@duckduckgo/autoconsent 16.48.0 → 16.50.0

package/scripts/benchmark-button-classification.ts

@@ -0,0 +1,75 @@
+/**
+ * Benchmark classifyButtonTextRegex against labelled button texts (exact label match, occurrence-weighted).
+ * New labelled rows are produced by the tracker-radar-collector post-processing scripts.
+ */
+import fs from 'fs';
+import path from 'path';
+import { parseArgs } from 'util';
+import { classifyButtonTextRegex } from '../lib/heuristics';
+import { buildLabelBenchmarks, classifyRows, formatExample, LabelBenchmark, parseButtonTextCsv, pct } from './button-text-benchmark';
+
+const DEFAULT_CSV_PATH = path.join(__dirname, '../data/labelled-button-texts.csv');
+const TOP_FAILURES = 25;
+const USAGE = `Usage: npm run benchmark-buttons -- [--input <path>] [--top <n>]
+
+Benchmark classifyButtonTextRegex against labelled button texts (exact label match, occurrence-weighted).
+
+  -i, --input <path>  path to labelled button text CSV (default: ${path.relative(process.cwd(), DEFAULT_CSV_PATH)})
+  --top <n>           number of failure examples to print per label (default: ${TOP_FAILURES})
+  -h, --help          show this help`;
+
+const { values } = parseArgs({
+    options: {
+        input: { type: 'string', short: 'i', default: DEFAULT_CSV_PATH },
+        top: { type: 'string', default: String(TOP_FAILURES) },
+        help: { type: 'boolean', short: 'h' },
+    },
+});
+if (values.help) {
+    console.log(USAGE);
+    process.exit(0);
+}
+const opts = { input: values.input, top: parseInt(values.top, 10) };
+
+function printLabelBenchmark(b: LabelBenchmark) {
+    console.log(`\n  ${b.label}`);
+    console.log(
+        `    correctly labelled: ${b.rowCorrect}/${b.rowSupport} rows (${pct(b.rowCorrect, b.rowSupport)}), ${b.weightedCorrect}/${b.weightedSupport} weighted (${pct(b.weightedCorrect, b.weightedSupport)})`,
+    );
+    console.log(
+        `    false positives (predicted ${b.label}, labelled otherwise): ${b.falsePositiveExamples.length} rows, ${b.weightedFalsePositives} weighted`,
+    );
+    for (const [title, examples] of [
+        ['top false positives', b.falsePositiveExamples],
+        ['top missed', b.missedExamples],
+    ] as const) {
+        console.log(`    ${title}:${examples.length === 0 ? ' (none)' : ''}`);
+        for (const example of examples.slice(0, opts.top)) {
+            console.log(`      ${formatExample(example)}`);
+        }
+    }
+}
+
+function main() {
+    const inputPath = path.resolve(opts.input);
+    const rows = parseButtonTextCsv(fs.readFileSync(inputPath, 'utf8'));
+    if (rows.length === 0) {
+        console.error('no labelled rows found in', inputPath);
+        process.exit(1);
+    }
+
+    const benchmarks = buildLabelBenchmarks(classifyRows(rows, classifyButtonTextRegex));
+    const total = (key: 'rowSupport' | 'weightedSupport' | 'rowCorrect' | 'weightedCorrect' | 'weightedFalsePositives') =>
+        benchmarks.reduce((sum, b) => sum + b[key], 0);
+
+    console.log(`Input: ${inputPath} (${rows.length} labelled rows)`);
+    console.log('\nclassifyButtonTextRegex benchmark (excluding other)');
+    console.log(
+        `  correctly labelled: ${total('rowCorrect')}/${total('rowSupport')} rows, ${total('weightedCorrect')}/${total('weightedSupport')} weighted (${pct(total('weightedCorrect'), total('weightedSupport'))})`,
+    );
+    console.log(`  false positives (weighted): ${total('weightedFalsePositives')}`);
+    console.log(`  missed (weighted): ${total('weightedSupport') - total('weightedCorrect')}`);
+    benchmarks.forEach(printLabelBenchmark);
+}
+
+main();

package/scripts/button-text-benchmark.ts

@@ -0,0 +1,102 @@
+/**
+ * Helpers for scoring the button classifier against data/labelled-button-texts.csv.
+ * Shared by the WTR accuracy test and scripts/benchmark-button-classification.ts, so must not use Node or DOM APIs.
+ */
+// browser ESM build, so the same import works in WTR and in Node
+import { parse } from 'csv-parse/browser/esm/sync';
+import { ButtonRegexClassification } from '../lib/types';
+
+const VALID_LABELS: readonly ButtonRegexClassification[] = ['settings', 'accept', 'reject', 'acknowledge', 'other'];
+
+export const BENCHMARK_LABELS = VALID_LABELS.filter((label) => label !== 'other');
+
+export type ButtonTextRow = {
+    buttonText: string;
+    occurences: number;
+    label: ButtonRegexClassification;
+};
+
+export type ClassifiedButtonTextRow = ButtonTextRow & {
+    predicted: ButtonRegexClassification;
+};
+
+export type LabelBenchmark = {
+    label: ButtonRegexClassification;
+    rowSupport: number;
+    weightedSupport: number;
+    rowCorrect: number;
+    weightedCorrect: number;
+    weightedFalsePositives: number;
+    falsePositiveExamples: ClassifiedButtonTextRow[];
+    missedExamples: ClassifiedButtonTextRow[];
+};
+
+/**
+ * Parse labelled-button-texts.csv (columns: button_text, occurences, label), keeping only rows with a valid label.
+ * Options match the reader in tracker-radar-collector, which also writes this file.
+ */
+export function parseButtonTextCsv(content: string): ButtonTextRow[] {
+    const records: Record<string, string>[] = parse(content, {
+        columns: true,
+        skip_empty_lines: true,
+        relax_column_count: true,
+        trim: true,
+    });
+
+    const rows: ButtonTextRow[] = [];
+    for (const record of records) {
+        const buttonText = record.button_text;
+        const occurences = Number.parseInt(record.occurences, 10);
+        const label = record.label?.toLowerCase() as ButtonRegexClassification;
+        if (!buttonText || Number.isNaN(occurences) || !VALID_LABELS.includes(label)) {
+            continue;
+        }
+        rows.push({ buttonText, occurences, label });
+    }
+    return rows;
+}
+
+export function classifyRows(
+    rows: ButtonTextRow[],
+    classifier: (buttonText: string) => ButtonRegexClassification,
+): ClassifiedButtonTextRow[] {
+    return rows.map((row) => ({ ...row, predicted: classifier(row.buttonText) }));
+}
+
+function sumOccurences(rows: ButtonTextRow[]): number {
+    return rows.reduce((sum, row) => sum + row.occurences, 0);
+}
+
+function byOccurence(a: ButtonTextRow, b: ButtonTextRow): number {
+    return b.occurences - a.occurences || a.buttonText.localeCompare(b.buttonText);
+}
+
+/**
+ * Per-label stats, in BENCHMARK_LABELS order. Rows labelled 'other' are excluded, matching the original collector benchmark.
+ */
+export function buildLabelBenchmarks(results: ClassifiedButtonTextRow[]): LabelBenchmark[] {
+    const benchmarkResults = results.filter((r) => r.label !== 'other');
+    return BENCHMARK_LABELS.map((label) => {
+        const support = benchmarkResults.filter((r) => r.label === label);
+        const correct = support.filter((r) => r.predicted === label);
+        const falsePositives = benchmarkResults.filter((r) => r.predicted === label && r.label !== label);
+        return {
+            label,
+            rowSupport: support.length,
+            weightedSupport: sumOccurences(support),
+            rowCorrect: correct.length,
+            weightedCorrect: sumOccurences(correct),
+            weightedFalsePositives: sumOccurences(falsePositives),
+            falsePositiveExamples: falsePositives.sort(byOccurence),
+            missedExamples: support.filter((r) => r.predicted !== label).sort(byOccurence),
+        };
+    });
+}
+
+export function pct(numerator: number, denominator: number): string {
+    return denominator === 0 ? 'n/a' : `${((numerator / denominator) * 100).toFixed(1)}%`;
+}
+
+export function formatExample(row: ClassifiedButtonTextRow): string {
+    return `[${row.label} -> ${row.predicted}] ${JSON.stringify(row.buttonText)} (x${row.occurences})`;
+}

package/tests/amazon-hiring.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('amazon-hiring', ['https://hiring.amazon.com/', 'https://hiring.amazon.ca/app']);

package/tests/avada-cookie-consent.spec.ts

@@ -0,0 +1,9 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('avada-cookie-consent', [
+    'https://www.nativeunion.com/products/belt-cable-usb-c-to-usb-c',
+    'https://www.marinbikes.com/',
+    'https://www.simplemodern.com/',
+    // Deny button opens the preferences popup
+    'https://www.outdoortechnology.com/',
+]);

package/tests/axeptio.spec.ts

@@ -1,9 +1,13 @@
 import generateCMPTests from '../playwright/runner';
 
 generateCMPTests('axeptio', [
-    'https://www.welcometothejungle.com/',
-    'https://www.dorcel.com/',
-    'https://mistral.ai/news/le-chat-dives-deep',
-    'https://lahalledesgourmets.com/pages/angelina',
+    'https://docs.mistral.ai/',
+    'https://www.prestashop-project.org/',
+    'https://www.spareka.fr/',
     'https://www.narbonneaccessoires.fr/fr-fr/',
 ]);
+
+// Step-based widget: "Close" rejects, "Accept all" only advances to the next step
+generateCMPTests('axeptio', ['https://mistral.ai/pricing', 'https://mistral.ai/news/le-chat-dives-deep'], {
+    testOptIn: false,
+});

package/tests/bigcrunch.spec.ts

@@ -0,0 +1,16 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests(
+    'bigcrunch',
+    [
+        'https://tvtropes.org/',
+        'https://www.mlbtraderumors.com/',
+        'https://www.alternet.org/',
+        'https://attackofthefanboy.com/',
+        'https://www.hoopsrumors.com/',
+        'https://crooksandliars.com/',
+    ],
+    {
+        skipRegions: ['US'],
+    },
+);

package/tests/borlabs.spec.ts

@@ -1,5 +1,9 @@
 import generateCMPTests from '../playwright/runner';
 
-generateCMPTests('borlabs', ['https://www.kesselheld.de/', 'https://cocoon-hotels.com/de/cocoon-salzburg/'], {
-    skipRegions: ['US'],
-});
+generateCMPTests(
+    'borlabs',
+    ['https://www.kesselheld.de/', 'https://cocoon-hotels.com/de/cocoon-salzburg/', 'https://www.headphonecheck.com/'],
+    {
+        skipRegions: ['US'],
+    },
+);

package/tests/bricklink.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('bricklink', ['https://www.bricklink.com/v2/main.page', 'https://store.bricklink.com/']);

package/tests/ceneo.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('ceneo', ['https://www.ceneo.pl/', 'https://www.ceneolokalnie.pl/']);

package/tests/conjointly.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('conjointly', ['https://conjointly.com/']);

package/tests/consentmo.spec.ts

@@ -8,6 +8,7 @@ generateCMPTests(
         'https://mgemi.com/products/the-felize-saddle-suede',
         'https://eu.blackdiamondequipment.com/',
         'https://www.directdoors.com/',
+        'https://goorin.com/',
     ],
     {
         skipRegions: ['US'],

package/tests/cookie-popup-close-cookies.spec.ts

@@ -0,0 +1,10 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('cookie-popup-close-cookies', [
+    'https://catflooringaccessories.com/product/fpb10x3/',
+    'https://www.coba.com/',
+    'https://www.copely.com/',
+    'https://www.movingedge.com/',
+    'https://www.arvitec.com/',
+    'https://www.cobaflooring.com/',
+]);

package/tests/cookieconsent2.spec.ts

@@ -1,3 +1,8 @@
 import generateCMPTests from '../playwright/runner';
 
-generateCMPTests('cookieconsent2', ['https://alpha-heidelberg.de/', 'https://antitec.fi/', 'https://servero.nl/onze-producten/']);
+generateCMPTests('cookieconsent2', [
+    'https://alpha-heidelberg.de/',
+    'https://antitec.fi/',
+    'https://servero.nl/onze-producten/',
+    'https://www.goaexperience.co.uk/guide/festivals-and-events',
+]);

package/tests/costco.spec.ts

@@ -0,0 +1,5 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('costco', ['https://www.costco.ca/', 'https://www.costcobusinesscentre.ca/'], {
+    testOptIn: false,
+});

package/tests/datasign.spec.ts

@@ -0,0 +1,8 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('datasign', [
+    'https://www.jfe-holdings.co.jp/en/',
+    'https://www.otafuku.co.jp/',
+    'https://www.billboard-japan.com/',
+    'https://www.dydo.co.jp/',
+]);

package/tests/dhbbank.spec.ts

@@ -0,0 +1,6 @@
+import generateCMPTests from '../playwright/runner';
+
+// rejecting reloads the page, which interrupts the self-test
+generateCMPTests('dhbbank', ['https://www.dhbbank.de/', 'https://www.dhbbank.nl/', 'https://www.dhbbank.com/', 'https://www.dhbbank.be/'], {
+    testSelfTest: false,
+});

package/tests/ee-consent-notice.spec.ts

@@ -0,0 +1,9 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('ee-consent-notice', [
+    'https://bryantpark.org/activities/category/winter-village',
+    'https://assuredguaranty.com/',
+    'https://www.janorth.org/',
+    'https://bellegrove.org/',
+    'https://artsmn.org/',
+]);

package/tests/ensembleartsphilly.spec.ts

@@ -0,0 +1,6 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('ensembleartsphilly', [
+    'https://www.ensembleartsphilly.org/tickets-and-events/ensemble-arts-philly-presents/gary-gulman',
+    'https://philorch.ensembleartsphilly.org/',
+]);

package/tests/evium.de.spec.ts

@@ -0,0 +1,5 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('evium.de', ['https://www.evium.de/'], {
+    testSelfTest: false,
+});

package/tests/fera-cookies.spec.ts

@@ -0,0 +1,6 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('fera-cookies', [
+    'https://planthealthportal.defra.gov.uk/trade/imports/imports-from-the-eu/bcpscps/faqs-for-animal-and-plant-health-agencies-apha-operational-hours/',
+    'https://www.cropmonitor.co.uk/',
+]);

package/tests/iubenda.spec.ts

@@ -3,3 +3,9 @@ import generateCMPTests from '../playwright/runner';
 generateCMPTests('iubenda', ['https://www.rossignol.com/us/', 'https://www.lofficielusa.com/', 'https://www.3bmeteo.com/'], {
     skipRegions: ['AU'],
 });
+
+// accept-only banner without reject or customize buttons
+generateCMPTests('iubenda', ['https://www.yogurtland.com/', 'https://www.greenplanet.net/'], {
+    skipRegions: ['AU'],
+    testSelfTest: false,
+});

package/tests/la7-cmp.spec.ts

@@ -0,0 +1,6 @@
+import generateCMPTests from '../playwright/runner';
+
+// the site reloads the page after opt-out, so the self-test cannot run in the original context
+generateCMPTests('la7-cmp', ['https://www.la7.it/', 'https://tg.la7.it/', 'https://www.sedanoallegro.it/', 'https://tv.torinofc.it/'], {
+    testSelfTest: false,
+});

package/tests/laravel-cookie-consent.spec.ts

@@ -0,0 +1,10 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('laravel-cookie-consent', [
+    'https://podstatus.com/',
+    'https://www.arzneimittel-datenbank.de/',
+    'https://www.opendag.no/',
+    'https://teatenerife.es/',
+    'https://radacini.ro/',
+    'https://30cc.be/',
+]);

package/tests/legislation-gov-uk.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('legislation-gov-uk', ['https://www.legislation.gov.uk/']);

package/tests/mackeeper.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('mackeeper', ['https://mackeeper.com/blog/all-you-need-to-know-about-mac-power-management/', 'https://mackeeper.com/']);

package/tests/ndl-go-jp.spec.ts

@@ -0,0 +1,8 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('ndl-go-jp', [
+    'https://ndlsearch.ndl.go.jp/imagebank',
+    'https://www.ndl.go.jp/',
+    'https://www.ndl.go.jp/en/',
+    'https://dl.ndl.go.jp/',
+]);

package/tests/pandectes.spec.ts

@@ -3,6 +3,25 @@ import generateCMPTests from '../playwright/runner';
 generateCMPTests('pandectes', [
     'https://hollandscountryclothing.co.uk/',
     'https://cluse.com/',
-    'https://www.parent.com/',
     'https://swedishstockings.com/', // full-page overlay ("blocking") banner variant
 ]);
+
+// shadow DOM <pandectes-cmp> banner variant
+generateCMPTests(
+    'pandectes',
+    [
+        'https://enron.com/',
+        'https://www.parent.com/',
+        'https://wetnwildbeauty.com/',
+        'https://www.yakima.com/',
+        'https://us.select-sport.com/collections/soccer-balls-training-series',
+    ],
+    {
+        testOptIn: false,
+    },
+);
+
+// shadow DOM variant without a decline button: opt out through preferences
+generateCMPTests('pandectes', ['https://asmc.de/', 'https://crossrope.com/'], {
+    testOptIn: false,
+});

package/tests/passle.spec.ts

@@ -0,0 +1,8 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('passle', [
+    'https://technologylaw.fkks.com/post/102jge9/california-requires-browsers-and-mobile-devices-to-support-opt-out-preference-sig',
+    'https://www.insurtechworld.org/',
+    'https://news.fintech.io/',
+    'https://www.bbeb.com/',
+]);

package/tests/paypal-us.spec.ts

@@ -2,6 +2,5 @@ import generateCMPTests from '../playwright/runner';
 
 generateCMPTests('paypal-us', ['https://www.paypal.com/us/home'], {});
 
-// This banner variant has no "Decline" button, so opting out goes through the cookie
-// preferences page, which navigates away and back, running the rule more than once.
-generateCMPTests('paypal-us', ['https://www.paypal.com/ca/home'], { expectedRuns: 3 });
+// This banner variant has no "Decline" button, so the rule hides it and no choice is stored.
+generateCMPTests('paypal-us', ['https://www.paypal.com/ca/home'], { testSelfTest: false });

package/tests/primeconsent.spec.ts

@@ -0,0 +1,11 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('primeconsent', [
+    'https://www.jclewislincoln.com/2025-lincoln-aviator-dimensions',
+    'https://jclewisford.net/',
+    'https://www.experiencepoint.com/',
+    'https://www.valenciaautocenter.com/',
+    'https://www.upl-ltd.com/',
+    'https://www.cornhuskernissan.com/',
+    'https://privacypillar.com/',
+]);

package/tests/shinden.pl.spec.ts

@@ -0,0 +1,5 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('shinden.pl', ['https://shinden.pl/episode/61872-tensei-shitara-slime-datta-ken-3rd-season/view/231781'], {
+    testSelfTest: false,
+});

package/tests/texas-instruments.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('texas-instruments', ['https://education.ti.com/', 'https://www.t3europe.eu/', 'https://www.t3deutschland.de/']);

package/tests/thegrandyork.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('thegrandyork', ['https://www.thegrandyork.co.uk/']);

package/tests/thenounproject.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('thenounproject', ['https://thenounproject.com/']);

package/tests/tri-table.spec.ts

@@ -0,0 +1,5 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('tri-table', ['https://ckm.pl/', 'https://tri-table.com/'], {
+    skipRegions: ['US'],
+});

package/tests/truevault-polaris.spec.ts

@@ -0,0 +1,9 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('truevault-polaris', [
+    'https://www.mcgeeandco.com/products/calvert-brass-tissue-box-cover',
+    'https://www.studio-mcgee.com/',
+    'https://www.thorne.com/',
+    'https://www.crossfit.com/',
+    'https://www.philzcoffee.com/',
+]);

package/tests/uft-cookie-notice.spec.ts

@@ -0,0 +1,3 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests('uft-cookie-notice', ['https://fribourg.ch/en/', 'https://www.la-gruyere.ch/', 'https://www.schwarzsee.ch/']);

package/tests/wbd-legal-terms.spec.ts

@@ -0,0 +1,18 @@
+import generateCMPTests from '../playwright/runner';
+
+generateCMPTests(
+    'wbd-legal-terms',
+    [
+        'https://www.foodnetwork.com/how-to/packages/shopping/product-reviews/best-food-processors',
+        'https://www.hgtv.com/',
+        'https://www.golfdigest.com/',
+        'https://www.tlc.com/',
+        'https://www.discovery.com/',
+    ],
+    {
+        // shown only in some US states
+        onlyRegions: ['US'],
+        testOptIn: false,
+        testSelfTest: false,
+    },
+);

package/tests/webflow.spec.ts

@@ -1,3 +1,9 @@
 import generateCMPTests from '../playwright/runner';
 
-generateCMPTests('webflow', ['https://olark.com/', 'https://www.nexaas.com/', 'https://www.rupahealth.com/', 'https://www.hk.co/']);
+generateCMPTests('webflow', [
+    'https://www.ctgt.ai/research/distillation-censorship-transfer',
+    'https://www.dock.io/',
+    'https://www.talkabroad.com/',
+    'https://www.pocoyo.com/',
+    'https://www.storypark.com/',
+]);

package/tests/wix.spec.ts

@@ -1,3 +1,10 @@
 import generateCMPTests from '../playwright/runner';
 
-generateCMPTests('wix', ['https://www.wix.com/']);
+generateCMPTests('wix', [
+    'https://www.wix.com/',
+    'https://www.poconopoolandspallc.com/',
+    'https://www.karlswarehousing.com/',
+    'https://www.stormkingpw.com/',
+    'https://www.iconicoffices.com/',
+    'https://www.steigertools.nl/',
+]);

package/tests-wtr/heuristics/button-classification-accuracy.test.ts

@@ -0,0 +1,41 @@
+import { expect } from '@esm-bundle/chai';
+import { classifyButtonTextRegex } from '../../lib/heuristics';
+import {
+    BENCHMARK_LABELS,
+    buildLabelBenchmarks,
+    classifyRows,
+    formatExample,
+    LabelBenchmark,
+    parseButtonTextCsv,
+    pct,
+} from '../../scripts/button-text-benchmark';
+
+// Minimum occurrence-weighted accuracy required per label.
+const MIN_WEIGHTED_ACCURACY = 0.9;
+
+describe('classifyButtonTextRegex against labelled button texts', () => {
+    let benchmarks: LabelBenchmark[];
+
+    before(async () => {
+        const response = await fetch('/data/labelled-button-texts.csv');
+        const rows = parseButtonTextCsv(await response.text());
+        expect(rows.length).to.be.greaterThan(0, 'expected labelled rows in the CSV');
+        benchmarks = buildLabelBenchmarks(classifyRows(rows, classifyButtonTextRegex));
+    });
+
+    BENCHMARK_LABELS.forEach((label, i) => {
+        it(`never predicts "${label}" for a button labelled differently`, () => {
+            const examples = benchmarks[i].falsePositiveExamples.slice(0, 10).map(formatExample).join('\n      ');
+            expect(benchmarks[i].falsePositiveExamples.length).to.equal(0, `false positives for "${label}":\n      ${examples}\n`);
+        });
+
+        it(`classifies >${MIN_WEIGHTED_ACCURACY * 100}% of weighted "${label}" buttons correctly`, () => {
+            const { weightedCorrect, weightedSupport } = benchmarks[i];
+            expect(weightedSupport).to.be.greaterThan(0, `expected labelled rows for "${label}"`);
+            expect(weightedCorrect / weightedSupport).to.be.greaterThan(
+                MIN_WEIGHTED_ACCURACY,
+                `weighted accuracy for "${label}" was ${pct(weightedCorrect, weightedSupport)} (${weightedCorrect}/${weightedSupport})`,
+            );
+        });
+    });
+});