fix(scoring): don't treat "register" in prose as signup navigation

src/removals/scoring.ts

@@ -44,7 +44,6 @@ const navigationIndicators = [
 	'popular',
 	'privacy',
 	'recommended',
-	'register',
 	'related',
 	'responses',
 	'share',
@@ -69,6 +68,22 @@ const datePattern = /(?:(?:Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec)[a-z]
 // Author attribution pattern — case-sensitive "By" + capitalized name
 const bylinePattern = /\bBy\s+[A-Z]/;
 
+// Bare "register" is also a CPU or quantum register. Count it only as a signup
+// prompt that opens a block, or as the label of a link or button.
+const registrationPromptPattern = /^\s*(?:please\s+)?register\s+(?:now|today|here|to\s+continue|for\s+(?:free|an?\s+account)|(?:an?|your)\s+account)\b/;
+const registrationControlPattern = /^(?:please\s+)?register(?:$|\s+(?:for|to)\b)/;
+
+function isRegistrationPrompt(element: Element): boolean {
+	const text = (element.textContent || '').trim().toLowerCase();
+	if (!text.includes('register')) return false;
+	if (registrationPromptPattern.test(text)) return true;
+	return Array.from(element.querySelectorAll('a[href], button, [role="button"]')).some(control => {
+		const label = (control.textContent || '').trim().toLowerCase();
+		return (registrationControlPattern.test(label) || registrationPromptPattern.test(label)) &&
+			!control.querySelector('code');
+	});
+}
+
 // Pre-compiled navigation indicator regexes for scoreNonContentBlock
 const navigationIndicatorRegexes = navigationIndicators.map(
 	indicator => new RegExp(`\\b${indicator.replace(/\s+/g, '\\s+')}\\b`)
@@ -355,6 +370,7 @@ export class ContentScorer {
 			if (headingText && headingText === text.trim()) {
 				const headingLower = headingText.toLowerCase();
 				if (!navigationHeadingPattern.test(headingLower) &&
+					!isRegistrationPrompt(headingChild) &&
 					!tocHeadingPattern.test(headingLower)) {
 					return true;
 				}
@@ -369,7 +385,7 @@ export class ContentScorer {
 			let hasNavigationHeading = false;
 			for (let i = 0; i < headings.length; i++) {
 				const headingText = (headings[i].textContent || '').toLowerCase().trim();
-				if (navigationHeadingPattern.test(headingText)) {
+				if (navigationHeadingPattern.test(headingText) || isRegistrationPrompt(headings[i])) {
 					hasNavigationHeading = true;
 					break;
 				}
@@ -479,6 +495,9 @@ export class ContentScorer {
 				indicatorMatches++;
 			}
 		}
+		if (isRegistrationPrompt(element)) {
+			indicatorMatches++;
+		}
 		score -= indicatorMatches * 10;
 
 		// Check for high link density (navigation)

tests/expected/scoring--algorithm-register.md

@@ -0,0 +1,40 @@
+```json
+{
+  "title": "Register operations",
+  "author": "",
+  "site": "example.com",
+  "published": ""
+}
+```
+
+This procedure describes how a receiver processes an encoded state after transmission. The sender prepares a collection of physical systems and applies an encoding operation before delivering the result. The receiver uses a decoding operation to recover the original information. Each operation acts on a specified group of systems rather than on the entire collection. The names attached to these groups distinguish the input from the output and make the instructions precise. Intermediate states may contain errors introduced during transmission. The decoding procedure identifies those errors and corrects them without changing the information that the sender intended to convey. The ordering of the operations is therefore important.
+
+The steps below describe the final part of that procedure. Read each instruction in order and keep the named systems separate throughout the calculation. An operation written beside a system name applies only to that system. A dagger denotes the inverse operation in this example. After the inverse operations have been applied the receiver can produce the decoded state. The final instruction identifies which system contains that state. These conventions also allow the same procedure to be described in ordinary prose without equations. Both descriptions specify the same sequence of operations and the same final result.
+
+## Algorithm
+
+1. Decode the received state.
+	1. Apply $D$ and output register $A$.
+		2. Apply $U^\dagger P^\dagger$ to the received register $B$.
+
+### Quantum register operations
+
+Initialize the register before decoding.
+
+Use the `register` keyword.
+
+See the [`register`](https://example.com/reference/register) keyword.
+
+Check `register_state` before returning.
+
+Registered values remain unchanged.
+
+Use the cash register carefully.
+
+The register now contains output.
+
+Call `register` to continue.
+
+We register an account in this example.
+
+Inspect register [state](https://example.com/state) before returning.

tests/expected/scoring--registration-heading.md

@@ -0,0 +1,16 @@
+```json
+{
+  "title": "Decoding experiment",
+  "author": "",
+  "site": "example.com",
+  "published": ""
+}
+```
+
+This procedure describes how a receiver processes an encoded state after transmission. The sender prepares a collection of physical systems and applies an encoding operation before delivering the result. The receiver uses a decoding operation to recover the original information. Each operation acts on a specified group of systems rather than on the entire collection. The names attached to these groups distinguish the input from the output and make the instructions precise. Intermediate states may contain errors introduced during transmission. The decoding procedure identifies those errors and corrects them without changing the information that the sender intended to convey. The ordering of the operations is therefore important.
+
+The steps below describe the final part of that procedure. Read each instruction in order and keep the named systems separate throughout the calculation. An operation written beside a system name applies only to that system. A dagger denotes the inverse operation in this example. After the inverse operations have been applied the receiver can produce the decoded state. The final instruction identifies which system contains that state. These conventions also allow the same procedure to be described in ordinary prose without equations. Both descriptions specify the same sequence of operations and the same final result.
+
+The experiment compares the recovered information with the original input under several controlled conditions. Each trial uses the same preparation procedure so that differences in the observed result can be attributed to transmission errors rather than to changes in the initial state. Measurements are collected after decoding and evaluated using the same criteria throughout the experiment. The comparison also checks whether the order of the recovery operations affects the outcome. Repeating the trials with different error patterns establishes which aspects of the procedure remain stable. The interpretation of these measurements depends on keeping the preparation and recovery stages distinct and recording the final state consistently.
+
+This is the last substantive conclusion of our experiment, and it explains the results to our readers.

tests/fixtures/scoring--algorithm-register.html

@@ -0,0 +1,40 @@
+<!-- {"url":"https://example.com/algorithm"} -->
+<!DOCTYPE html>
+<html lang="en">
+<head><meta charset="UTF-8"><title>Register operations</title></head>
+<body>
+<article>
+	<h1>Register operations</h1>
+	<p>This procedure describes how a receiver processes an encoded state after transmission. The sender prepares a collection of physical systems and applies an encoding operation before delivering the result. The receiver uses a decoding operation to recover the original information. Each operation acts on a specified group of systems rather than on the entire collection. The names attached to these groups distinguish the input from the output and make the instructions precise. Intermediate states may contain errors introduced during transmission. The decoding procedure identifies those errors and corrects them without changing the information that the sender intended to convey. The ordering of the operations is therefore important.</p>
+	<p>The steps below describe the final part of that procedure. Read each instruction in order and keep the named systems separate throughout the calculation. An operation written beside a system name applies only to that system. A dagger denotes the inverse operation in this example. After the inverse operations have been applied the receiver can produce the decoded state. The final instruction identifies which system contains that state. These conventions also allow the same procedure to be described in ordinary prose without equations. Both descriptions specify the same sequence of operations and the same final result.</p>
+	<h2>Algorithm</h2>
+	<ol>
+		<li>Decode the received state.
+			<ol>
+				<li><div><p>Apply <math><semantics><mi>D</mi><annotation encoding="application/x-tex">D</annotation></semantics></math> and output register <math><semantics><mi>A</mi><annotation encoding="application/x-tex">A</annotation></semantics></math>.</p></div></li>
+				<li><div><p>Apply <math><semantics><mrow><msup><mi>U</mi><mo>†</mo></msup><msup><mi>P</mi><mo>†</mo></msup></mrow><annotation encoding="application/x-tex">U^\dagger P^\dagger</annotation></semantics></math> to the received register <math><semantics><mi>B</mi><annotation encoding="application/x-tex">B</annotation></semantics></math>.</p></div></li>
+			</ol>
+		</li>
+	</ol>
+	<div><h3>Quantum register operations</h3></div>
+	<div><p>Initialize the register before decoding.</p></div>
+	<div><p>Use the <code>register</code> keyword.</p></div>
+	<div><p>See the <a href="/reference/register"><code>register</code></a> keyword.</p></div>
+	<div><p>Check <code>register_state</code> before returning.</p></div>
+	<div><p>Registered values remain unchanged.</p></div>
+	<div><p>Use the cash register carefully.</p></div>
+	<div><p>The register now contains output.</p></div>
+	<div><p>Call <code>register</code> to continue.</p></div>
+	<div><p>We register an account in this example.</p></div>
+	<div><p>Inspect register <a href="/state">state</a> before returning.</p></div>
+	<div><p>Account access <a href="/register">Register</a></p></div>
+	<div><p>Register now for free access.</p></div>
+	<div><p>Register an account to proceed.</p></div>
+	<div><p>Please register to continue.</p></div>
+	<div><p>Account access <a href="/register">Register now</a></p></div>
+	<div><p>Sign up for updates.</p></div>
+	<div><p>Register now for <math><mi>X</mi></math> access.</p></div>
+	<nav><p>Account access <a href="/register">Register</a> <math><mi>Y</mi></math></p></nav>
+</article>
+</body>
+</html>

tests/fixtures/scoring--registration-heading.html

@@ -0,0 +1,12 @@
+<!-- {"url":"https://example.com/experiment"} -->
+<!DOCTYPE html>
+<html lang="en">
+<head><meta charset="UTF-8"><title>Decoding experiment</title></head>
+<body><article>
+<p>This procedure describes how a receiver processes an encoded state after transmission. The sender prepares a collection of physical systems and applies an encoding operation before delivering the result. The receiver uses a decoding operation to recover the original information. Each operation acts on a specified group of systems rather than on the entire collection. The names attached to these groups distinguish the input from the output and make the instructions precise. Intermediate states may contain errors introduced during transmission. The decoding procedure identifies those errors and corrects them without changing the information that the sender intended to convey. The ordering of the operations is therefore important.</p>
+<p>The steps below describe the final part of that procedure. Read each instruction in order and keep the named systems separate throughout the calculation. An operation written beside a system name applies only to that system. A dagger denotes the inverse operation in this example. After the inverse operations have been applied the receiver can produce the decoded state. The final instruction identifies which system contains that state. These conventions also allow the same procedure to be described in ordinary prose without equations. Both descriptions specify the same sequence of operations and the same final result.</p>
+<p>The experiment compares the recovered information with the original input under several controlled conditions. Each trial uses the same preparation procedure so that differences in the observed result can be attributed to transmission errors rather than to changes in the initial state. Measurements are collected after decoding and evaluated using the same criteria throughout the experiment. The comparison also checks whether the order of the recovery operations affects the outcome. Repeating the trials with different error patterns establishes which aspects of the procedure remain stable. The interpretation of these measurements depends on keeping the preparation and recovery stages distinct and recording the final state consistently.</p>
+<div><h3><a href="/register">Register for updates</a></h3></div>
+<p>This is the last substantive conclusion of our experiment, and it explains the results to our readers.</p>
+</article></body>
+</html>