diff --git a/pkgs/sdk_triage_bot/bin/triage.dart b/pkgs/sdk_triage_bot/bin/triage.dart
index 17968b58..ad46ff35 100644
--- a/pkgs/sdk_triage_bot/bin/triage.dart
+++ b/pkgs/sdk_triage_bot/bin/triage.dart
@@ -44,7 +44,7 @@ void main(List<String> arguments) async {
 
   var issue = results.rest.first;
   final dryRun = results.flag('dry-run');
-  final force = results.flag('force');
+  final forceTriage = results.flag('force');
 
   // Accept either an issue number or a url (i.e.,
   // https://github.com/dart-lang/sdk/issues/55816).
@@ -69,7 +69,7 @@ void main(List<String> arguments) async {
   await triage(
     int.parse(issue),
     dryRun: dryRun,
-    force: force,
+    forceTriage: forceTriage,
     githubService: githubService,
     geminiService: geminiService,
     logger: Logger(),
diff --git a/pkgs/sdk_triage_bot/lib/src/gemini.dart b/pkgs/sdk_triage_bot/lib/src/gemini.dart
index 7f2b5982..e87f0303 100644
--- a/pkgs/sdk_triage_bot/lib/src/gemini.dart
+++ b/pkgs/sdk_triage_bot/lib/src/gemini.dart
@@ -6,6 +6,10 @@ import 'package:google_generative_ai/google_generative_ai.dart';
 import 'package:http/http.dart' as http;
 
 class GeminiService {
+  // gemini-1.5-pro-latest, gemini-1.5-flash-latest, gemini-1.0-pro-latest
+  static const String classificationModel = 'models/gemini-1.5-flash-latest';
+  static const String summarizationModel = 'models/gemini-1.5-flash-latest';
+
   final GenerativeModel _summarizeModel;
   final GenerativeModel _classifyModel;
 
@@ -13,15 +17,15 @@ class GeminiService {
     required String apiKey,
     required http.Client httpClient,
   })  : _summarizeModel = GenerativeModel(
-          model: 'models/gemini-1.5-flash-latest',
+          model: summarizationModel,
           apiKey: apiKey,
           generationConfig: GenerationConfig(temperature: 0.2),
           httpClient: httpClient,
         ),
         _classifyModel = GenerativeModel(
-          // TODO(devconcarew): substitute our tuned model
+          // TODO(devoncarew): substitute our tuned model
           // model: 'tunedModels/autotune-sdk-triage-tuned-prompt-1l96e2n',
-          model: 'models/gemini-1.5-flash-latest',
+          model: classificationModel,
           apiKey: apiKey,
           generationConfig: GenerationConfig(temperature: 0.2),
           httpClient: httpClient,
@@ -45,6 +49,6 @@ class GeminiService {
 
   Future<String> _query(GenerativeModel model, String prompt) async {
     final response = await model.generateContent([Content.text(prompt)]);
-    return response.text!.trim();
+    return (response.text ?? '').trim();
   }
 }
diff --git a/pkgs/sdk_triage_bot/lib/src/github.dart b/pkgs/sdk_triage_bot/lib/src/github.dart
index a6a5e081..25bba1ee 100644
--- a/pkgs/sdk_triage_bot/lib/src/github.dart
+++ b/pkgs/sdk_triage_bot/lib/src/github.dart
@@ -43,10 +43,15 @@ class GithubService {
 
 Future<FetchIssuesResult> fetchIssues(
   String areaLabel, {
+  required bool includeClosed,
   String? cursor,
 }) async {
   final result = await _query(QueryOptions(
-    document: gql(_buildQueryString(areaLabel, cursor: cursor)),
+    document: gql(_buildQueryString(
+      areaLabel,
+      cursor: cursor,
+      includeClosed: includeClosed,
+    )),
     fetchPolicy: FetchPolicy.noCache,
     parserFn: (data) {
       final search = data['search'] as Map<String, dynamic>;
@@ -104,41 +109,46 @@ Future<QueryResult<T>> _query<T>(QueryOptions<T> options) {
   return _client.query<T>(options);
 }
 
-String _buildQueryString(String areaLabel, {String? cursor}) {
-  final cursorRef = cursor == null ? null : '"$cursor"';
+String _buildQueryString(
+  String areaLabel, {
+  required bool includeClosed,
+  String? cursor,
+}) {
+  final cursorTerm = cursor == null ? '' : 'after: "$cursor"';
+  final isOpen = includeClosed ? '' : 'is:open';
 
   return '''{
-  search(
-    query: "repo:dart-lang/sdk is:issue is:open label:$areaLabel"
-    type: ISSUE
-    first: 100,
-    after: $cursorRef
-  ) {
-    edges {
-      node {
-        ... on Issue {
-          title
-          number
-          state
-          bodyText
-          labels(first: 10) {
-            edges {
-              node {
-                name
+    search(
+      query: "repo:dart-lang/sdk is:issue $isOpen label:$areaLabel"
+      type: ISSUE
+      first: 100
+      $cursorTerm
+    ) {
+      edges {
+        node {
+          ... on Issue {
+            title
+            number
+            state
+            bodyText
+            labels(first: 10) {
+              edges {
+                node {
+                  name
+                }
               }
             }
           }
         }
       }
+      pageInfo {
+        endCursor
+        startCursor
+        hasNextPage
+        hasPreviousPage
+      }
     }
-    pageInfo {
-      endCursor
-      startCursor
-      hasNextPage
-      hasPreviousPage
-    }
-  }
-}''';
+  }''';
 }
 
 final GraphQLClient _client = _initGraphQLClient();
@@ -158,4 +168,17 @@ extension IssueExtension on Issue {
   ///
   /// Note that the original text for the issue is returned in the `body` field.
   bool get hasComments => commentsCount > 0;
+
+  /// Returns whether this issue has already been triaged.
+  ///
+  /// Generally, this means the the issue has had an `area-` label applied to
+  /// it, has had `needs-info` applied to it, or was closed.
+  bool get alreadyTriaged {
+    if (isClosed) return true;
+
+    return labels.any((label) {
+      final name = label.name;
+      return name == 'needs-info' || name.startsWith('area-');
+    });
+  }
 }
diff --git a/pkgs/sdk_triage_bot/lib/src/prompts.dart b/pkgs/sdk_triage_bot/lib/src/prompts.dart
index e363949a..0afa2517 100644
--- a/pkgs/sdk_triage_bot/lib/src/prompts.dart
+++ b/pkgs/sdk_triage_bot/lib/src/prompts.dart
@@ -28,10 +28,11 @@ area-infrastructure: Use area-infrastructure for SDK infrastructure issues, like
 area-intellij: Tracking issues for the Dart IntelliJ plugin.
 area-language: Dart language related items (some items might be better tracked at github.com/dart-lang/language).
 area-meta: Cross-cutting, high-level issues (for tracking many other implementation issues, ...).
+area-native-interop: Used for native interop related issues, including FFI.
 area-pkg: Used for miscellaneous pkg/ packages not associated with specific area- teams.
 area-sdk: Use area-sdk for general purpose SDK issues (packaging, distribution, …).
 area-test: Cross-cutting test issues (use area- labels for specific failures; not used for package:test).
-area-vm: Use area-vm for VM related issues, including code coverage, FFI, and the AOT and JIT backends.
+area-vm: Use area-vm for VM related issues, including code coverage, and the AOT and JIT backends.
 area-web: Use area-web for Dart web related issues, including the DDC and dart2js compilers and JS interop.
 
 Don't make up a new area.
diff --git a/pkgs/sdk_triage_bot/lib/triage.dart b/pkgs/sdk_triage_bot/lib/triage.dart
index 94f1fa9c..9529e98a 100644
--- a/pkgs/sdk_triage_bot/lib/triage.dart
+++ b/pkgs/sdk_triage_bot/lib/triage.dart
@@ -17,7 +17,7 @@ final sdkSlug = RepositorySlug('dart-lang', 'sdk');
 Future<void> triage(
   int issueNumber, {
   bool dryRun = false,
-  bool force = false,
+  bool forceTriage = false,
   required GithubService githubService,
   required GeminiService geminiService,
   required Logger logger,
@@ -63,21 +63,22 @@ ${trimmedBody(comment.body ?? '')}
   }
 
   // decide if we should triage
-  final alreadyTriaged = labels.any((l) => l.startsWith('area-'));
-  if (alreadyTriaged && !force) {
-    logger.log('Exiting (issue is already triaged).');
-    return;
+  if (!forceTriage) {
+    if (issue.alreadyTriaged) {
+      logger.log('Exiting (issue is already triaged).');
+      return;
+    }
   }
 
   // ask for the summary
   var bodyTrimmed = trimmedBody(issue.body);
   String summary;
   try {
-    // Failures here can include things like gemini safety issues, ...
     summary = await geminiService.summarize(
       summarizeIssuePrompt(title: issue.title, body: bodyTrimmed),
     );
   } on GenerativeAIException catch (e) {
+    // Failures here can include things like gemini safety issues, ...
     stderr.writeln('gemini: $e');
     exit(1);
   }
@@ -88,21 +89,21 @@ ${trimmedBody(comment.body ?? '')}
   logger.log('');
 
   // ask for the 'area-' classification
-  List<String> classification;
+  List<String> newLabels;
   try {
-    // Failures here can include things like gemini safety issues, ...
-    classification = await geminiService.classify(
+    newLabels = await geminiService.classify(
       assignAreaPrompt(
           title: issue.title, body: bodyTrimmed, lastComment: lastComment),
     );
   } on GenerativeAIException catch (e) {
+    // Failures here can include things like gemini safety issues, ...
     stderr.writeln('gemini: $e');
     exit(1);
   }
 
   logger.log('## gemini classification');
   logger.log('');
-  logger.log(classification.toString());
+  logger.log(newLabels.toString());
   logger.log('');
 
   if (dryRun) {
@@ -113,7 +114,7 @@ ${trimmedBody(comment.body ?? '')}
   // perform changes
   logger.log('## github comment');
   logger.log('');
-  logger.log('labels: $classification');
+  logger.log('labels: $newLabels');
   logger.log('');
   logger.log(summary);
 
@@ -122,17 +123,16 @@ ${trimmedBody(comment.body ?? '')}
   // create github comment
   await githubService.createComment(sdkSlug, issueNumber, comment);
 
-  final allLabels = await githubService.getAllLabels(sdkSlug);
-  var newLabels = filterExistingLabels(allLabels, classification);
-  if (newLabels.any((l) => l.startsWith('area-'))) {
-    newLabels.add('triage-automation');
+  final allRepoLabels = (await githubService.getAllLabels(sdkSlug)).toSet();
+  final labelAdditions = newLabels.toSet().union(allRepoLabels).toList()
+    ..sort();
+  if (labelAdditions.isNotEmpty) {
+    labelAdditions.add('triage-automation');
   }
-  // remove any duplicates
-  newLabels = newLabels.toSet().toList();
 
   // apply github labels
   if (newLabels.isNotEmpty) {
-    await githubService.addLabelsToIssue(sdkSlug, issueNumber, newLabels);
+    await githubService.addLabelsToIssue(sdkSlug, issueNumber, labelAdditions);
   }
 
   logger.log('');
diff --git a/pkgs/sdk_triage_bot/test/triage_test.dart b/pkgs/sdk_triage_bot/test/triage_test.dart
index 67a57017..c07f737f 100644
--- a/pkgs/sdk_triage_bot/test/triage_test.dart
+++ b/pkgs/sdk_triage_bot/test/triage_test.dart
@@ -63,7 +63,7 @@ void main() {
 
     await triage(
       mockIssueNumber,
-      force: true,
+      forceTriage: true,
       githubService: githubService,
       geminiService: geminiService,
       logger: TestLogger(),
diff --git a/pkgs/sdk_triage_bot/tool/bench.dart b/pkgs/sdk_triage_bot/tool/bench.dart
new file mode 100644
index 00000000..62d597ab
--- /dev/null
+++ b/pkgs/sdk_triage_bot/tool/bench.dart
@@ -0,0 +1,121 @@
+// Copyright (c) 2024, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+// This script benchmarks the issues listed in tool/bench.md using the current
+// issue triage prompt and writes the results back into bench.md.
+
+import 'dart:io';
+
+import 'package:github/github.dart';
+import 'package:google_generative_ai/google_generative_ai.dart';
+import 'package:http/http.dart' as http;
+import 'package:sdk_triage_bot/src/common.dart';
+import 'package:sdk_triage_bot/src/gemini.dart';
+import 'package:sdk_triage_bot/src/github.dart';
+import 'package:sdk_triage_bot/src/prompts.dart';
+
+final sdkSlug = RepositorySlug('dart-lang', 'sdk');
+
+void main(List<String> args) async {
+  print('Running benchmark against current prompt...');
+  print('');
+
+  final client = http.Client();
+
+  final github = GitHub(
+    auth: Authentication.withToken(githubToken),
+    client: client,
+  );
+  final githubService = GithubService(github: github);
+  final geminiService = GeminiService(
+    apiKey: geminiKey,
+    httpClient: client,
+  );
+
+  // read issues
+  final benchmarkFile = File('tool/bench.md');
+  final lines = benchmarkFile
+      .readAsLinesSync()
+      .where((l) => l.startsWith('| #'))
+      .toList();
+
+  final expectations = lines.map(ClassificationResults.parseFrom).toList();
+  var predicted = 0;
+
+  print('${expectations.length} issues read.');
+  print('');
+
+  for (var expectation in expectations) {
+    final issue =
+        await githubService.fetchIssue(sdkSlug, expectation.issueNumber);
+    final bodyTrimmed = trimmedBody(issue.body);
+
+    print('#${issue.number}');
+
+    try {
+      final labels = await geminiService.classify(
+        assignAreaPrompt(title: issue.title, body: bodyTrimmed),
+      );
+      if (expectation.satisfiedBy(labels)) {
+        predicted++;
+      } else {
+        var title = issue.title.length > 100
+            ? '${issue.title.substring(0, 100)}...'
+            : issue.title;
+        print('  "$title"');
+        print('     labeled: ${expectation.expectedLabels.join(', ')}');
+        print('  prediction: ${labels.join(', ')}');
+      }
+    } on GenerativeAIException catch (e) {
+      // Failures here can include things like gemini safety issues, ...
+      stderr.writeln('gemini: $e');
+    }
+  }
+
+  final result = predicted * 100.0 / expectations.length;
+  final today = DateTime.now().toIso8601String().substring(0, 10);
+  final percent = result.toStringAsFixed(1);
+  final model = GeminiService.classificationModel.split('/')[1];
+
+  benchmarkFile.writeAsStringSync(
+    '$today: $percent% using $model\n',
+    mode: FileMode.append,
+  );
+
+  print('');
+  print('$today: $percent% using $model');
+
+  client.close();
+}
+
+class ClassificationResults {
+  final int issueNumber;
+  final List<String> expectedLabels;
+
+  ClassificationResults({
+    required this.issueNumber,
+    required this.expectedLabels,
+  });
+
+  static ClassificationResults parseFrom(String line) {
+    // | #56366 | `area-dart2wasm`, `type-enhancement` |
+    final sections = line.split('|').skip(1).take(2).toList();
+    final number = sections[0].trim();
+    final labels = sections[1].trim();
+
+    return ClassificationResults(
+      issueNumber: int.parse(number.substring(1)),
+      expectedLabels: labels.split(',').map((label) {
+        label = label.trim();
+        return label.substring(1, label.length - 1);
+      }).toList(),
+    );
+  }
+
+  bool satisfiedBy(List<String> labels) {
+    final filtered = labels.where((l) => !l.startsWith('type-')).toSet();
+    final expected = expectedLabels.where((l) => !l.startsWith('type-'));
+    return expected.every(filtered.contains);
+  }
+}
diff --git a/pkgs/sdk_triage_bot/tool/bench.md b/pkgs/sdk_triage_bot/tool/bench.md
new file mode 100644
index 00000000..c5d189a5
--- /dev/null
+++ b/pkgs/sdk_triage_bot/tool/bench.md
@@ -0,0 +1,110 @@
+## What's this?
+
+This is a benchmark of existing, previously triaged issues. We use it to
+measure changes to our triage classification prompt - to see if changes
+materially improve the classification performance.
+
+## General issues
+
+| Issue | Expected |
+| --- | --- |
+| #56366 | `area-dart2wasm`, `type-enhancement`            |
+| #56365 | `area-core-library`, `type-enhancement`         |
+| #56364 | `area-front-end`, `type-bug`                    |
+| #56363 | `area-core-library`, `type-enhancement`         |
+| #56362 | `area-vm`, `type-bug`                           |
+| #56361 | `area-analyzer`                                 |
+| #56360 | `area-core-library`                             |
+| #56358 | `area-web`                                      |
+| #56355 | `area-analyzer`, `type-bug`, `type-performance` |
+| #56354 | `area-web`, `type-bug`                          |
+| #56353 | `area-dart2wasm`                                |
+| #56350 | `area-analyzer`, `type-enhancement`             |
+| #56348 | `area-intellij`                                 |
+| #56347 | `area-dart-cli`, `type-bug`                     |
+| #56346 | `area-pkg`, `pkg-json`, `type-enhancement`      |
+| #56345 | `area-analyzer`, `type-enhancement`             |
+| #56344 | `area-language`, `type-question`                |
+| #56343 | `area-analyzer`, `type-enhancement`             |
+| #56342 | `area-analyzer`, `type-enhancement`             |
+| #56341 | `area-analyzer`                                 |
+| #56340 | `area-vm`                                       |
+| #56339 | `area-core-library`                             |
+| #56337 | `area-language`, `type-enhancement`             |
+| #56330 | `area-native-interop`, `type-bug`               |
+| #56329 | `area-language`, `type-enhancement`             |
+| #56327 | `area-web`, `type-question`                     |
+| #56325 | `area-web`, `type-question`                     |
+| #56324 | `area-dart-cli`                                 |
+| #56323 | `area-core-library`, `type-question`            |
+| #56322 | `area-web`, `type-bug`                          |
+| #56321 | `area-dart2wasm`, `type-bug`                    |
+| #56319 | `area-vm`, `type-enhancement`                   |
+| #56317 | `area-core-library`, `type-enhancement`         |
+| #56316 | `area-web`                                      |
+| #56315 | `area-web`                                      |
+| #56314 | `area-web`, `type-bug`                          |
+| #56308 | `area-vm`                                       |
+| #56306 | `area-vm`, `type-bug`                           |
+| #56305 | `area-front-end`, `type-bug`, `type-question`   |
+| #56304 | `area-core-library`, `type-enhancement`         |
+| #56303 | `area-dart-cli`, `type-bug`                     |
+| #56302 | `area-vm`                                       |
+| #56297 | `area-language`, `type-enhancement`             |
+| #56284 | `area-core-library`, `type-enhancement`         |
+| #56283 | `area-dart2wasm`                                |
+| #56256 | `area-front-end`, `type-bug`                    |
+| #56254 | `area-pkg`, `pkg-vm-service`, `type-bug`        |
+| #56246 | `area-intellij`                                 |
+| #56240 | `area-intellij`                                 |
+| #56229 | `area-infrastructure`                           |
+| #56227 | `area-native-interop`                           |
+| #56220 | `area-infrastructure`, `type-code-health`       |
+| #56217 | `area-meta`                                     |
+| #56216 | `area-intellij`                                 |
+| #56214 | `area-native-interop`                           |
+| #56208 | `area-google3`, `type-enhancement`              |
+| #56207 | `area-google3`                                  |
+| #56206 | `area-google3`, `type-enhancement`              |
+| #56196 | `area-meta`, `type-enhancement`                 |
+| #56195 | `area-test`, `type-enhancement`                 |
+| #56193 | `area-front-end`, `type-enhancement`            |
+| #56192 | `area-test`, `type-enhancement`                 |
+| #56189 | `area-meta`, `type-enhancement`                 |
+| #56188 | `area-meta`, `type-enhancement`                 |
+| #56178 | `area-pkg`, `pkg-dtd`, `type-enhancement`       |
+| #56177 | `area-pkg`, `pkg-dtd`, `type-enhancement`       |
+| #56168 | `area-google3`                                  |
+| #56161 | `area-build`, `type-enhancement`                |
+| #56156 | `area-test`                                     |
+| #56154 | `area-front-end`, `type-enhancement`            |
+| #56153 | `area-test`, `type-enhancement`                 |
+| #56138 | `area-front-end`, `type-bug`                    |
+| #56122 | `area-front-end`, `area-language`               |
+| #56107 | `area-front-end`, `type-bug`                    |
+| #56081 | `area-sdk`, `type-enhancement`, `type-infra`    |
+| #56058 | `area-native-interop`                           |
+| #56008 | `area-infrastructure`                           |
+| #55990 | `area-infrastructure`                           |
+| #55727 | `area-google3`                                  |
+| #55669 | `area-build`                                    |
+| #55385 | `area-sdk`                                      |
+| #54306 | `area-build`                                    |
+| #53635 | `area-build`, `area-vm`                         |
+
+## Needs info issues
+
+We need more information from the user before we can triage these issues.
+
+| Issue | Expected | Notes |
+| --- | --- | --- |
+| #55291 | `needs-info` | unmodified 'create an issue' template |
+| #55144 | `needs-info` | |
+| #54595 | `needs-info` | |
+| #54134 | `needs-info` | unmodified 'create an issue' template |
+| #56348 | `needs-info` | 'Analyzer Feedback from IntelliJ' |
+| #56246 | `needs-info` | 'Analyzer Feedback from IntelliJ' |
+| #56349 | `needs-info` | Or, potentially `should-close` |
+
+## Results
+2024-08-27: 55.6% using gemini-1.5-flash-latest
diff --git a/pkgs/sdk_triage_bot/tool/create_tuning_data.dart b/pkgs/sdk_triage_bot/tool/create_tuning_data.dart
index 05ac6e95..8950eaee 100644
--- a/pkgs/sdk_triage_bot/tool/create_tuning_data.dart
+++ b/pkgs/sdk_triage_bot/tool/create_tuning_data.dart
@@ -17,25 +17,27 @@ import 'package:sdk_triage_bot/src/prompts.dart';
 //   - make sure we have at least 10 items from each area
 
 const Map<String, int> areaSampleCount = {
-  'area-vm': 100,
   'area-analyzer': 100,
-  'area-web': 100,
   'area-core-library': 100,
   'area-front-end': 100,
+  'area-vm': 100,
+  'area-web': 100,
   //
-  'area-language': 50,
+  'area-dart-cli': 50,
   'area-infrastructure': 50,
+  'area-language': 50,
   'area-test': 50,
-  'area-dart-cli': 50,
   //
-  'area-meta': 25,
   'area-dart2wasm': 25,
+  'area-meta': 25,
+  'area-pkg': 25,
   //
-  'area-sdk': 10,
-  'area-intellij': 10,
-  'area-tools': 10,
   'area-build': 10,
   'area-google3': 10,
+  'area-intellij': 10,
+  'area-native-interop': 10,
+  'area-sdk': 10,
+  'area-tools': 10,
 };
 
 void main(List<String> args) async {
@@ -84,8 +86,12 @@ void main(List<String> args) async {
   exit(0);
 }
 
-Future<List<Issue>> downloadIssues(String areaLabel, int count) async {
-  var result = await fetchIssues(areaLabel);
+Future<List<Issue>> downloadIssues(
+  String areaLabel,
+  int count, {
+  bool includeClosed = false,
+}) async {
+  var result = await fetchIssues(areaLabel, includeClosed: includeClosed);
 
   final issues = <Issue>[];
 
@@ -101,7 +107,11 @@ Future<List<Issue>> downloadIssues(String areaLabel, int count) async {
     if (!result.hasNext) {
       break;
     } else {
-      result = await fetchIssues(areaLabel, cursor: result.cursor);
+      result = await fetchIssues(
+        areaLabel,
+        includeClosed: includeClosed,
+        cursor: result.cursor,
+      );
     }
   }
 
@@ -150,6 +160,19 @@ extension on Issue {
 
     return '[$number] "$shortTitle": ${filteredLabels.join(', ')}';
   }
+
+  // ignore: unused_element
+  String get markdownDesc {
+    final filteredLabels = labels.map((l) => l.name).where((label) {
+      return label.startsWith('area-') ||
+          label.startsWith('type-') ||
+          label == 'needs-info';
+    }).toList()
+      ..sort();
+
+    final descriptions = filteredLabels.map((l) => '`$l`').toList();
+    return '| #$number | ${descriptions.join(', ')} |';
+  }
 }
 
 String csvEncode(String str) {