswrong/cli": "^0.18.5",
         "@runwisp/pubv": "^1.5.1",
         "@tsconfig/node26": "^26.0.1",
         "@tsconfig/strictest": "^2.0.8",
-        "@types/node": "^26.4.1",
+        "@types/node": "^26.6.1",
         "c8": "^12.0.0",
         "check-outdated": "^3.0.0",
-        "cspell": "^10.2.2",
+        "cspell": "^10.3.2",
         "expect-type": "1.4.0",
         "knip": "^6.34.0",
         "lefthook": "^2.1.12",
-        "oxfmt": "^0.68.0",
+        "oxfmt": "^0.70.0",
         "oxlint": "^1.83.0",
         "oxlint-tsgolint": "^7.0.2001",
         "publint": "^0.3.24",
-        "typescript": "^7.0.2"
+        "typescript": "^7.0.2",
+        "vscode-languageserver-protocol": "^3.18.3"
     },
     "engines": {
         "node": ">=26"

scripts/branch.sh

@@ -4,43 +4,13 @@ set -eu
 
 # Branch front-door. Run as `npm run create:branch -- <prefix>/<desc>`.
 #
-# How we got here (short): the branching model says every change starts from a
+# Asserts the branch preconditions — clean tree, no in-progress operation,
-# clean, current `main`, and the type-driven loop only produces trustworthy
+# current `main`, green baseline — and only then creates the branch, so the
-# results if the baseline was green *before* the first edit. Both facts were
+# expensive `npm run test` is not paid on a tree that was never eligible.
-# prose. Prose rots silently — a rule nobody checks is a suggestion — so the
-# precondition became this script: it asserts, then branches, and the branch
-# only appears if the assertions passed. Cheap checks run first, `npm run test`
-# runs last: the expensive gate is not paid on a tree that was never eligible.
 #
-# Rejected for the prefix name: `run:` / `perform:` (both mean only "do the
+# Why the front door exists, the rejected prefix names, and the `create:`
-# thing named after them", so every script in the repo would fit under them and
+# decision: development/workflow.md § Branching model and § Script prefix
-# the taxonomy collapses); `git:` (names the tool, not the lifecycle moment, and
+# convention.
-# advertises passthrough aliases); `start:` (describes this half, not the
-# release); `cut:` (idiomatic for both, but it needs VCS slang to decode, and a
-# signpost that has to be explained is not one); `flow:` (overloaded in a library
-# about type-level matching); and the existing families — `check:*` is read-only
-# and aggregated by `check`, so CI would run a command that mutates repo state;
-# `fix:*`'s review surface is a file diff, not a branch; `maintain:*` is advisory
-# and explicitly never a gate.
-#
-# `create:` was kept because both members really do create something: a branch,
-# a release. It was added to both prefix lists in CONTRIBUTING.md in the same
-# commit as its first members, because a prefix missing from those lists is
-# invisible — which was the `use:` mistake this repo carried in backlog.tasks (since retired into `setup:`).
-# There is deliberately no bare `create` aggregator: "run all the workflows"
-# describes nothing anyone wants, and `publish:*` already sets the precedent for
-# a prefix without one.
-#
-# Also rejected here: reusing `pubv`'s preflight (release-shaped, third-party,
-# and it would make branch start pay a build + pack it has no use for). The
-# merge half was originally rejected too ("review the diff yourself" is
-# judgment), but it now has its own front door — `create:finish` — which owns
-# the merge-side preconditions and the post-merge `verify`, so the start half
-# does not have to carry that burden.
-#
-# Every refusal is non-mutating except the baseline test, which runs on `main`
-# after we switch there — so a red `main` restores the branch you started on
-# rather than stranding you on it.
 
 BASE="main"
 PREFIXES="feature fix chore"
@@ -103,9 +73,9 @@ git show-ref --verify --quiet "refs/heads/${BASE}" || {
     exit 1
 }
 
-# Derive the remote rather than hardcoding it: this repo has `origin` (ssh) and
+# Derive the remote rather than hardcoding it: `main` tracks `origin` (ssh)
-# `origin_https`, and `main` tracks the latter — `git fetch origin main` would
+# here; a hardcoded name would check currency against a ref that may not
-# check currency against a ref that is never updated here.
+# exist on a differently configured clone.
 # `--quiet` echoes the unresolved `main@{upstream}` literal to stdout on
 # failure, so it cannot be paired with a `$(...) || fallback`; the non-quiet
 # form prints nothing on failure and the fallback runs.
@@ -116,12 +86,14 @@ if [ -n "${UPSTREAM}" ]; then
         echo "       refusing to branch on a possibly stale '${BASE}'." >&2
         exit 1
     }
+    # Only *behind* is a problem. `create:finish` deliberately leaves the local
+    # merge on '${BASE}' until `create:release` pushes it, so being ahead is the
+    # normal state between a merge and the release that ships it; branching from
+    # those commits is intended. Missing remote commits is not.
     BEHIND=$(git rev-list --count "${BASE}..${UPSTREAM}")
-    AHEAD=$(git rev-list --count "${UPSTREAM}..${BASE}")
+    if [ "${BEHIND}" -ne 0 ]; then
-    if [ "${BEHIND}" -ne 0 ] || [ "${AHEAD}" -ne 0 ]; then
+        echo "error: '${BASE}' is behind '${UPSTREAM}' (by ${BEHIND})." >&2
-        echo "error: '${BASE}' has diverged from '${UPSTREAM}' (ahead ${AHEAD}, behind ${BEHIND})." >&2
+        echo "       update it: git switch ${BASE} && git pull --ff-only" >&2
-        [ "${AHEAD}" -ne 0 ] && echo "       not yet pushed commits on '${BASE}': push them, or rebase this work onto them." >&2
-        [ "${BEHIND}" -ne 0 ] && echo "       update it: git switch ${BASE} && git pull --ff-only" >&2
         exit 1
     fi
 else

scripts/finish.sh

@@ -4,29 +4,14 @@ set -eu
 
 # Feature-finish front door. Run as `npm run create:finish`.
 #
-# Why this exists: `create:branch` opens a unit of work, but the close half
+# Mirror image of `create:branch`: asserts the merge-side preconditions, merges
-# (`git checkout main && git merge --no-ff <branch>`) stayed prose in the
+# the current `feature/`/`fix/`/`chore/` branch into `main` with `--no-ff`,
-# branching model, so it drifted per contributor and per session. This is the
+# proves the result with `npm run verify`, and only then deletes the branch. The
-# mirror image of `create:branch`: it asserts the same preconditions (clean
+# push is owned by `create:release`, so the merge stays local and reviewable.
-# tree, no in-progress operation, `main` matching its upstream), merges the
+# On a conflict it aborts and returns to the feature branch.
-# current `feature/`/`fix/`/`chore/` branch into `main`, proves the result with
-# `npm run verify`, and only then deletes the branch.
 #
-# `create:branch`'s comment argued against wrapping the merge as "judgment —
+# Rationale and the rejected alternatives: development/workflow.md § Branching
-# review the diff yourself". That judgment still lives here, just moved: the
+# model.
-# maintainer reviews the handover *before* invoking this, and the script only
-# commits the merge, never the push. The push is owned by `create:release`, so
-# the release commit and its tag leave together and a local merge stays
-# reviewable (and can be reverted with `git revert -m 1`) until then. `--no-ff`
-# keeps the unit of work visible in `git log`.
-#
-# Unlike `create:branch` a stale `main` is fast-forwarded instead of refused:
-# the tree is clean (checked above) and `main` is not the checked-out branch
-# yet, so there is no local state to lose. True divergence (local commits *and*
-# upstream commits) is still refused — that needs a human.
-#
-# On a merge conflict we abort and return to the feature branch, so a failed
-# finish never strands you on a half-merged `main`.
 
 BASE="main"
 PREFIXES="feature fix chore"

scripts/precompress.ts

@@ -9,6 +9,9 @@ import zlib from "node:zlib";
  * matching `Accept-Encoding` and falls back to the original for the rest.
  *
  * Usage: node --strip-types scripts/precompress.ts <dir> [<dir>...]
+ *
+ * Why sidecars rather than per-request compression: development/ci.md § Coverage
+ * serving.
  */
 
 /**

scripts/release-notes.sh

@@ -9,6 +9,8 @@ set -eu
 # `v1.2.3` and `1.2.3` match the `## [1.2.3]` heading. Prints to stdout and
 # exits non-zero when the tag has no section, so a release can never publish
 # with an empty body.
+#
+# Why: development/publishing.md § Release notes are extracted from the changelog.
 
 TAG="${1:-}"
 CHANGELOG="${CHANGELOG:-CHANGELOG.md}"

scripts/release.sh

@@ -4,29 +4,13 @@ set -eu
 
 # Release front-door. Run as `npm run create:release`.
 #
-# How we got here (short): we want hand-written Keep-a-Changelog notes, an
+# Graduates the [Unreleased] changelog notes, bumps package.json + the lockfile,
-# [Unreleased] -> "## [x.y.z] - DATE" graduation, and a tag that marks the exact
+# and commits then tags the exact SHA that CI publishes. The notes are finalized
-# commit on main that gets published. No single tool did BOTH the [Unreleased]
+# in VS Code before pubv because pubv's bump heuristic reads the [Unreleased]
-# graduation AND the package.json bump. So split by strength: `pubv` (tiny,
+# body.
-# changelog-driven) owns preflight + the interactive major/minor/patch heuristic
-# + graduating/committing CHANGELOG.md (no tag, no push); `npm version` syncs
-# package.json + the lockfile; `--amend` folds them into pubv's single commit;
-# tag AFTER the amend (so the tag is never orphaned) and push.
 #
-# The notes are finalized in VS Code *before* pubv: the [Unreleased] body is
+# Tooling rationale, the rejected alternatives, and the version source of truth:
-# what pubv's bump heuristic reads, so editing afterwards would inform the
+# development/publishing.md.
-# changelog only, not the version choice. pubv refuses a dirty tree, so that
-# edit is committed as a staging commit and folded back into the single release
-# commit below.
-#
-# Rejected: the conventional-commits family (our history is gitmoji, not
-# Conventional; and we want hand-written notes); changesets/rtk (config + a
-# heavier version/publish flow that fights our CI-only publish); knope/kacl/
-# bestikk (changelog-only — don't bump package.json; plus 5yr/2yr/brand-new
-# maintenance); pubv alone (verified it never writes package.json). We also
-# tried `versions` (silverwind) — great Gitea support — but pairing it with a
-# hand-rolled promote became a ~180-line script we'd have to maintain, which is
-# exactly what this ~30-line version replaces.
 
 CHANGELOG="CHANGELOG.md"
 BASE="main"

scripts/runner-image.sh

@@ -8,6 +8,9 @@ set -euo pipefail
 # pulls the image by name.
 #
 # Usage: scripts/runner-image.sh [--push]
+#
+# Why the image is baked, its two invariants, and the coordinated Node-bump
+# steps: development/ci.md.
 
 IMAGE_REPO="gitea.e1nsnull.de/tmu/act-ci"
 

src/index.test.ts

@@ -1,56 +1,27 @@
-/* oxlint-disable typescript/no-floating-promises -- expectTypeOf() is a sync type-assertion library that the type-aware linter misidentifies as a promise */
 import { strict as assert } from "node:assert";
 import { test } from "node:test";
 
 import { expectTypeOf } from "expect-type";
 
-import { type Matcher, P, match } from "./index.ts";
+import {
+    getPrimitiveUnionMatcher,
+    getPrimitiveUnionMatcherW,
+    getTaggedUnionMatcher,
+    getTaggedUnionMatcherW,
+} from "./index.ts";
 
-test("match returns a builder", () => {
+// The published entry point is the barrel (`package.json` exports
-    const builder = match("x");
+// `./dist/index.js`), so every factory must be reachable from here. Importing it
-    expectTypeOf(builder).toHaveProperty("with");
+// also loads the module, which is what lets c8's `--all` measure it — see
-    expectTypeOf(builder).toHaveProperty("exhaustive");
+// development/ci.md § Coverage threshold.
-    expectTypeOf(builder).toHaveProperty("otherwise");
+test("index: the public entry point re-exports every matcher factory", () => {
-});
+    // Assert
-
+    expectTypeOf(getPrimitiveUnionMatcher).toBeFunction();
-test("P.literal narrows to its literal type", () => {
+    assert.equal(typeof getPrimitiveUnionMatcher, "function");
-    const matcher = P.literal("yes");
+    expectTypeOf(getPrimitiveUnionMatcherW).toBeFunction();
-    expectTypeOf(matcher).toMatchTypeOf<Matcher<"yes">>();
+    assert.equal(typeof getPrimitiveUnionMatcherW, "function");
-    assert.equal(matcher.matches("yes"), true);
+    expectTypeOf(getTaggedUnionMatcher).toBeFunction();
-    assert.equal(matcher.matches("no"), false);
+    assert.equal(typeof getTaggedUnionMatcher, "function");
-});
+    expectTypeOf(getTaggedUnionMatcherW).toBeFunction();
-
+    assert.equal(typeof getTaggedUnionMatcherW, "function");
-test("P.type narrows to the typeof target", () => {
-    const matcher = P.type<string>("string");
-    expectTypeOf(matcher).toMatchTypeOf<Matcher<string>>();
-    assert.equal(matcher.matches("hi"), true);
-    assert.equal(matcher.matches(42), false);
-});
-
-test("exhaustive() returns the union of handler return types", () => {
-    const result = match<"a" | "b">("a")
-        .with(P.literal("a"), () => 1 as const)
-        .with(P.literal("b"), () => "two" as const)
-        .exhaustive();
-
-    expectTypeOf(result).toEqualTypeOf<1 | "two">();
-    assert.equal(result, 1);
-});
-
-test("otherwise() falls back when no case matches", () => {
-    const result = match<"x" | "y" | "z">("z")
-        .with(P.literal("x"), (v): string => `got ${v}`)
-        .otherwise((v): string => `fallback ${v}`);
-    assert.equal(result, "fallback z");
-});
-
-test("exhaustive throws when no case matches", () => {
-    assert.throws(
-        () =>
-            match<"a" | "b" | "c">("c")
-                .with(P.literal("a"), () => "A")
-                .with(P.literal("b"), () => "B")
-                .exhaustive(),
-        /no matching case/,
-    );
 });

src/index.ts

@@ -1,2 +1,8 @@
-export { match, P } from "./match.ts";
+export {
-export type { Matcher, Pattern } from "./pattern.ts";
+    getPrimitiveUnionMatcher,
+    getPrimitiveUnionMatcherW,
+} from "./primitive-union.ts";
+export {
+    getTaggedUnionMatcher,
+    getTaggedUnionMatcherW,
+} from "./tagged-union.ts";

src/match.ts

@@ -1,62 +0,0 @@
-import { P, type Matcher, type Pattern } from "./pattern.ts";
-
-type Cases<R> = readonly (readonly [Matcher<unknown>, (value: unknown) => R])[];
-
-interface MatchBuilder<T, R> {
-    with<U extends T, V>(
-        pattern: Matcher<U>,
-        handler: (value: U) => V,
-    ): MatchBuilder<T, R | V>;
-    exhaustive(): R;
-    otherwise(handler: (value: T) => R): R;
-}
-
-const buildMatch = <T, R>(value: T, cases: Cases<R>): MatchBuilder<T, R> => {
-    const apply = (): R | undefined => {
-        for (const [matcher, handler] of cases) {
-            if (matcher.matches(value)) {
-                return handler(value);
-            }
-        }
-        return undefined;
-    };
-
-    const builder = {
-        with<U extends T, V>(
-            pattern: Matcher<U>,
-            handler: (value: U) => V,
-        ): MatchBuilder<T, R | V> {
-            const nextCases: Cases<R | V> = [
-                ...cases,
-                // oxlint-disable-next-line typescript/no-unsafe-type-assertion
-                [pattern, handler as (value: unknown) => R | V],
-            ];
-            return buildMatch(value, nextCases);
-        },
-        exhaustive(): R {
-            const result = apply();
-            if (result === undefined) {
-                throw new Error(
-                    "tiny-pattern-ts: match.exhaustive() called with no matching case",
-                );
-            }
-            return result;
-        },
-        otherwise(handler: (value: T) => R): R {
-            for (const [matcher, run] of cases) {
-                if (matcher.matches(value)) {
-                    return run(value);
-                }
-            }
-            return handler(value);
-        },
-    };
-
-    return builder;
-};
-
-export const match = <T>(value: T): MatchBuilder<T, never> =>
-    buildMatch<T, never>(value, []);
-
-export type { Matcher, Pattern };
-export { P };

src/matcher-shared.ts

@@ -0,0 +1,102 @@
+/* c8 ignore start -- types only: the module has no runtime image to cover */
+import type { IsLiteral, IsNever, ValueOf } from "type-fest";
+
+// The primitive-union and tagged-union matchers differ in their universe, but the
+// handler/fallback plumbing is identical; these are the shared pieces. The
+// boundary is deliberate: `Handlers`, `Fallback` and `MustBePartial` stay with
+// each matcher because they are built from its universe. See development/library.md.
+
+// A handler: one universe member in, one return value out.
+export type UnaryFn<T, R> = (shape: T) => R;
+
+// The primitive universe a matcher can discriminate. The primitive-union matcher
+// uses it directly; the tagged-union matcher uses it as the set of allowed
+// discriminant (`Tag`) values. `boolean` is admitted as the pair `true | false`;
+// see README § Caveats for the unsupported members.
+export type Matchable = string | number | boolean | null | undefined;
+
+// `boolean`, `null` and `undefined` cannot be property keys, so a mapped type
+// over a universe that includes one keys each such member by its
+// stringification. `Member` inverts that projection against the universe, so a
+// handler callback still receives the *real* member (`true`, not `"true"`).
+// Both matchers use the projection: the primitive-union matcher over its
+// universe, the tagged-union matcher over a discriminant property's values. See
+// README § Caveats for the limits.
+export type PatternKey<T> = T extends boolean
+    ? T extends true
+        ? "true"
+        : "false"
+    : T extends null
+      ? "null"
+      : T extends undefined
+        ? "undefined"
+        : T;
+
+// The member(s) of `T` whose `PatternKey` is `K`: the universe-keyed inverse of
+// `PatternKey`. The key alone cannot recover the member (`"true"` and `true`
+// share it), so the handler parameter is derived from `T` instead. For a
+// supported (injective) universe the result is a single member.
+export type Member<
+    T extends Matchable,
+    K extends PropertyKey,
+> = T extends Matchable ? (PatternKey<T> extends K ? T : never) : never;
+
+// The property key a `Matchable` member takes at runtime: booleans, `null` and
+// `undefined` stringify, and a numeric literal becomes its decimal string.
+export type Stringified<T> = T extends boolean
+    ? T extends true
+        ? "true"
+        : "false"
+    : T extends null
+      ? "null"
+      : T extends undefined
+        ? "undefined"
+        : T extends number
+          ? `${T}`
+          : never;
+
+// The members of `T` that are also the stringification of another member, so
+// `PatternKey` cannot invert them. `never` means the universe is injective.
+export type Collisions<T> = Extract<T, Stringified<T>>;
+
+// Why `T` is not a supported universe, or `never` when it is. A supported
+// universe is a finite union of literals with no value/stringification
+// collision: only then can `PatternKey` be inverted unambiguously.
+export type UnsupportedReason<T extends Matchable> =
+    IsLiteral<PatternKey<T>> extends true
+        ? [Collisions<T>] extends [never]
+            ? never
+            : `a value and its stringification collide. Value is "${Collisions<T> & string}"`
+        : "broad types like string, number and template literals are not supported";
+
+// The diagnostic for an unsupported universe. Extends `HandlerMap` so the
+// implementation's `handlers: HandlerMap` stays assignable when the gate is
+// intersected into a parameter; the property name is the message.
+export type UnsupportedUniverse<Reason extends string> = HandlerMap &
+    Readonly<Record<`unsupported universe: ${Reason}`, never>>;
+
+// `unknown` for a supported universe (an intersection no-op), the diagnostic
+// otherwise. Intersecting rather than branching keeps `R` inference intact.
+export type UniverseGate<T extends Matchable> =
+    IsNever<UnsupportedReason<T>> extends true
+        ? unknown
+        : UnsupportedUniverse<UnsupportedReason<T>>;
+
+// `Extract` drops optional handlers (`undefined`) so `PatternReturns` also works
+// when `P`'s constraint has optional keys.
+export type PatternReturns<P> = ReturnType<
+    Extract<ValueOf<P>, (...args: never[]) => unknown>
+>;
+
+// The diagnostic raised when a fallback is supplied for an already-exhaustive
+// handler map. Each matcher's `MustBePartial` folds it into `Handled`'s
+// constraint so the guard is checked after inference.
+export interface RedundantFallback {
+    readonly "every case is already handled, so the fallback is redundant": never;
+}
+
+// The runtime dispatch map the handler maps and fallback erase to.
+export type HandlerMap = Record<
+    string | number,
+    UnaryFn<never, unknown> | undefined
+>;

src/pattern.ts

@@ -1,105 +0,0 @@
-/**
- * Pattern matching primitives. Each constructor returns a lightweight
- * matcher object whose `matches` method returns a type guard.
- */
-
-export interface Matcher<T> {
-    readonly matches: (value: unknown) => value is T;
-}
-
-const literalMatcher = <
-    const L extends string | number | boolean | null | undefined,
->(
-    value: L,
-): Matcher<L> => ({
-    matches: (candidate): candidate is L => candidate === value,
-});
-const typeMatcher = <T>(
-        type:
-            | "string"
-            | "number"
-            | "boolean"
-            | "bigint"
-            | "symbol"
-            | "undefined"
-            | "object"
-            | "function",
-    ): Matcher<T> => {
-        const matches = (value: unknown): value is T => {
-            if (type === "undefined") {
-                return value === undefined;
-            }
-            if (type === "object") {
-                return (
-                    (typeof value === "object" && value !== null) ||
-                    typeof value === "function"
-                );
-            }
-            return typeof value === type;
-        };
-        return { matches };
-    },
-    whenMatcher = <T>(
-        predicate: (value: unknown) => value is T,
-    ): Matcher<T> => ({
-        matches: predicate,
-    }),
-    whenMatcherAny = <T>(
-        predicate: (value: unknown) => boolean,
-    ): Matcher<T> => ({
-        matches: (value: unknown): value is T => predicate(value),
-    }),
-    isNestedMatcher = (expected: unknown): expected is Matcher<unknown> =>
-        typeof expected === "object" &&
-        expected !== null &&
-        "matches" in expected,
-    // oxlint-disable-next-line typescript/no-unnecessary-type-parameters
-    keysMatch = <S extends object>(shape: S, candidate: object): boolean => {
-        // oxlint-disable-next-line typescript/no-unsafe-type-assertion
-        for (const key of Object.keys(shape) as (keyof S)[]) {
-            if (!(key in candidate)) {
-                return false;
-            }
-            // oxlint-disable-next-line typescript/no-unsafe-type-assertion
-            const expected = shape[key],
-                // oxlint-disable-next-line typescript/no-unsafe-type-assertion
-                actual = candidate[key as keyof object];
-            if (isNestedMatcher(expected)) {
-                if (!expected.matches(actual)) {
-                    return false;
-                }
-            } else if (actual !== expected) {
-                return false;
-            }
-        }
-        return true;
-    },
-    structuralMatcher = <S extends object, T extends S>(
-        shape: S,
-        refine?: (value: S) => value is T,
-    ): Matcher<T> => ({
-        matches: (value: unknown): value is T => {
-            if (typeof value !== "object" || value === null) {
-                return false;
-            }
-            // oxlint-disable-next-line typescript/no-unsafe-type-assertion
-            const candidate = value as S;
-            if (!keysMatch(shape, candidate)) {
-                return false;
-            }
-            if (refine && !refine(candidate)) {
-                return false;
-            }
-            return true;
-        },
-    });
-
-export const P = {
-    literal: literalMatcher,
-    type: typeMatcher,
-    when: whenMatcher,
-    any: whenMatcherAny,
-    shape: structuralMatcher,
-} as const;
-
-export type Pattern<T> = Matcher<T>;

src/primitive-union.ts

@@ -0,0 +1,104 @@
+import type { Exact } from "type-fest";
+
+import type {
+    HandlerMap,
+    Matchable,
+    Member,
+    PatternKey,
+    PatternReturns,
+    RedundantFallback,
+    UnaryFn,
+    UniverseGate,
+} from "./matcher-shared.ts";
+
+type Handlers<T extends Matchable, R> = {
+    [K in PatternKey<T>]: UnaryFn<Member<T, K>, R>;
+};
+
+// The fallback is a *second argument*, not a property of the handler map,
+// because its parameter is the remainder `Exclude<T, Member<T, keyof Handled>>` and TypeScript
+// fixes a property's contextual type before it infers its sibling keys. A later
+// argument, by contrast, is contextually typed from inference on an earlier
+// one, so the split is what makes the remainder expressible at all.
+// See development/library.md.
+type Fallback<T extends Matchable, Handled, R> = UnaryFn<
+    Exclude<T, Member<T, keyof Handled>>,
+    R
+>;
+
+// A fallback is redundant once the handler map covers `T`. The guard is folded
+// into `Handled`'s own (self-referential) constraint so it is checked *after*
+// inference; a conditional in the fallback's parameter type is evaluated while
+// `Handled` is still its constraint and would reject context-sensitive partial
+// maps. That placement also fixes where the diagnostic lands: the constraint
+// failure is reported on the argument that inferred `Handled` (the handler
+// map), so the required property is spelled as the message instead of relying
+// on its position. See development/library.md.
+type MustBePartial<T extends Matchable, Handled> =
+    PatternKey<T> extends keyof Handled ? RedundantFallback : unknown;
+
+// TypeScript does not apply the excess-property check to a generic constraint,
+// so `Exact` restores it for the generic forms: a handler map can otherwise
+// carry keys outside `T`.
+
+// Strict returns: one common `R`. Overload order is load-bearing:
+//   #1 Handlers (first) -> the exhaustive form and the autocomplete popup
+//   #2 Fallback (last)  -> accepts a partial handler map plus a fallback
+interface PrimitiveUnionMatcherStrict<T extends Matchable> {
+    <R>(handlers: Handlers<T, R> & UniverseGate<T>): UnaryFn<T, R>;
+    <
+        R,
+        Handled extends Exact<Partial<Handlers<T, R>>, Handled> &
+            MustBePartial<T, Handled>,
+    >(
+        handlers: Handled & Partial<Handlers<T, R>> & UniverseGate<T>,
+        fallback: Fallback<T, Handled, R> & UniverseGate<T>,
+    ): UnaryFn<T, R>;
+}
+
+// Widened returns: the union of every handler's return type. `P` is inferred
+// from the whole handler map, whose closed constraint supplies the
+// contextual/autocomplete type.
+interface PrimitiveUnionMatcherWidening<T extends Matchable> {
+    <P extends Exact<Handlers<T, unknown>, P>>(
+        handlers: P & UniverseGate<T>,
+    ): UnaryFn<T, PatternReturns<P>>;
+    <
+        R,
+        Handled extends Exact<Partial<Handlers<T, unknown>>, Handled> &
+            MustBePartial<T, Handled>,
+    >(
+        handlers: Handled & UniverseGate<T>,
+        fallback: Fallback<T, Handled, R> & UniverseGate<T>,
+    ): UnaryFn<T, PatternReturns<Handled> | R>;
+}
+
+const dispatch =
+    (handlers: HandlerMap, fallback?: UnaryFn<never, unknown>) =>
+    (shape: Matchable): unknown =>
+        // `handlers[true]` already coerces to the `"true"` property at runtime,
+        // identical to `handlers[String(shape)]`, so indexing with `shape`
+        // directly is sound: `shape` is a facade-checked universe member and
+        // `PatternKey` only ever produces valid property keys. The assertion is
+        // needed solely because TypeScript forbids indexing with
+        // `boolean`/`null`/`undefined` (TS2538); it buys the number fast path.
+        (
+            handlers[
+                // oxlint-disable-next-line typescript/no-unsafe-type-assertion
+                shape as string | number
+            ] ??
+            fallback ??
+            (() => {
+                throw new Error(`Unhandled shape: ${String(shape)}`);
+            })
+        )(
+            // oxlint-disable-next-line typescript/no-unsafe-type-assertion
+            shape as never,
+        );
+
+export const getPrimitiveUnionMatcher = <
+    T extends Matchable,
+>(): PrimitiveUnionMatcherStrict<T> => dispatch;
+export const getPrimitiveUnionMatcherW = <
+    T extends Matchable,
+>(): PrimitiveUnionMatcherWidening<T> => dispatch;

src/tagged-union.ts

@@ -0,0 +1,161 @@
+import type { Exact, SetRequired, UnknownRecord } from "type-fest";
+
+import type {
+    HandlerMap,
+    Matchable,
+    Member,
+    PatternKey,
+    PatternReturns,
+    RedundantFallback,
+    UnaryFn,
+    UniverseGate,
+} from "./matcher-shared.ts";
+
+// A tagged union is discriminated by one property whose values are the tags (a
+// `Matchable`). `string` and `number` tags key a handler map directly;
+// `boolean`, `null` and `undefined` are admitted too but are not property keys,
+// so they go through the `PatternKey` projection. `symbol` has no literal syntax
+// to write a handler under, and `bigint` is not a property key.
+
+// The discriminant values of `T` under `K`. `Extract` keeps the finite literal
+// tags and leaves a widened `string`/`number` as itself; a broad tag is then
+// rejected by the universe gate.
+type Tags<T extends object, K extends keyof T> = Extract<T[K], Matchable>;
+
+// The keys of `T` that can act as a discriminant. `getTaggedUnionMatcher<T>()`
+// accepts only these, so the factory rejects a key whose values are not tags.
+type Discriminated<T extends object> = {
+    [K in keyof T]: T[K] extends Matchable ? K : never;
+}[keyof T];
+
+// The member(s) of `T` narrowed to the tag(s) `V`. Distributes over `T`, so a
+// duplicated tag maps to a union of members rather than silently dropping one.
+// A member whose `K` cannot take `V` drops out; the rest have `K` narrowed to
+// `Extract<T[K], V>`. `T[K] extends V` keeps an exact member (a discriminated
+// union's declared interface) untouched, so only a property that is itself a
+// union — or optional — goes through the mapped form. `SetRequired` makes `K`
+// required unless `V` admits `undefined`: a defined tag yields `{ type: "x" }`,
+// while the `undefined` tag keeps `{ type?: never }` (absence) or
+// `{ type?: undefined }` when the property admits an explicit `undefined`.
+// Keyed by `PatternKey`, so `boolean`/`null`/`undefined` tags can key a mapped
+// type; `Member` inverts the projection against the tag set to recover the tag.
+type Narrowed<T extends object, K extends keyof T, V> = T extends object
+    ? [Extract<T[K], V>] extends [never]
+        ? never
+        : T[K] extends V
+          ? T
+          : undefined extends V
+            ? { [P in keyof T]: P extends K ? Extract<T[P], V> : T[P] }
+            : SetRequired<
+                  { [P in keyof T]: P extends K ? Extract<T[P], V> : T[P] },
+                  K
+              >
+    : never;
+
+type MapTaggedUnion<T extends object, K extends keyof T> = {
+    [P in PatternKey<Tags<T, K>>]: Narrowed<T, K, Member<Tags<T, K>, P>>;
+};
+
+type Handlers<T extends object, K extends keyof T, R> = {
+    [P in PatternKey<Tags<T, K>>]: UnaryFn<MapTaggedUnion<T, K>[P], R>;
+};
+
+// The tag values whose `PatternKey` is handled.
+type HandledTags<T extends object, K extends keyof T, Handled> = Member<
+    Tags<T, K>,
+    keyof Handled
+>;
+
+// The fallback is a *second argument*, not a property of the handler map, so
+// its parameter can be the remainder the map left uncovered: the members
+// narrowed to the tags the map did not handle. See development/library.md.
+type Fallback<T extends object, K extends keyof T, Handled, R> = UnaryFn<
+    Narrowed<T, K, Exclude<Tags<T, K>, HandledTags<T, K, Handled>>>,
+    R
+>;
+
+// A fallback is redundant once the handler map covers every tag of `T`. Folded
+// into `Handled`'s own (self-referential) constraint so it is checked *after*
+// inference; see the primitive-union matcher for why a conditional in the fallback's
+// parameter is evaluated too early.
+type MustBePartial<T extends object, K extends keyof T, Handled> =
+    PatternKey<Tags<T, K>> extends keyof Handled ? RedundantFallback : unknown;
+
+// TypeScript does not apply the excess-property check to a generic constraint,
+// so `Exact` restores it for the generic forms.
+
+// Strict returns: one common `R`. Overload order is load-bearing:
+//   #1 Handlers (first) -> the exhaustive form and the autocomplete popup
+//   #2 Fallback (last)  -> accepts a partial handler map plus a fallback
+interface TaggedUnionMatcherStrict<T extends object, K extends keyof T> {
+    <R>(handlers: Handlers<T, K, R> & UniverseGate<Tags<T, K>>): UnaryFn<T, R>;
+    <
+        R,
+        Handled extends Exact<Partial<Handlers<T, K, R>>, Handled> &
+            MustBePartial<T, K, Handled>,
+    >(
+        handlers: Handled &
+            Partial<Handlers<T, K, R>> &
+            UniverseGate<Tags<T, K>>,
+        fallback: Fallback<T, K, Handled, R> & UniverseGate<Tags<T, K>>,
+    ): UnaryFn<T, R>;
+}
+
+// Widened returns: the union of every handler's return type. `P` is inferred
+// from the whole handler map, whose closed constraint supplies the
+// contextual/autocomplete type.
+interface TaggedUnionMatcherWidening<T extends object, K extends keyof T> {
+    <P extends Exact<Handlers<T, K, unknown>, P>>(
+        handlers: P & UniverseGate<Tags<T, K>>,
+    ): UnaryFn<T, PatternReturns<P>>;
+    <
+        R,
+        Handled extends Exact<Partial<Handlers<T, K, unknown>>, Handled> &
+            MustBePartial<T, K, Handled>,
+    >(
+        handlers: Handled & UniverseGate<Tags<T, K>>,
+        fallback: Fallback<T, K, Handled, R> & UniverseGate<Tags<T, K>>,
+    ): UnaryFn<T, PatternReturns<Handled> | R>;
+}
+
+// The key-taking step of the curried factory. Naming it lets the factory return
+// `dispatch` directly, the tacit twin of the primitive-union factory's bare
+// `=> dispatch`.
+type TaggedUnionMatcherFactory<T extends object> = <K extends Discriminated<T>>(
+    k: K,
+) => TaggedUnionMatcherStrict<T, K>;
+
+type TaggedUnionMatcherWideningFactory<T extends object> = <
+    K extends Discriminated<T>,
+>(
+    k: K,
+) => TaggedUnionMatcherWidening<T, K>;
+
+const dispatch =
+    (k: PropertyKey) =>
+    (handlers: HandlerMap, fallback?: UnaryFn<never, unknown>) =>
+    (shape: object): unknown => {
+        // `object` carries no index signature, so the read needs the assertion;
+        // the factory admits only keys whose values are tags, and the map keys
+        // them by `PatternKey`, so the result is narrowed to the map's key space.
+        // oxlint-disable-next-line typescript/no-unsafe-type-assertion
+        const tag = (shape as UnknownRecord)[k] as string | number;
+        return (
+            handlers[tag] ??
+            fallback ??
+            (() => {
+                throw new Error(`Unhandled tag: ${String(tag)}`);
+            })
+        )(
+            // oxlint-disable-next-line typescript/no-unsafe-type-assertion
+            shape as never,
+        );
+    };
+
+export const getTaggedUnionMatcher = <
+    T extends object,
+>(): TaggedUnionMatcherFactory<T> => dispatch;
+
+export const getTaggedUnionMatcherW = <
+    T extends object,
+>(): TaggedUnionMatcherWideningFactory<T> => dispatch;

src/util/__tests__/lsp-completion.test.ts

@@ -0,0 +1,147 @@
+import { strict as assert } from "node:assert";
+import path from "node:path";
+import { test } from "node:test";
+
+import { expectTypeOf } from "expect-type";
+
+import {
+    type CompletionResult,
+    type CompletionTarget,
+    LspSession,
+} from "#test-utils/lsp-completion.ts";
+
+const REPO_ROOT = path.resolve(import.meta.dirname, "../../..");
+
+// The `#test-utils/*` self-reference (package.json#imports) is the one route
+// scattered test files take to reach test helpers; this pins that it resolves
+// and types without starting a language server
+// (development/testing.md § Test helpers).
+test("test-helper home: `#test-utils/…` resolves to the helper and types it", () => {
+    // Arrange
+    const target: CompletionTarget = {
+        file: "src/util/__tests__/lsp-completion.ts",
+        source: "",
+    };
+
+    // Assert — no session is constructed: the constructor spawns `tsc --lsp`,
+    // so only the resolved values and their types are checked here.
+    expectTypeOf(LspSession).toBeConstructibleWith("repo-root");
+    expectTypeOf<CompletionResult>().toMatchTypeOf<{
+        labels: readonly string[];
+    }>();
+    assert.equal(typeof LspSession, "function");
+    assert.equal(target.file, "src/util/__tests__/lsp-completion.ts");
+});
+
+// The helper drives a language server, so its contract is asserted against the
+// server. Every document below is probed from memory and carries its
+// contextual type inline, so the helper is tested without the library's code.
+const probe = (source: string, marker?: string): Promise<CompletionResult> => {
+    const session = new LspSession(REPO_ROOT);
+    const target: CompletionTarget =
+        marker === undefined
+            ? { file: "src/__probe.ts", source }
+            : { file: "src/__probe.ts", source, marker };
+    return session.completionLabelsAt(target).finally(() => session.close());
+};
+
+const HANDLERS = [
+    "type Handlers = { a: () => number; b: () => number; _?: () => number };",
+    "declare const apply: (handlers: Handlers) => Handlers;",
+];
+
+test("helper: a fresh object literal completes with its contextual keys", () => {
+    // Arrange
+    const source = [
+        ...HANDLERS,
+        "const done = apply({",
+        "    /*COMPLETE*/",
+        "});",
+        "export { done };",
+    ].join("\n");
+
+    // Act
+    const probed = probe(source);
+
+    // Assert — the position is the marker's, which the helper strips
+    expectTypeOf(probed).toEqualTypeOf<Promise<CompletionResult>>();
+    return probed.then((result) => {
+        assert.deepEqual(result.position, { line: 3, character: 4 });
+        assert.deepEqual([...result.labels], ["_?", "a", "b"]);
+    });
+});
+
+test("helper: a handled key drops out of the popup", () => {
+    // Arrange
+    const source = [
+        ...HANDLERS,
+        "const done = apply({",
+        "    b: () => 1,",
+        "    /*COMPLETE*/",
+        "});",
+        "export { done };",
+    ].join("\n");
+
+    // Act
+    const probed = probe(source);
+
+    // Assert
+    expectTypeOf<CompletionResult["labels"]>().toEqualTypeOf<
+        readonly string[]
+    >();
+    return probed.then((result) => {
+        assert.deepEqual([...result.labels], ["_?", "a"]);
+    });
+});
+
+test("helper: a custom marker is located at the line start", () => {
+    // Arrange
+    const source = [
+        "type Handlers = { a: () => number; _?: () => number };",
+        "declare const apply: (handlers: Handlers) => Handlers;",
+        "const done = apply({",
+        "@@@",
+        "});",
+        "export { done };",
+    ].join("\n");
+
+    // Act
+    const probed = probe(source, "@@@");
+
+    // Assert
+    expectTypeOf(probed).resolves.toEqualTypeOf<CompletionResult>();
+    return probed.then((result) => {
+        assert.deepEqual(result.position, { line: 3, character: 0 });
+        assert.deepEqual([...result.labels], ["_?", "a"]);
+    });
+});
+
+test("helper: labels are read from the server, not from a pattern literal", () => {
+    // Arrange
+    const source = [
+        'const text = "x";',
+        "const upper = text./*COMPLETE*/;",
+        "export { upper };",
+    ].join("\n");
+
+    // Act
+    const probed = probe(source);
+
+    // Assert
+    expectTypeOf(probed).resolves.toHaveProperty("labels");
+    return probed.then((result) => {
+        assert.ok(result.labels.includes("toUpperCase"));
+    });
+});
+
+test("helper: a source without the marker rejects", () => {
+    // Arrange
+    const source = "const text = 1;\nexport { text };\n";
+
+    // Act
+    const probed = probe(source);
+
+    // Assert
+    expectTypeOf(probed).resolves.toEqualTypeOf<CompletionResult>();
+    return assert.rejects(probed, /marker not found/);
+});

src/util/__tests__/lsp-completion.ts

@@ -0,0 +1,254 @@
+// oxlint-disable no-magic-numbers unicorn/no-null - tolerable here, this is a helper
+import { spawn, type ChildProcessWithoutNullStreams } from "node:child_process";
+import fs from "node:fs";
+import path from "node:path";
+import { setTimeout as delay } from "node:timers/promises";
+import { pathToFileURL } from "node:url";
+
+import {
+    type CompletionItem,
+    type CompletionList,
+    CompletionRequest,
+    ConfigurationRequest,
+    createMessageConnection,
+    DidOpenTextDocumentNotification,
+    InitializedNotification,
+    InitializeRequest,
+    type MessageConnection,
+    ShutdownRequest,
+    StreamMessageReader,
+    StreamMessageWriter,
+} from "vscode-languageserver-protocol/node";
+
+/**
+ * Print the completion labels TypeScript's language server offers at a marker
+ * inside a source file.
+ *
+ * Usage: node --strip-types src/util/__tests__/lsp-completion.ts <file> [<marker>]
+ *
+ * The marker (default `DEFAULT_MARKER`, a `COMPLETE` block comment) is stripped
+ * from the source before it is sent; the request position is where the marker
+ * stood. Point it at a scratch file whose expected type is a matcher pattern to
+ * inspect the popup.
+ *
+ * `LspSession` exposes the same probe to the test suite, so completion is
+ * asserted against the server rather than the type system.
+ *
+ * Why a script and not a type assertion: completion is a contextual-type
+ * property that `expect-type` cannot observe, and `Parameters<…>` resolves only
+ * a single overload. The server is the only ground truth. See
+ * development/testing.md § Autocomplete.
+ */
+
+const DEFAULT_MARKER = "/*COMPLETE*/";
+const ARGV_PREFIX_LENGTH = 2;
+const EXIT_DELAY_MS = 200;
+const FAILURE_EXIT_CODE = 1;
+const USAGE = "usage: lsp-completion.ts <file> [<marker>]";
+
+export interface Position {
+    readonly line: number;
+    readonly character: number;
+}
+
+/** A document to probe: `source` carries `marker`, which is stripped. */
+export interface CompletionTarget {
+    /** Path relative to the session's repo root; drives URI and resolution. */
+    readonly file: string;
+    readonly source: string;
+    readonly marker?: string;
+}
+
+export interface CompletionResult {
+    readonly position: Position;
+    readonly labels: readonly string[];
+}
+
+const completionLabels = (
+    result: CompletionItem[] | CompletionList | null,
+): readonly string[] => {
+    if (result === null) {
+        return [];
+    }
+    const items = Array.isArray(result) ? result : result.items;
+    return items
+        .map((item) => item.label)
+        .toSorted((left, right) => left.localeCompare(right));
+};
+
+const markerPosition = (source: string, marker: string): Position => {
+    const index = source.indexOf(marker);
+    if (index === -1) {
+        throw new Error(`marker not found: ${marker}`);
+    }
+    const lines = source.slice(0, index).split("\n");
+    const lineIndex = lines.length - 1;
+    const line = lines[lineIndex];
+    return { line: lineIndex, character: line === undefined ? 0 : line.length };
+};
+
+/**
+ * One language server, initialized on first use, shared across probes. Callers
+ * own the lifecycle and must `close()` it.
+ */
+export class LspSession {
+    readonly #child: ChildProcessWithoutNullStreams;
+    readonly #connection: MessageConnection;
+    readonly #repoRoot: string;
+    #ready: Promise<void> | undefined;
+
+    public constructor(repoRoot: string) {
+        this.#repoRoot = repoRoot;
+        this.#child = spawn(
+            path.join(repoRoot, "node_modules", ".bin", "tsc"),
+            ["--lsp", "--stdio"],
+            { cwd: repoRoot },
+        );
+        this.#child.stderr.on("data", (chunk: Buffer) => {
+            process.stderr.write(chunk);
+        });
+        this.#connection = createMessageConnection(
+            new StreamMessageReader(this.#child.stdout),
+            new StreamMessageWriter(this.#child.stdin),
+        );
+        // Server -> client requests the server waits on: answer so it proceeds.
+        // `workspace/configuration` wants one reply per requested item; a
+        // catch-all covers the rest (e.g. `client/registerCapability`).
+        this.#connection.onRequest(ConfigurationRequest.type, ({ items }) =>
+            items.map(() => null),
+        );
+        this.#connection.onRequest(() => null);
+        this.#connection.listen();
+    }
+
+    public completionLabelsAt(
+        target: CompletionTarget,
+    ): Promise<CompletionResult> {
+        return this.#ensureInitialized()
+            .then(() => this.#open(target))
+            .then(({ uri, position }) =>
+                this.#connection
+                    .sendRequest(CompletionRequest.type, {
+                        textDocument: { uri },
+                        position,
+                        context: { triggerKind: 1 },
+                    })
+                    .then((result) => ({
+                        position,
+                        labels: completionLabels(result),
+                    })),
+            );
+    }
+
+    /** `shutdown` + close stdin, then kill the server; safe after a failed probe. */
+    public close(): Promise<void> {
+        return (
+            this.#connection
+                .sendRequest(ShutdownRequest.type)
+                // The TS 7 Go server logs a bare `context canceled` to stderr
+                // when it handles `exit`; EOF on stdin shuts it down cleanly
+                // (exit 0, no output) instead.
+                .then(() => {
+                    this.#child.stdin.end();
+                })
+                .then(() => delay(EXIT_DELAY_MS))
+                .finally(() => {
+                    this.#connection.dispose();
+                    this.#child.kill();
+                })
+        );
+    }
+
+    #ensureInitialized(): Promise<void> {
+        this.#ready ??= this.#initialize();
+        return this.#ready;
+    }
+
+    #initialize(): Promise<void> {
+        return this.#connection
+            .sendRequest(InitializeRequest.type, {
+                processId: process.pid,
+                rootUri: pathToFileURL(this.#repoRoot).href,
+                workspaceFolders: [
+                    { uri: pathToFileURL(this.#repoRoot).href, name: "repo" },
+                ],
+                capabilities: {
+                    textDocument: {
+                        completion: {
+                            completionItem: { snippetSupport: false },
+                        },
+                        publishDiagnostics: {},
+                    },
+                },
+            })
+            .then(() =>
+                this.#connection.sendNotification(
+                    InitializedNotification.type,
+                    {},
+                ),
+            );
+    }
+
+    #open(target: CompletionTarget): Promise<{
+        readonly uri: string;
+        readonly position: Position;
+    }> {
+        const absolute = path.resolve(this.#repoRoot, target.file);
+        const marker = target.marker ?? DEFAULT_MARKER;
+        const position = markerPosition(target.source, marker);
+        const text = target.source.replace(marker, "");
+        const uri = pathToFileURL(absolute).href;
+        // The server handles `didOpen` in order before the completion request,
+        // so no settle delay is needed.
+        return this.#connection
+            .sendNotification(DidOpenTextDocumentNotification.type, {
+                textDocument: {
+                    uri,
+                    languageId: "typescript",
+                    version: 1,
+                    text,
+                },
+            })
+            .then(() => ({ uri, position }));
+    }
+}
+
+const main = (args: readonly string[]): Promise<void> => {
+    const [file, markerArgument] = args;
+    if (file === undefined) {
+        process.stderr.write(`${USAGE}\n`);
+        process.exitCode = FAILURE_EXIT_CODE;
+        return Promise.resolve();
+    }
+
+    const repoRoot = path.resolve(import.meta.dirname, "../../..");
+    const absolute = path.resolve(repoRoot, file);
+    const source = fs.readFileSync(absolute, "utf8");
+    const session = new LspSession(repoRoot);
+    return session
+        .completionLabelsAt({
+            file,
+            source,
+            marker: markerArgument ?? DEFAULT_MARKER,
+        })
+        .then(({ position, labels }) => {
+            process.stdout.write(
+                `\n[${file}] completions @ ${position.line}:${position.character}:\n${labels.join(", ")}\n`,
+            );
+        })
+        .finally(() => session.close());
+};
+
+const isEntryPoint = (): boolean => {
+    const [entry] = process.argv.slice(1, 2);
+    return entry !== undefined && import.meta.url === pathToFileURL(entry).href;
+};
+
+if (isEntryPoint()) {
+    main(process.argv.slice(ARGV_PREFIX_LENGTH)).catch((error: unknown) => {
+        process.stderr.write(
+            `${error instanceof Error ? error.message : String(error)}\n`,
+        );
+        process.exitCode = FAILURE_EXIT_CODE;
+    });
+}