@invisibleloop/pulse 0.20.2 → 0.20.4

package/.claude/commands/verify.md

@@ -1,12 +1,18 @@
 # Verify
 
-Run the full verification loop on the current page or the page specified in $ARGUMENTS.
+Run the verification loop on the current page or the page specified in $ARGUMENTS.
+
+**Modes:**
+- `/verify` — full loop including Lighthouse (desktop + mobile), perf trace, and mobile layout check. Use when the user approves the design or signals they're ready to ship.
+- `/verify --quick` — skips Lighthouse and perf trace; just validate → screenshot → console → review → stamp. Use during iterative building: design rounds, debugging visual issues, mid-build checkpoints. Fast enough to run between every change.
+
+If `--quick` appears anywhere in $ARGUMENTS, run in quick mode (steps 6, 7, 8, and 9 are skipped). The route/file path is whatever remains in $ARGUMENTS after removing `--quick`.
 
 ## Steps
 
 ### 1. Identify the target
 
-If $ARGUMENTS is provided, use it as the file path or route. Otherwise, identify the most recently edited spec file from context.
+If $ARGUMENTS is provided, use it (minus `--quick` if present) as the file path or route. Otherwise, identify the most recently edited spec file from context.
 
 ### 2. Design approval (new builds only)
 
@@ -30,7 +36,7 @@ Use `pulse_fetch_page` to confirm the server is responding for the route. If it
 
 Use `mcp__chrome-devtools__navigate_page` to load the page route, then `mcp__chrome-devtools__take_screenshot` to capture the result. Describe what you see — layout, content, any obvious rendering issues.
 
-### 6. Lighthouse — desktop
+### 6. Lighthouse — desktop *(skip in quick mode)*
 
 **Pre-flight (required before every Lighthouse run):**
 1. Call `pulse_build` to produce a production build and start the production server on port 3001.
@@ -40,7 +46,7 @@ Then run `mcp__chrome-devtools__lighthouse_audit` with `{ "device": "desktop" }`
 
 **Pass bar: Accessibility, Best Practices, and SEO must all be 100.** Performance is measured and reported but is not a hard requirement (it varies with machine load). Report the actual scores. If Accessibility, Best Practices, or SEO is below 100, identify the failing audit(s), fix the issue, and restart from step 3.
 
-### 7. Lighthouse — mobile
+### 7. Lighthouse — mobile *(skip in quick mode)*
 
 The browser should still be on `http://localhost:3001/` from step 6. Run `mcp__chrome-devtools__lighthouse_audit` with `{ "device": "mobile" }`.
 
@@ -48,7 +54,7 @@ The browser should still be on `http://localhost:3001/` from step 6. Run `mcp__c
 
 **Stay on the production server (port 3001)** — steps 8 and 9 use it too. Do not call `pulse_restart_server` yet.
 
-### 8. Mobile layout check
+### 8. Mobile layout check *(skip in quick mode)*
 
 Emulate a mobile viewport and take a screenshot to catch wrapping, overflow, and layout issues before they become Lighthouse failures.
 
@@ -70,7 +76,7 @@ Fix any issues, then reset to desktop before continuing:
 mcp__chrome-devtools__emulate  viewport: "1440x900x1"
 ```
 
-### 9. Performance
+### 9. Performance *(skip in quick mode)*
 
 **Run the trace against the production server, not dev.** Navigate to `http://localhost:3001/<route>` first — if the production server is no longer running, run `pulse_build` again. Tracing the dev server (port 3000) reports misleading LCP/CLS. Then run `mcp__chrome-devtools__performance_start_trace` with `reload: true` and `autoStop: true`. Report LCP and CLS. Flag any LCP insight that suggests a fixable problem (render-blocking resources, large image load delay, etc.).
 
@@ -102,11 +108,11 @@ Call `pulse_stamp`. This writes `.pulse-verified` via the MCP server, which is m
 
 Summarise:
 - Validation: pass or fail (with errors if any)
-- Visual: desktop screenshot and mobile layout check — any issues found and fixed
-- Lighthouse desktop: scores for Accessibility, Best Practices, SEO
-- Lighthouse mobile: scores for Accessibility, Best Practices, SEO
-- Performance: LCP and CLS values, any flagged insights
+- Visual: desktop screenshot — any issues found and fixed
 - Console: any errors
 - Review: pass or issues found and fixed
+- **Full mode only:** Lighthouse desktop/mobile scores, mobile layout check, LCP and CLS values
+
+In quick mode, end with: "Quick check passed — run `/verify` when you're ready for the full Lighthouse audit."
 
-Only confirm the page is good when validation passes, both Lighthouse runs are 100/100/100, CLS is 0.00, and there are no console errors. Otherwise, fix and run `/verify` again.
+In full mode, only confirm the page is good when validation passes, both Lighthouse runs are 100/100/100, CLS is 0.00, and there are no console errors. Otherwise, fix and run `/verify` again.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@invisibleloop/pulse",
-  "version": "0.20.2",
+  "version": "0.20.4",
   "type": "module",
   "description": "AI-first frontend framework. The spec is the source of truth.",
   "license": "MIT",

package/public/.pulse-ui-version

@@ -1 +1 @@
-0.20.2
+0.20.4

package/scripts/css-purge.js

@@ -12,9 +12,9 @@
 export function extractUsedClasses(htmlContents, jsContents, safelist = []) {
   const used = new Set()
 
-  // Extract from class="..." in both rendered HTML and JS template literals
-  for (const content of [...htmlContents, ...jsContents]) {
-    const re = /class(?:Name)?=["']([^"'\n]+)["']/g
+  // Extract from class="..." in rendered HTML (no template expressions to worry about)
+  for (const content of htmlContents) {
+    const re = /class=["']([^"'\n]+)["']/g
     let m
     while ((m = re.exec(content)) !== null) {
       for (const cls of m[1].trim().split(/\s+/)) {
@@ -23,6 +23,41 @@ export function extractUsedClasses(htmlContents, jsContents, safelist = []) {
     }
   }
 
+  // Extract from class= attributes in JS source. Template literal class attributes
+  // like class="foo${expr ? ' bar' : ''}" have their expressions interrupted by
+  // single quotes inside the ternary, so [^"'\n]+ stops mid-expression and misses
+  // the base class. Instead: scan for class= followed by a double-quoted string
+  // OR a backtick-delimited segment, extract everything up to the first ${ or the
+  // closing delimiter, and tokenise only the static prefix.
+  for (const js of jsContents) {
+    // Match class="...anything up to closing quote or ${..."
+    // Using a two-pass approach: find class= then manually scan to get the static part
+    let i = 0
+    while (i < js.length) {
+      const idx = js.indexOf('class=', i)
+      if (idx === -1) break
+      i = idx + 6 // past 'class='
+
+      // Skip optional whitespace (shouldn't be any but be safe)
+      while (i < js.length && (js[i] === ' ' || js[i] === '\t')) i++
+
+      const quote = js[i]
+      if (quote !== '"' && quote !== "'" && quote !== '`') continue
+      i++ // past opening quote
+
+      // Collect characters until we hit the closing quote, a ${, or end of string
+      let staticPart = ''
+      while (i < js.length && js[i] !== quote && !(js[i] === '$' && js[i + 1] === '{')) {
+        staticPart += js[i++]
+      }
+
+      // Tokenise the static prefix
+      for (const cls of staticPart.trim().split(/\s+/)) {
+        if (cls && /^[a-zA-Z][a-zA-Z0-9_-]*$/.test(cls)) used.add(cls)
+      }
+    }
+  }
+
   // Scan JS source for quoted strings that look like CSS class names (contain hyphens)
   // Catches conditionally-applied classes like 'ui-btn--disabled' in ternaries
   for (const js of jsContents) {
@@ -50,6 +85,33 @@ export function extractUsedClasses(htmlContents, jsContents, safelist = []) {
     }
   }
 
+  // Pulse re-renders via innerHTML — classes added after mutations appear as
+  // template literal expressions inside class= attributes, not classList calls.
+  // Pattern: class="${state.x ? 'foo' : 'bar'}" or class="base ${state.x ? 'active' : ''}"
+  // The hyphen-only scan above misses single-word classes like 'active', 'open',
+  // 'visible', 'selected', 'hidden' that are common state-driven class names.
+  // Extract every quoted string that appears inside a class= template expression.
+  for (const js of jsContents) {
+    // Match: class=`...` or class="${...}" template attribute values
+    const attrRe = /class(?:Name)?=`([^`]*)`/g
+    let m
+    while ((m = attrRe.exec(js)) !== null) {
+      const attrContent = m[1]
+      // Collect static class tokens (outside ${...} expressions)
+      for (const cls of attrContent.replace(/\$\{[^}]*\}/g, ' ').trim().split(/\s+/)) {
+        if (cls && /^[a-zA-Z][a-zA-Z0-9_-]*$/.test(cls)) used.add(cls)
+      }
+      // Collect quoted strings inside ${...} expressions — these are the
+      // conditional class names: ${state.open ? 'open' : ''} → 'open'
+      for (const [, inner] of attrContent.matchAll(/\$\{([^}]*)\}/g)) {
+        for (const [, cls] of inner.matchAll(/['"]([a-zA-Z][a-zA-Z0-9_-]*)['"]|`([a-zA-Z][a-zA-Z0-9_-]*)`/g)) {
+          const name = cls
+          if (name && /^[a-zA-Z][a-zA-Z0-9_-]*$/.test(name)) used.add(name)
+        }
+      }
+    }
+  }
+
   // Apply safelist from pulse.config.js — css.safelist accepts strings or RegExp.
   // String entries are added directly; RegExp entries are matched against all
   // class names defined in the CSS (resolved later in purgeCss via selectorUsed

package/scripts/css-purge.test.js

@@ -119,6 +119,22 @@ test('multiple classList arguments in one call are all collected', () => {
   for (const cls of ['one', 'two-x', 'three']) assert.ok(used.has(cls), cls)
 })
 
+// Regression: class name immediately followed by ${ (no space before the expression)
+// was not extracted — "saved-palette${...}" yielded "saved-palette${expanded" as a
+// single token, so the rule was silently purged in prod while dev (no purging) looked fine.
+test('class name immediately before ${ is extracted correctly', () => {
+  const js = [
+    // No space before ${ — the broken pattern
+    `const v = \`<div class="saved-palette\${expanded ? ' saved-palette--expanded' : ''}"></div>\``,
+    // Multiple static classes, last one butts into ${
+    `const v = \`<button class="btn btn--primary\${loading ? ' btn--loading' : ''}"></button>\``,
+  ]
+  const used = extractUsedClasses([], js)
+  for (const cls of ['saved-palette', 'btn', 'btn--primary']) {
+    assert.ok(used.has(cls), `Expected "${cls}" to be extracted but it was missing`)
+  }
+})
+
 test('runtime-toggled compound rule survives the purge', () => {
   const css = `.docs-sidebar { transform: translateX(-100%); }\n.docs-sidebar.open { transform: translateX(0); }\n.unused { color: red; }`
   const used = extractUsedClasses(

package/src/agent/commands/verify.md

@@ -1,12 +1,18 @@
 # Verify
 
-Run the full verification loop on the current page or the page specified in $ARGUMENTS.
+Run the verification loop on the current page or the page specified in $ARGUMENTS.
+
+**Modes:**
+- `/verify` — full loop including Lighthouse (desktop + mobile), perf trace, and mobile layout check. Use when the user approves the design or signals they're ready to ship.
+- `/verify --quick` — skips Lighthouse and perf trace; just validate → screenshot → console → review → stamp. Use during iterative building: design rounds, debugging visual issues, mid-build checkpoints. Fast enough to run between every change.
+
+If `--quick` appears anywhere in $ARGUMENTS, run in quick mode (steps 6, 7, 8, and 9 are skipped). The route/file path is whatever remains in $ARGUMENTS after removing `--quick`.
 
 ## Steps
 
 ### 1. Identify the target
 
-If $ARGUMENTS is provided, use it as the file path or route. Otherwise, identify the most recently edited spec file from context.
+If $ARGUMENTS is provided, use it (minus `--quick` if present) as the file path or route. Otherwise, identify the most recently edited spec file from context.
 
 ### 2. Design approval (new builds only)
 
@@ -30,7 +36,7 @@ Use `pulse_fetch_page` to confirm the server is responding for the route. If it
 
 Use `mcp__chrome-devtools__navigate_page` to load the page route, then `mcp__chrome-devtools__take_screenshot` to capture the result. Describe what you see — layout, content, any obvious rendering issues.
 
-### 6. Lighthouse — desktop
+### 6. Lighthouse — desktop *(skip in quick mode)*
 
 **Pre-flight (required before every Lighthouse run):**
 1. Call `pulse_build` to produce a production build and start the production server on port 3001.
@@ -40,7 +46,7 @@ Then run `mcp__chrome-devtools__lighthouse_audit` with `{ "device": "desktop" }`
 
 **Pass bar: Accessibility, Best Practices, and SEO must all be 100.** Performance is measured and reported but is not a hard requirement (it varies with machine load). Report the actual scores. If Accessibility, Best Practices, or SEO is below 100, identify the failing audit(s), fix the issue, and restart from step 3.
 
-### 7. Lighthouse — mobile
+### 7. Lighthouse — mobile *(skip in quick mode)*
 
 The browser should still be on `http://localhost:3001/` from step 6. Run `mcp__chrome-devtools__lighthouse_audit` with `{ "device": "mobile" }`.
 
@@ -48,7 +54,7 @@ The browser should still be on `http://localhost:3001/` from step 6. Run `mcp__c
 
 **Stay on the production server (port 3001)** — steps 8 and 9 use it too. Do not call `pulse_restart_server` yet.
 
-### 8. Mobile layout check
+### 8. Mobile layout check *(skip in quick mode)*
 
 Emulate a mobile viewport and take a screenshot to catch wrapping, overflow, and layout issues before they become Lighthouse failures.
 
@@ -70,7 +76,7 @@ Fix any issues, then reset to desktop before continuing:
 mcp__chrome-devtools__emulate  viewport: "1440x900x1"
 ```
 
-### 9. Performance
+### 9. Performance *(skip in quick mode)*
 
 **Run the trace against the production server, not dev.** Navigate to `http://localhost:3001/<route>` first — if the production server is no longer running, run `pulse_build` again. Tracing the dev server (port 3000) reports misleading LCP/CLS. Then run `mcp__chrome-devtools__performance_start_trace` with `reload: true` and `autoStop: true`. Report LCP and CLS. Flag any LCP insight that suggests a fixable problem (render-blocking resources, large image load delay, etc.).
 
@@ -102,11 +108,11 @@ Call `pulse_stamp`. This writes `.pulse-verified` via the MCP server, which is m
 
 Summarise:
 - Validation: pass or fail (with errors if any)
-- Visual: desktop screenshot and mobile layout check — any issues found and fixed
-- Lighthouse desktop: scores for Accessibility, Best Practices, SEO
-- Lighthouse mobile: scores for Accessibility, Best Practices, SEO
-- Performance: LCP and CLS values, any flagged insights
+- Visual: desktop screenshot — any issues found and fixed
 - Console: any errors
 - Review: pass or issues found and fixed
+- **Full mode only:** Lighthouse desktop/mobile scores, mobile layout check, LCP and CLS values
+
+In quick mode, end with: "Quick check passed — run `/verify` when you're ready for the full Lighthouse audit."
 
-Only confirm the page is good when validation passes, both Lighthouse runs are 100/100/100, CLS is 0.00, and there are no console errors. Otherwise, fix and run `/verify` again.
+In full mode, only confirm the page is good when validation passes, both Lighthouse runs are 100/100/100, CLS is 0.00, and there are no console errors. Otherwise, fix and run `/verify` again.

package/src/agent/guide-components.md

@@ -194,7 +194,7 @@ To add a missing icon: copy a Lucide path (MIT) into `src/ui/icons.js` using `s(
 | `list` | `items` (array of HTML strings), `ordered`, `gap` (xs/sm/md) |
 | `prose` | `content` (raw HTML — not escaped), `size` (sm/base/lg) — for CMS/markdown content you don't control |
 
-**Never write raw `<h1>`–`<h6>` or `<ul>`/`<ol>` without styling.** Use `heading()` and `list()` instead.
+**Never write raw `<h1>`–`<h6>` or `<ul>`/`<ol>` without styling.** Use `heading()` and `list()` instead. *(Mode A only — in a creative override, raw headings styled directly with utility classes or custom CSS are correct. The `heading()` component adds wrapper markup that can fight custom layouts.)*
 
 **Heading order across components.** `card`, `feature`, `section`, `pricing`, `cta`, and `modal` all accept a `level` prop. Use it to keep the document outline correct without changing visual style. For example, if a page has an h2 section title and each card inside it has a title, pass `level: 3` on the cards (the default) — but if the cards are the first heading on the page, pass `level: 2`. The CSS class stays the same (`ui-card-title` etc.) so appearance is unaffected.
 
@@ -284,7 +284,7 @@ This is by design: vibes are a paint job, not a structural transformation. Compo
 
 | Component | Key props |
 |-----------|-----------|
-| `nav` | `logo` (HTML slot), `logoHref`, `links` ([{label,href}]), `action` (HTML slot), `sticky`, `background` (any CSS value — overrides default surface colour), `color` (foreground colour — set when custom background doesn't contrast with default muted text) |
+| `nav` | `logo` (HTML slot), `logoHref`, `links` ([{label,href}]), `action` (HTML slot), `sticky`, `background` (any CSS value — overrides default surface colour), `color` (foreground colour — set when custom background doesn't contrast with default muted text) — ⚠ **renders a `<nav>` landmark internally. Never wrap it in another `<nav>` or `<header><nav>` — that creates a duplicate landmark Lighthouse flags as an accessibility failure. Wrapping in `<header>` alone is fine.** |
 | `hero` | `eyebrow`, `title`, `subtitle`, `actions` (HTML slot), `image` (HTML slot — activates image layout), `imageAlign` (right/left), `align` (center/left), `size` (md/sm), `layout` (split/asymmetric/overlap — when image provided: 'split'=50/50, 'asymmetric'=60/40 text-heavy, 'overlap'=image fills section with text overlay), `gradient` (true/false or preset: 'purple'\|'blue'\|'green'\|'rose'\|'orange'), `background` (any CSS value), `eyebrowColor` |
 | `phoneFrame` | `content` (HTML slot — rendered inside the phone screen), `animate` (boolean, default `false` — enables a gentle 3-D tilt on hover). Pure CSS phone mockup with dynamic island, side buttons, home indicator. Use as the `image` slot in `hero()` for app landing pages. Add `animate: true` whenever the phone is a prominent hero element. |
 | `feature` | `image` (HTML slot — rendered above icon), `icon` (HTML slot), `title`, `level` (1–6, default 3), `description` (or `body` as alias), `center` |

package/src/agent/identity.md

@@ -44,8 +44,8 @@ The component rules marked *(Mode A)* below are suspended under a declared creat
 - Hardcode hex colours in CSS. Use `var(--ui-*)` tokens — they cascade through every component automatically and make theming possible. The one place hex values belong is `public/theme.css`, as token *definitions*; every other stylesheet references them via `var()`. A lint hook enforces this.
 - Use inline `style=""` attributes. Use utility classes or `var(--ui-*)` tokens in a stylesheet. *Exception:* structural image styles with no utility-class equivalent (`aspect-ratio`, `object-fit`) are permitted inline — these are layout facts, not theming, and the design gallery uses them this way.
 - Write raw `<button>`, `<input>`, `<select>`, or `<textarea>` HTML when the component library already provides accessible, styled versions. *(Mode A — under a creative override, functional atoms should still come from components unless there is a specific design reason.)*
-- Write raw `<h1>`–`<h6>` without styling. Use `heading({ level, text })` instead. *(Mode A)*
-- Write raw `<ul>` or `<ol>` without styling. Use `list({ items })` instead. *(Mode A)*
+- Write raw `<h1>`–`<h6>` without styling. Use `heading({ level, text })` instead. *(Mode A — under a creative override, raw headings styled with utility classes or custom CSS are correct; `heading()` adds wrapper markup that fights custom layouts.)*
+- Write raw `<ul>` or `<ol>` without styling. Use `list({ items })` instead. *(Mode A — under a creative override, raw lists are permitted when the component markup would interfere with the layout.)*
 - Output CMS or database HTML without a `prose()` wrapper. Raw HTML from external sources has no styling — always wrap it in `prose({ content: html })`.
 - Use `data-event` on text inputs. Re-rendering on every keystroke destroys focus. Use uncontrolled inputs and read values from `FormData` in `action.onStart` or `action.run`.
 - Skip `onError` in an action. It is required. Always handle failure.

package/src/agent/workflow.md

@@ -9,9 +9,10 @@
 | New site, new branded page, or "build me X from scratch" | `pulse_intake` → `pulse_sketch` → `pulse_intent` → build |
 | Editing an existing page, adding a section, fixing a bug | Go straight to Step 1 — Understand |
 | "Add X to the existing Y" (one-liner) | Read the relevant file, make the change, run `/verify` |
+| Iterating on design before user approves | `/verify --quick` between rounds; full `/verify` once approved |
 | Stuck on which component to use | `pulse_intent("describe what I'm building")` |
 | Colours failing contrast / unsure about palette | `pulse_check_contrast` |
-| Something looks wrong in the browser | `pulse_restart_server`, then `/verify` |
+| Something looks wrong in the browser | `pulse_restart_server`, then `/verify --quick` |
 | Lighthouse < 100 | Fix the flagged audit, run `/verify` again — do not skip to commit |
 
 ---
@@ -224,6 +225,8 @@ After the Write tool completes, call `pulse_create_page(name)` to register the p
 
 After writing the first draft, call `pulse_suggest(content)` before running the hard validator. It catches obvious omissions collaboratively — a second opinion, not a gate. Then call `pulse_review(file, { quick: true })` for a lightweight structural pass (main landmark, data-event on inputs, missing onError, hex in view). Neither is a substitute for the full review, but catching issues here is faster than discovering them at Lighthouse.
 
+If the user is iterating on the design (multiple rounds of changes before giving a final go-ahead), use `/verify --quick` between rounds — it validates, screenshots, checks console, and runs a quick code review without the 90-second Lighthouse cost. Save the full `/verify` for after the user explicitly approves the design.
+
 **Rule: never output `✓` before the tool call that confirms the result.**
 
 ---
@@ -278,7 +281,7 @@ Do this before every screenshot, console check, or visual inspection. Never debu
 
 ### Phase 5b — Full verification (pass gate)
 
-**Run `/verify` now.** Do not execute these steps manually.
+**Run `/verify` now** (full mode — not `--quick`). Do not execute these steps manually.
 
 `/verify` validates, screenshots, runs Lighthouse (desktop + mobile), runs the performance trace (LCP + CLS), checks console errors, runs `pulse_review`, and writes the `.pulse-verified` stamp. Running steps manually does not write the stamp.
 

package/src/cli/dev.js

@@ -118,6 +118,15 @@ if (specs.length === 0) {
   process.exit(1)
 }
 
+// Warn if a production build exists — public/dist/ is ignored by the dev server
+// (manifest: {} forces source-file hydration) but its presence can cause confusion
+// because static files in public/dist/ ARE served by the static file handler.
+// If the user ran pulse build and sees stale styles, deleting public/dist/ fixes it.
+const DIST_DIR = path.join(ROOT, 'public', 'dist')
+if (fs.existsSync(path.join(DIST_DIR, 'manifest.json'))) {
+  console.warn(`\n  ⚠  A production build exists in public/dist/\n     Dev mode ignores the manifest and always serves source files, but if\n     styles look wrong after a build, run: rm -rf public/dist/\n`)
+}
+
 // ---------------------------------------------------------------------------
 // Hot reload — SSE clients + file watcher
 // ---------------------------------------------------------------------------
@@ -257,6 +266,14 @@ const { updateSpecs, updateStore } = await createServer(specs, {
   onRequest(req, res) {
     const url = req.url.split('?')[0]
 
+    // MCP hot-reload trigger — POST to this instead of killing the process
+    if (url === '/_pulse/trigger-reload' && req.method === 'POST') {
+      triggerReload('MCP trigger')
+      res.writeHead(204)
+      res.end()
+      return false
+    }
+
     // SSE endpoint — browser connects here to receive reload events
     if (url === '/_pulse/reload') {
       res.writeHead(200, {

package/src/mcp/server.js

@@ -37,6 +37,20 @@ import { execFileSync, spawn, spawnSync } from 'child_process'
 
 import { loadPages } from '../cli/discover.js'
 
+// ---------------------------------------------------------------------------
+// Crash guards — an uncaught error in any tool handler must not kill the
+// stdio transport. Without these, Claude Code sees the process exit and
+// reports "pulse: disconnected" instead of a tool-level error. stderr only —
+// stdout is the JSON-RPC channel and any stray write corrupts the stream.
+// ---------------------------------------------------------------------------
+
+process.on('uncaughtException', (err) => {
+  console.error('[pulse-mcp] uncaught exception:', err?.stack || err)
+})
+process.on('unhandledRejection', (err) => {
+  console.error('[pulse-mcp] unhandled rejection:', err?.stack || err)
+})
+
 // ---------------------------------------------------------------------------
 // Project root
 // ---------------------------------------------------------------------------
@@ -49,6 +63,17 @@ const ROOT    = rootArg !== -1
 const PAGES_DIR      = path.join(ROOT, 'src', 'pages')
 const COMPONENTS_DIR = path.join(ROOT, 'src', 'components')
 
+// Detect whether ROOT is actually a Pulse project — used to warn the agent
+// early rather than silently misbehaving in unrelated repos.
+const IS_PULSE_PROJECT = (() => {
+  if (fs.existsSync(path.join(ROOT, 'pulse.config.js'))) return true
+  if (fs.existsSync(PAGES_DIR)) return true
+  try {
+    const pkg = JSON.parse(fs.readFileSync(path.join(ROOT, 'package.json'), 'utf8'))
+    return !!(pkg?.dependencies?.['@invisibleloop/pulse'] || pkg?.devDependencies?.['@invisibleloop/pulse'])
+  } catch { return false }
+})()
+
 const PKG_VERSION = JSON.parse(
   fs.readFileSync(new URL('../../package.json', import.meta.url).pathname, 'utf8')
 ).version
@@ -241,10 +266,10 @@ Fetch \`pulse://quickstart\` — workflow phases, spec skeleton, components, and
 Quick checklist:
 1. Read the file you're changing
 2. Make the edit
-3. \`pulse_validate\` → \`/verify\`
+3. \`pulse_validate\` → \`/verify --quick\` (mid-iteration) or \`/verify\` (final check)
 
 ### C — Targeted one-liner (label change, prop swap, CSS tweak)
-No resource fetch needed. Read the file, make the change, run \`/verify\`.
+No resource fetch needed. Read the file, make the change, run \`/verify --quick\` to check it looks right, then \`/verify\` when the user is happy.
 
 ### D — Stuck mid-build
 - Wrong component props? → \`pulse://guide/components\`
@@ -252,7 +277,7 @@ No resource fetch needed. Read the file, make the change, run \`/verify\`.
 - CSS / theming? → \`pulse://guide/styles\`
 - Routing or layout? → \`pulse://guide/routing\`
 - Lighthouse < 100? → Fix the failing audit, re-run \`/verify\`
-- Something looks wrong visually? → \`pulse_restart_server\` then navigate the browser
+- Something looks wrong visually? → \`pulse_restart_server\` (hot-reloads specs instantly) then navigate the browser
 
 ---
 
@@ -261,7 +286,7 @@ No resource fetch needed. Read the file, make the change, run \`/verify\`.
 - **\`meta.theme\` defaults to dark.** If the design is light, set \`theme: 'light'\` in the plan — not at screenshot time.
 - **Never set \`hydrate\`** — the framework sets it automatically from the URL entry.
 - **\`meta\` is always a plain object** — individual fields (\`title\`, \`description\`) can be async functions, but \`meta\` itself is \`{}\`.
-- **After every file edit:** \`pulse_restart_server\` → navigate browser → check result.
+- **After every file edit:** \`pulse_restart_server\` → navigate browser → check result. (Hot-reloads specs in ~200 ms — no process restart needed.)
 - **\`/verify\` is always last** — it writes the \`.pulse-verified\` stamp. Without it, the stop hook blocks.
 
 ---
@@ -318,9 +343,15 @@ server.registerResource(
 
 \`\`\`
 New page:   pulse_intake → pulse_sketch → pulse_intent → build → validate → screenshot → design approval → /verify
-Edit/fix:   read file → change → validate → /verify
+Edit/fix:   read file → change → validate → /verify --quick → (iterate) → /verify
 \`\`\`
 
+**\`/verify\` has two modes:**
+- \`/verify --quick\` — validate + screenshot + console + code review. No Lighthouse, no build. Use this during active iteration, between design rounds, or any time you're not yet at the "ready to ship" point. Fast (~10 s).
+- \`/verify\` — full loop: Lighthouse desktop + mobile + perf trace + all of the above. Use once the user has approved the design and you're done building. Slow (~90 s).
+
+**Rule: default to \`/verify --quick\` while building. Only run \`/verify\` (full) when the user signals they're done or happy with the result.**
+
 Tier 1 (static/read-only): phases 3 → 4 → 5a (design approval) → 5b (/verify)
 Tier 2 (interactive):       phases 3 → 4 → 5a → 5b → 6 (tests) → 7 (review)
 
@@ -372,6 +403,43 @@ export default {
 
 ---
 
+## Event binding — input patterns
+
+**The right pattern for each input type:**
+
+| Input type | Pattern | Why |
+|---|---|---|
+| Button click | \`<button data-event="increment">\` | click fires mutation |
+| Select / radio / checkbox | \`<select data-event="change:setFilter">\` | \`change\` fires on blur-with-changed-value — the correct commit semantic |
+| Color picker | \`<input type="color" data-event="change:setColor">\` | same as select |
+| Range slider | \`<input type="range" data-event="input:setVolume">\` | \`input\` fires on every drag step |
+| Text / email / password / search | **No \`data-event\`** — read via FormData | \`data-event\` on text inputs re-renders on every keystroke, destroying cursor position and focus |
+
+**Text input pattern (uncontrolled):**
+\`\`\`js
+// view — no data-event on the input
+view: (state) => \`
+  <form data-action="submit">
+    <input type="text" name="email" placeholder="Email">
+    <button type="submit">Save</button>
+  </form>
+\`,
+// action — read FormData in onStart (before validation)
+actions: {
+  submit: {
+    onStart:   (state, formData) => ({ status: 'loading', email: formData.get('email') }),
+    validate:  true,
+    run:       async (state) => { /* use state.email */ },
+    onSuccess: (state) => ({ status: 'success' }),
+    onError:   (state, err) => ({ status: 'error', error: err.message }),
+  }
+}
+\`\`\`
+
+**\`change:\` is the blur/commit event** — it fires when a non-text input loses focus with a changed value. This is correct for selects, color pickers, checkboxes, and radios. It does NOT fire on every keystroke (use \`input:\` for live-updating sliders or search fields where re-render on every character is intentional and the element is not a free-text input).
+
+---
+
 ## Key components (quick reference)
 
 Import from \`@invisibleloop/pulse/ui\`. Always include \`/pulse-ui.css\` in \`meta.styles\`.
@@ -457,6 +525,9 @@ server.registerTool(
     inputSchema: {},
   },
   async () => {
+    if (!IS_PULSE_PROJECT) {
+      return { content: [{ type: 'text', text: `⚠ This does not appear to be a Pulse project (no pulse.config.js, no src/pages/, no @invisibleloop/pulse dependency found in ${ROOT}).\n\nThe pulse MCP server is project-specific — it is started by the \`pulse\` CLI from inside a Pulse project directory. If you are working on a Pulse project, check that you launched with \`pulse\` from the correct directory.` }] }
+    }
     const specs      = await loadPages(ROOT)
     const components = findComponents()
     const lines      = []
@@ -528,6 +599,9 @@ server.registerTool(
     inputSchema: {},
   },
   async () => {
+    if (!IS_PULSE_PROJECT) {
+      return { content: [{ type: 'text', text: `⚠ This does not appear to be a Pulse project (no pulse.config.js, no src/pages/, no @invisibleloop/pulse dependency found in ${ROOT}).\n\nThe pulse MCP server is project-specific — it is started by the \`pulse\` CLI from inside a Pulse project directory. If you are working on a Pulse project, check that you launched with \`pulse\` from the correct directory.` }] }
+    }
     const lines = []
 
     // ── Pages ──────────────────────────────────────────────────────────────
@@ -1093,7 +1167,7 @@ server.registerTool(
 server.registerTool(
   'pulse_restart_server',
   {
-    description: 'Stop and restart the Pulse dev server. Use after adding new pages or making changes that require a restart.',
+    description: 'Reload specs in the running dev server. Tries a fast hot-reload first (no process restart); falls back to full kill/restart only if the server is not running.',
     inputSchema: {},
   },
   async () => {
@@ -1106,15 +1180,32 @@ server.registerTool(
       } catch { /* use default */ }
     }
 
-    // Kill any process on the port
+    // Fast path — POST to the running dev server's hot-reload endpoint.
+    // This reloads specs in-process and sends an SSE reload event to the browser
+    // without killing the server process (~200 ms vs ~3 s for a full restart).
+    const hotReloaded = await new Promise(resolve => {
+      try {
+        const req = http.request(
+          { hostname: '127.0.0.1', port, path: '/_pulse/trigger-reload', method: 'POST', timeout: 2000 },
+          res => resolve(res.statusCode === 204)
+        )
+        req.on('error', () => resolve(false))
+        req.on('timeout', () => { req.destroy(); resolve(false) })
+        req.end()
+      } catch { resolve(false) }
+    })
+
+    if (hotReloaded) {
+      return text(`Dev server specs reloaded (hot) on port ${port}`)
+    }
+
+    // Slow path — server not running or unresponsive, do a full kill/restart
     try { execFileSync('sh', ['-c', `lsof -ti:${port} | xargs kill -9 2>/dev/null; true`]) } catch { /* nothing running */ }
 
-    // Start fresh dev server detached so it outlives the MCP tool call
     const devScript = new URL('../cli/dev.js', import.meta.url).pathname
     const proc = spawn(process.execPath, [devScript, '--root', ROOT], { detached: true, stdio: 'ignore' })
     proc.unref()
 
-    // Wait until the server is actually accepting requests
     const ready = await waitForServer(port)
     return text(ready
       ? `Dev server restarted on port ${port}`
@@ -1202,6 +1293,7 @@ any feature build.
 
     let source = content
     if (file) {
+      if (!path.isAbsolute(file)) file = path.resolve(process.cwd(), file)
       if (!fs.existsSync(file)) return text(`File not found: ${file}`)
       source = fs.readFileSync(file, 'utf8')
     }
@@ -1232,9 +1324,16 @@ any feature build.
           issues.push('✗ React patterns found (className / htmlFor / onClick) — use class, for, data-event')
         if (/tabindex=["']?([1-9]\d*)/.test(renderedHtml))
           issues.push('✗ Positive tabindex found — remove, reorder DOM instead')
+        const rawNavInSource = (source.match(/<nav[\s>]/gi) || []).length
+        if (rawNavInSource > 0)
+          issues.push(`✗ Raw <nav> tag in spec source — nav() renders <nav> internally; wrapping it in another <nav> creates duplicate landmarks that fail Lighthouse accessibility. Remove the outer <nav>.`)
         const viewBlockStripped = renderedHtml
           .replace(/href=["']#[^"']*["']/g, 'href="#"')
           .replace(/id=["'][^"']*["']/g, 'id=""')
+          // Strip CSS custom property assignments from style attrs (--foo:#hex is a runtime
+          // data binding, not an authored colour — e.g. style="--swatch-bg:#E8524A").
+          // Standard property assignments like color:#000 are left intact and still flagged.
+          .replace(/style="([^"]*)"/g, (_, v) => `style="${v.replace(/--[a-zA-Z][^:]*:[^;"]*/g, '')}"`)
         if (/#[0-9a-fA-F]{6}\b|#[0-9a-fA-F]{3}\b(?![0-9a-fA-F])/.test(viewBlockStripped))
           warnings.push('⚠ Possible hex colour in rendered HTML — use var(--ui-*) tokens')
 
@@ -1247,19 +1346,49 @@ any feature build.
         }
       }
 
-      // Source-level checks (don't need rendered HTML)
-      if (source.includes('actions:') && !source.includes('onError:'))
+      // Source-level checks (don't need rendered HTML).
+      // Use line-anchored regex for actions: to avoid false positives from component
+      // prop strings like hero({ actions: '...' }) or card({ actions: '...' }).
+      const hasActionsBlock = /^\s{0,4}actions\s*:/m.test(source)
+      if (hasActionsBlock && !source.includes('onError:'))
         issues.push('✗ Action missing `onError` — required, will throw at runtime')
-      if (source.includes('actions:') && !source.includes('onSuccess:'))
+      if (hasActionsBlock && !source.includes('onSuccess:'))
         issues.push('✗ Action missing `onSuccess` — required')
       if (source.includes('hydrate:'))
         issues.push('✗ `hydrate` is set manually — remove it, the framework sets it automatically')
       if (/meta\s*:\s*async/.test(source))
         issues.push('✗ `meta` is an async function — `meta` must be a plain object; make individual fields async instead')
 
+      // Structural confirmations — always shown regardless of issues, so the agent
+      // gets positive signal on what's correct, not just a list of failures.
+      const confirms = []
+      if (renderedHtml) {
+        confirms.push(/<main[^>]*id=["']?main-content/.test(renderedHtml) ? '✓ `<main id="main-content">` landmark present' : null)
+        const rawNavInSource = (source.match(/<nav[\s>]/gi) || []).length
+        if (rawNavInSource === 0) confirms.push('✓ No raw <nav> in spec source — nav() component handles landmarks correctly')
+        const headings = [...renderedHtml.matchAll(/<h([1-6])[^>]*>/g)].map(m => parseInt(m[1]))
+        if (headings.length > 0) {
+          const orderOk = headings.every((h, i) => i === 0 || h <= headings[i - 1] + 1)
+          confirms.push(orderOk ? `✓ Heading order correct (${headings.map(h => `h${h}`).join(' → ')})` : `⚠ Heading order may be skipping levels (${headings.map(h => `h${h}`).join(' → ')})`)
+        }
+        const interactiveWithoutLabel = [...renderedHtml.matchAll(/<button(?![^>]*aria-label)[^>]*>\s*<\/button>/g)]
+        confirms.push(interactiveWithoutLabel.length === 0 ? '✓ No empty buttons without aria-label detected' : `⚠ ${interactiveWithoutLabel.length} button(s) appear empty — check aria-label`)
+        const dataEvents = [...renderedHtml.matchAll(/data-event=/g)].length
+        if (dataEvents > 0) confirms.push(`✓ ${dataEvents} data-event binding(s) found`)
+        const creativeOverride = /component.free|creative\s+override/i.test(source)
+        if (creativeOverride) confirms.push('✓ Creative override declared — component pattern checks are advisory')
+      }
+
       const lines = ['## Quick review\n']
+
+      if (confirms.filter(Boolean).length > 0) {
+        lines.push('### Structural checks\n')
+        for (const c of confirms.filter(Boolean)) lines.push(c)
+        lines.push('')
+      }
+
       if (issues.length === 0 && warnings.length === 0) {
-        lines.push('✓ No obvious issues found. Run `pulse_validate` next, then `/verify`.')
+        lines.push('✓ No issues found. Run `pulse_validate` next, then `/verify`.')
       } else {
         if (issues.length) {
           lines.push('### Fix before proceeding\n')
@@ -1271,7 +1400,7 @@ any feature build.
           for (const w of warnings) lines.push(w)
           lines.push('')
         }
-        lines.push('---\nFix issues, then run `pulse_validate` → `pulse_suggest` → `/verify`.')
+        lines.push('---\nFix issues, then run `pulse_validate` → `/verify --quick`.')
       }
 
       return text(lines.join('\n'))
@@ -1392,13 +1521,14 @@ ${(() => {
   }
   
   // Check spec source
-  // Hex colours in view — strip anchor hrefs and id attributes before checking so
-  // href="#fixtures" / id="main-content" / href="#join" are never flagged as hex colours.
-  // Only flag actual colour values: # followed by exactly 3 or 6 hex chars.
+  // Hex colours in view — strip anchor hrefs, id attributes, template literal expressions,
+  // and style attributes before checking so runtime-generated values (e.g. style="--swatch-bg:${hex}")
+  // are never flagged. Only flag literal hex values authored directly in the view string.
   const viewBlock = source.slice(source.indexOf('view:'))
   const viewBlockStripped = viewBlock
     .replace(/href=["']#[^"']*["']/g, 'href="#"')   // href="#anchor" → href="#"
     .replace(/id=["'][^"']*["']/g, 'id=""')          // id="foo" → id=""
+    .replace(/\$\{[^}]*\}/g, '${…}')                 // ${expr} — runtime values, not authored hex
   const hexInView = /#[0-9a-fA-F]{6}\b|#[0-9a-fA-F]{3}\b(?![0-9a-fA-F])/.test(viewBlockStripped)
   if (hexInView) {
     checks.push('⚠ **Possible hex colour in view** — use var(--ui-*) tokens only')
@@ -2674,7 +2804,9 @@ The extracted values map directly to pulse_intake fields (palette, vibe, styleNo
       lines.push(`Navigate to: **${source}**`)
       lines.push('Observe the rendered page at desktop width (~1440px) and again at mobile (~375px).')
       lines.push('Take note of the above-the-fold area, then scroll through the full page.')
-      lines.push('**Fill in the sections below based on what you observe. Return a completed extraction, not an empty template.**\n')
+      lines.push('**Fill in the sections below based on what you observe. Return a completed extraction, not an empty template.**')
+      lines.push('')
+      lines.push('> **If the URL is blocked or unreachable:** do not return a blank template. Instead, use your training knowledge of the site/brand to populate as many fields as you can, mark each cell with "(from knowledge, not live)" and note at the top that the URL was unreachable. A knowledge-based extraction is more useful than an empty template.\n')
     } else {
       lines.push('## Step 1 — Analyse the source (do this now)')
       lines.push(`Source: **${source}**`)
@@ -3973,7 +4105,7 @@ If you know your context already:
 - \`pulse_run_tests\` — run the project test suite (npm test). Use after writing or editing specs.
 - \`pulse_stamp\` — write the \`.pulse-verified\` stamp. Call as the **last step** of \`/verify\`, after Lighthouse and \`pulse_review\` both pass. Using this MCP tool avoids the mtime race that \`date +%s > .pulse-verified\` can cause when the stamp and the last spec write land in the same filesystem second.
 - \`pulse_fetch_page(url)\` — HTTP GET the dev server URL. Use to verify SSR output.
-- \`pulse_restart_server\` — stop and restart the dev server.
+- \`pulse_restart_server\` — hot-reload specs in the running dev server (~200 ms). Falls back to full kill/restart if the server is not running.
 - \`pulse_build\` — production build + starts prod server on devPort+1 for Lighthouse. Returns the URL. Call \`pulse_restart_server\` after to return to dev. **Slow — takes 30–60 s. Tell the user before calling.**
 - \`pulse_check_version\` — check installed package version, static asset version, and latest on npm. Use this instead of running npm commands when the user asks about updates.
 - \`pulse_update\` — install the latest \`@invisibleloop/pulse\` package and re-copy \`pulse-ui.css\`, \`pulse-ui.js\`, and the agent checklist into \`public/\`. One command does the full upgrade.
@@ -3989,9 +4121,13 @@ If you know your context already:
 
 ## MANDATORY: Verify every build
 
-**After writing or editing any page spec, you MUST run \`/verify\` before declaring done.**
+**After writing or editing any page spec, you MUST run \`/verify\` or \`/verify --quick\` before declaring done.**
+
+**Choose the right mode:**
+- \`/verify --quick\` — use during active building and iteration. Runs: validate → screenshot → console check → code review → stamp. No Lighthouse, no production build. Fast (~10 s). Use this between every round of changes.
+- \`/verify\` (full) — use when the user signals they're done or happy with the result. Runs everything: validate → screenshot → production build → Lighthouse desktop → Lighthouse mobile → performance trace → console check → code review → stamp. Slow (~90 s). Run this once at the end.
 
-\`/verify\` is the single canonical verification workflow. It runs: validate → screenshot → console check → network check → SSR check → production build → Lighthouse desktop → Lighthouse mobile → performance trace → code review → writes the \`.pulse-verified\` stamp. The stop hook checks this stamp — if it is missing or older than the last spec edit, the hook will block and make you run verification again.
+The stop hook checks the \`.pulse-verified\` stamp — \`/verify --quick\` writes the stamp too, so it satisfies the hook for mid-build turns. The full \`/verify\` is required before the final commit.
 
 **Do not replicate \`/verify\` steps manually.** Running the individual tools yourself does not write the stamp. The agent will be blocked at the end regardless.
 

package/src/mcp/validate-worker.js

@@ -11,8 +11,11 @@
 
 import { validateSpec } from '../spec/schema.js'
 
+import { existsSync } from 'fs'
+
 const tmpFile = process.argv[2]
 if (!tmpFile) { process.stdout.write('Invalid: no file path provided'); process.exit(0) }
+if (!existsSync(tmpFile)) { process.stdout.write(`Invalid: file not found — ${tmpFile}`); process.exit(0) }
 
 let mod
 try {

package/src/runtime/index.js

@@ -296,8 +296,15 @@ function bindEvents(el, dispatch, signal) {
     }
 
     // Dialog open — data-dialog-open="dialogId"
+    // If the element also carries data-event, fire the mutation first so the view
+    // re-renders with updated state before the dialog opens.
     const dialogOpenTarget = e.target?.closest?.('[data-dialog-open]')
     if (dialogOpenTarget) {
+      const eventAttr = dialogOpenTarget.dataset.event
+      if (eventAttr) {
+        const [type, name] = parseEventAttr(eventAttr)
+        if (type === 'click') { e.preventDefault(); dispatch(name, e) }
+      }
       const dialog = document.getElementById(dialogOpenTarget.dataset.dialogOpen)
       if (dialog?.showModal) { e.preventDefault(); dialog.showModal() }
       return
@@ -472,6 +479,8 @@ function morphAttrs(cur, nxt) {
     }
   }
   for (const { name } of Array.from(cur.attributes)) {
+    // `open` on <dialog> is managed by the native showModal/close API — never remove it
+    if (cur.tagName === 'DIALOG' && name === 'open') continue
     if (!nxt.hasAttribute(name)) cur.removeAttribute(name)
   }
 }

package/src/server/index.js

@@ -611,9 +611,25 @@ export async function createServer(entries, options = {}) {
   const brandCache = new TtlCache(500)
   brandCache.startEviction(30_000)  // brand TTL is 60s, scan every 30s
 
-  // Load manifest — maps source hydrate paths to production bundle paths
-  const hydrateMap     = loadManifest(manifest, staticDir)
-  const runtimeBundle  = hydrateMap['_runtime'] || ''
+  // Load manifest — maps source hydrate paths to production bundle paths.
+  // When staticDir is set and no manifest is explicitly provided, the manifest is
+  // re-read from disk on every request so that `pulse build` takes effect immediately
+  // without a server restart. The file is small (~1 kB) and the read is synchronous,
+  // so the overhead is negligible.
+  // When manifest is an explicit object (including {} from the dev server to disable
+  // manifest-based hydration), it is used as-is without disk reads. This ensures that
+  // passing manifest: {} always means "no manifest" — even when staticDir is set.
+  const _manifestStatic   = (manifest !== null && typeof manifest === 'object') ? manifest : null
+  const _manifestPath     = _manifestStatic ? null
+    : typeof manifest === 'string' ? manifest
+    : (manifest === null && staticDir) ? path.join(staticDir, 'dist', 'manifest.json') : null
+  const getHydrateMap = _manifestStatic
+    ? () => _manifestStatic
+    : () => {
+        if (!_manifestPath) return {}
+        try { return JSON.parse(fs.readFileSync(_manifestPath, 'utf8')) } catch { return {} }
+      }
+
 
   // Auto-detect favicon in staticDir
   const faviconPath = staticDir
@@ -905,6 +921,8 @@ export async function createServer(entries, options = {}) {
         ctx.store = await resolveStoreState(storeDef, ctx)
       }
 
+      const hydrateMap     = getHydrateMap()
+      const runtimeBundle  = hydrateMap['_runtime'] || ''
       const spec = resolveSpec(match.spec, hydrateMap)
 
       // Per-spec timeout overrides the global default
@@ -1691,33 +1709,6 @@ function mergeCtxHeaders(ctx, headers) {
 // Manifest & spec resolution
 // ---------------------------------------------------------------------------
 
-/**
- * Load a hydrate manifest.
- * Accepts a manifest object, a path to a JSON file, or auto-detects from
- * staticDir/dist/manifest.json when staticDir is set.
- *
- * @param {Object|string|null} manifest
- * @param {string|null} staticDir
- * @returns {Object} map of source paths → bundle paths
- */
-function loadManifest(manifest, staticDir) {
-  if (!manifest && !staticDir) return {}
-
-  if (manifest && typeof manifest === 'object') return manifest
-
-  const manifestPath = typeof manifest === 'string'
-    ? manifest
-    : staticDir ? path.join(staticDir, 'dist', 'manifest.json') : null
-
-  if (!manifestPath) return {}
-
-  try {
-    return JSON.parse(fs.readFileSync(manifestPath, 'utf8'))
-  } catch {
-    return {}
-  }
-}
-
 /**
  * Return a copy of the spec with the hydrate path and any meta.styles paths
  * resolved to their production bundle paths (if manifest entries exist).