chore: cut 1.0.0

DESIGN.md

@@ -532,6 +532,36 @@ keyframes declare `from` and nothing else, so the browser animates back to
 whatever the element's real style is — which fades out correctly on every
 variant, focused or not, without hard-coding the resting colours.
 
+## What 1.0.0 promises
+
+A version number is a promise about change, so it is worth saying which one.
+
+**The exported names are the contract.** Every entry point lists its exports by
+hand, and a test asserts each list exactly — `index.test.ts` and
+`entryPoints.test.ts`. Removing or renaming any of them is a major version.
+That list is deliberately short: the package exported 66 names by accident
+before 1.0.0, through four `export *` lines, and shipping that would have
+frozen reducer internals and a cache-clearing test seam as public API forever.
+
+**What is not exported may change in a patch.** The reducer, the validation
+helpers, the caret arithmetic — all internal. If you need one of them, ask for
+it rather than reaching into `dist/`; a name that is asked for can be exported
+on purpose.
+
+**The CSS class names are part of the contract too**, because the stylesheet is
+opt-in and people build on the markup. `rfi-input`, `rfi-field`, `rfi-label`,
+`rfi-adornment`, `rfi-flag` and the rest behave like exports.
+
+**Behaviour that is documented here is behaviour you can rely on.** The
+canonical-versus-display split, `null` never being `NaN`, the caret rules, the
+grouping following the locale. Where a platform forced an unusual decision it is
+written down with the reason, so it does not look like something safe to
+"clean up" later.
+
+**What is not promised** is in [Limits](#limits) below: numbers above 2^53,
+digit glyphs outside ASCII, and the support-matrix rows nobody has hardware to
+verify. Those are stated rather than implied.
+
 ## Limits
 
 What this library does not do, stated plainly, because finding out later is

HISTORY.md

@@ -139,3 +139,20 @@ because nobody owns every phone.
 
 The headline used to say "works on every device". It says something narrower
 now, and truer.
+
+## 1.0.0
+
+Cut after seven rounds of looking for reasons not to.
+
+Each round found something, and the list is a fair description of what a
+"finished" side project actually hides: an API of 66 names nobody had chosen,
+TypeScript declarations that were wrong for CommonJS consumers, two critical
+ARIA violations in the widget the docs told people to copy, a `'use client'`
+banner that made the parsing unusable in the server action the examples
+demonstrated, a load-bearing rule with no test holding it, and grouping that
+was wrong for every Indian locale.
+
+The seventh round found nothing, and a deliberate hunt for memory leaks found
+one uncancelled animation frame that was not a leak.
+
+None of that was visible from the outside. The library worked the whole time.

release-please-config.json

@@ -1,25 +1,55 @@
 {
   "$schema": "https://raw.githubusercontent.com/googleapis/release-please/main/schemas/config.json",
-  "$comment": "separate-pull-requests must stay true. With false, the Merge plugin builds one PR on a branch carrying no component, but getBranchComponent() falls back to the package.json name — so buildRelease compares 'undefined' against 'react-financial-input', returns undefined, and no release is ever tagged. There is only one package here, so this still produces a single release PR.",
+  "$comment": "separate-pull-requests must stay true. With false, the Merge plugin builds one PR on a branch carrying no component, but getBranchComponent() falls back to the package.json name \u2014 so buildRelease compares 'undefined' against 'react-financial-input', returns undefined, and no release is ever tagged. There is only one package here, so this still produces a single release PR.",
   "separate-pull-requests": true,
   "packages": {
     ".": {
       "release-type": "node",
       "changelog-path": "CHANGELOG.md",
       "include-component-in-tag": false,
-      "bump-minor-pre-major": true,
-      "bump-patch-for-minor-pre-major": false,
       "changelog-sections": [
-        { "type": "feat", "section": "Features" },
-        { "type": "fix", "section": "Bug fixes" },
-        { "type": "perf", "section": "Performance" },
-        { "type": "docs", "section": "Documentation" },
-        { "type": "refactor", "section": "Refactoring" },
-        { "type": "test", "section": "Tests", "hidden": true },
-        { "type": "build", "section": "Build", "hidden": true },
-        { "type": "ci", "section": "CI", "hidden": true },
-        { "type": "chore", "section": "Chores", "hidden": true }
+        {
+          "type": "feat",
+          "section": "Features"
+        },
+        {
+          "type": "fix",
+          "section": "Bug fixes"
+        },
+        {
+          "type": "perf",
+          "section": "Performance"
+        },
+        {
+          "type": "docs",
+          "section": "Documentation"
+        },
+        {
+          "type": "refactor",
+          "section": "Refactoring"
+        },
+        {
+          "type": "test",
+          "section": "Tests",
+          "hidden": true
+        },
+        {
+          "type": "build",
+          "section": "Build",
+          "hidden": true
+        },
+        {
+          "type": "ci",
+          "section": "CI",
+          "hidden": true
+        },
+        {
+          "type": "chore",
+          "section": "Chores",
+          "hidden": true
+        }
       ]
     }
-  }
+  },
+  "$comment2": "bump-minor-pre-major and bump-patch-for-minor-pre-major were removed at 1.0.0. They only apply below 1.0, where they kept breaking changes at a minor bump. Past 1.0.0 a BREAKING CHANGE footer means a major, which is the whole point of the version."
 }