Skip to content

The ultimate harness for coding agents and beyond. All your agents. All your machines. One command center. Start with code, then follow your curiosity and build across disciplines: CAD, circuits, robots, games and music.

Leaderboard Models: each model's share of the week's tasks, with engine icons (#2203)

0xken15b1aa3a5c729b
diff --git a/backend/src/superTabs/leaderboard/contract.ts b/backend/src/superTabs/leaderboard/contract.ts
index f02daebb2..a428619fa 100644
--- a/backend/src/superTabs/leaderboard/contract.ts
+++ b/backend/src/superTabs/leaderboard/contract.ts
@@ -69,6 +69,8 @@ export interface ModelScore {
/** SHIPPED_POINTS per request it shipped this week, REVIEW_POINTS per review verdict it gave. */
points: number
rank: number
+ /** This week's attempts on it at requests, never reviews: its share of the week's tasks (Dee, 2026-10-29). */
+ attempts: number
/** Requests its attempt shipped this week; median times to ship from that attempt's assignment, and from the request's first. */
shipped: { count: number; medianAttemptMs: number | null; medianRequestMs: number | null }
/** This week's attempts on it with a first review on record, and how many passed it. */
diff --git a/backend/src/superTabs/leaderboard/stats.test.ts b/backend/src/superTabs/leaderboard/stats.test.ts
index 50504d398..087bda338 100644
--- a/backend/src/superTabs/leaderboard/stats.test.ts
+++ b/backend/src/superTabs/leaderboard/stats.test.ts
@@ -313,6 +313,18 @@ describe('Models: temporary workers by engine, model and effort, never by agent
expect(model(view, 'opus').firstReview).toEqual({ reviewed: 3, passed: 2 })
})
+ it('counts each model\'s attempts of the week, reviewed or not, never a review group\'s or another week\'s', () => {
+ const view = board(swarm(), [
+ group({ status: 'working', ledger: { runs: [run('a1', WEEK, 'opus', 'high', { firstReview: true }), run('a2', WEEK + HOUR),
+ run('a3', LAST), run('a4', WEEK + 2 * HOUR, 'gpt-6.1', 'medium')] } }),
+ group({ status: 'done', ledger: { runs: [run('b1', WEEK + 3 * HOUR)] } }),
+ // A review group's attempts are reviews, not tasks.
+ group({ status: 'done', reviewOf: 'g0', ledger: { runs: [run('c1', WEEK + HOUR, 'gpt-6.1', 'medium')] } }),
+ ])
+ expect(model(view, 'opus').attempts).toBe(3)
+ expect(model(view, 'gpt-6.1', 'medium').attempts).toBe(1)
+ })
+
it('gives a reviewer model 1 point per finished review verdict of the week', () => {
const counted = (week: number, count: number) => ({ week, engine: 'codex', model: 'gpt-5.5', effort: 'medium', count })
const view = board(swarm(), [group({ status: 'review', ledger: { reviews: [counted(LAST, 1), counted(WEEK, 2)] } }),
diff --git a/backend/src/superTabs/leaderboard/stats.ts b/backend/src/superTabs/leaderboard/stats.ts
index d7c6326f3..54adebb4f 100644
--- a/backend/src/superTabs/leaderboard/stats.ts
+++ b/backend/src/superTabs/leaderboard/stats.ts
@@ -134,7 +134,7 @@ function agentScores(state: SuperTabState, projectId: string, shippedThisWeek: C
return scored.map(entry => ({ ...entry, rank: 1 + scored.filter(other => other.points > entry.points).length }))
}
-/** Every model that worked this week: the requests its attempts shipped, the first reviews of its attempts, and the review
+/** Every model that worked this week: its attempts, the requests they shipped, their first reviews, and the review
* verdicts it gave. A request ships on its last attempt before it shipped. */
function modelScores(requests: Conversation[], shippedThisWeek: Conversation[], inWeek: (at: number) => boolean): ModelScore[] {
const rows = new Map<string, ModelScore & { attemptMs: number[]; requestMs: number[] }>()
@@ -142,7 +142,7 @@ function modelScores(requests: Conversation[], shippedThisWeek: Conversation[],
const key = JSON.stringify([on.engine, on.model, on.effort ?? null])
let entry = rows.get(key)
if (!entry) rows.set(key, entry = { engine: on.engine, model: on.model, effort: on.effort ?? null, grid: !!on.grid, points: 0, rank: 0,
- shipped: { count: 0, medianAttemptMs: null, medianRequestMs: null }, firstReview: { reviewed: 0, passed: 0 }, reviews: 0,
+ attempts: 0, shipped: { count: 0, medianAttemptMs: null, medianRequestMs: null }, firstReview: { reviewed: 0, passed: 0 }, reviews: 0,
attemptMs: [], requestMs: [] })
return entry
}
@@ -162,6 +162,7 @@ function modelScores(requests: Conversation[], shippedThisWeek: Conversation[],
for (const run of group.ledger?.runs ?? []) {
if (!inWeek(run.at)) continue
const entry = row(run)
+ entry.attempts++
if (run.firstReview === undefined) continue
entry.firstReview.reviewed++
if (run.firstReview) entry.firstReview.passed++
diff --git a/desktop/lib/super_tabs/leaderboard/model.dart b/desktop/lib/super_tabs/leaderboard/model.dart
index cdd6dea97..84211e84a 100644
--- a/desktop/lib/super_tabs/leaderboard/model.dart
+++ b/desktop/lib/super_tabs/leaderboard/model.dart
@@ -79,6 +79,7 @@ class SuperTabModelScore {
grid = json['grid'] == true,
points = _int(json['points']),
rank = _int(json['rank']),
+ attempts = _int(json['attempts']),
shipped = _int(_map(json['shipped'])['count']),
medianAttemptMs = _optional(_map(json['shipped'])['medianAttemptMs']),
medianRequestMs = _optional(_map(json['shipped'])['medianRequestMs']),
@@ -96,6 +97,9 @@ class SuperTabModelScore {
final bool grid;
final int points, rank, shipped;
+ /// This week's tasks it ran, never reviews; 0 from a server before them.
+ final int attempts;
+
/// Median time to ship from the shipping attempt's assignment, and from the
/// request's first assignment.
final int? medianAttemptMs, medianRequestMs;
diff --git a/desktop/lib/super_tabs/leaderboard/models_section.dart b/desktop/lib/super_tabs/leaderboard/models_section.dart
index 69e70532d..9ea40c930 100644
--- a/desktop/lib/super_tabs/leaderboard/models_section.dart
+++ b/desktop/lib/super_tabs/leaderboard/models_section.dart
@@ -1,6 +1,7 @@
import 'package:flutter/material.dart';
import '../../core/runtime_model_name.dart';
+import '../../shared/theme/app_icons.dart';
import '../../shared/theme/app_theme.dart' as grid;
import '../../widgets/desktop_chrome.dart';
import '../../widgets/engine_identity.dart';
@@ -14,33 +15,60 @@ import 'reader.dart';
const _gap = DesktopChrome.controlGap;
const _rowHeight = 36.0;
-/// Fewer first reviews than this say nothing about a model.
-const leaderboardModelMinReviewed = 3;
-
/// How many models the sidebar lists until its heading shows them all.
const leaderboardModelsShown = 3;
/// A model as the sidebar's Models lists it: every engine and effort it ran
-/// at as one, and how many of its attempts reviewers approved on first review.
+/// at as one, its share of the week's tasks, and how many of its tasks
+/// reviewers approved on first review.
class LeaderboardModelShare {
const LeaderboardModelShare({
required this.name,
+ required this.engine,
+ required this.attempts,
+ required this.total,
+ required this.percent,
required this.reviewed,
required this.passed,
});
/// "Opus 5.5", "GPT-6.1 Sol"; the engine's name when it reported no model.
final String name;
+
+ /// The engine that ran most of its tasks: its icon leads the row. Null for
+ /// [LeaderboardModelShare.other], which no one engine ran.
+ final String? engine;
+
+ /// Its tasks, of the [total] every model ran.
+ final int attempts, total;
+
+ /// Its share of [total], rounded so the week's rows add up to 100.
+ final int percent;
final int reviewed, passed;
- int get percent => (passed * 100 / reviewed).round();
+ /// The models behind the heading as one row, so the rows shown add up to
+ /// 100 too: its percent is the sum of theirs, never a rounding of its own.
+ factory LeaderboardModelShare.other(List<LeaderboardModelShare> rest) =>
+ LeaderboardModelShare(
+ name: 'Other',
+ engine: null,
+ attempts: rest.fold(0, (sum, share) => sum + share.attempts),
+ total: rest.first.total,
+ percent: rest.fold(0, (sum, share) => sum + share.percent),
+ reviewed: rest.fold(0, (sum, share) => sum + share.reviewed),
+ passed: rest.fold(0, (sum, share) => sum + share.passed),
+ );
}
-/// The week's models with at least [leaderboardModelMinReviewed] first
-/// reviews, the most often approved first; of two alike, the one reviewed
-/// more, then by name.
+/// The week's models that ran tasks, the largest share first, then by name.
+/// Shares round by largest remainder: the floors, then one more point each
+/// to the largest remainders until they add up to 100.
List<LeaderboardModelShare> leaderboardModelShares(SuperTabLeaderboard board) {
- final merged = <String, ({String name, int reviewed, int passed})>{};
+ final merged =
+ <
+ String,
+ ({Map<String, int> engines, int attempts, int reviewed, int passed})
+ >{};
for (final score in board.byModel) {
final name = switch (score.model) {
final String model => modelDisplayName(model),
@@ -49,35 +77,60 @@ List<LeaderboardModelShare> leaderboardModelShares(SuperTabLeaderboard board) {
// One row per model, whichever engine ran it at whichever effort:
// rows are keyed by name.
final before = merged[name];
+ final engines = before?.engines ?? <String, int>{};
+ engines[score.engine] = (engines[score.engine] ?? 0) + score.attempts;
merged[name] = (
- name: name,
+ engines: engines,
+ attempts: (before?.attempts ?? 0) + score.attempts,
reviewed: (before?.reviewed ?? 0) + score.reviewed,
passed: (before?.passed ?? 0) + score.passed,
);
}
- final shares = [
- for (final model in merged.values)
- if (model.reviewed >= leaderboardModelMinReviewed)
- LeaderboardModelShare(
- name: model.name,
- reviewed: model.reviewed,
- passed: model.passed,
- ),
+ final ran =
+ [
+ for (final MapEntry(key: name, :value) in merged.entries)
+ if (value.attempts > 0) (name: name, model: value),
+ ]..sort((a, b) {
+ final share = b.model.attempts - a.model.attempts;
+ return share != 0 ? share : a.name.compareTo(b.name);
+ });
+ final total = ran.fold<int>(0, (sum, row) => sum + row.model.attempts);
+ final percents = [for (final row in ran) row.model.attempts * 100 ~/ total];
+ final short = 100 - percents.fold<int>(0, (sum, p) => sum + p);
+ // Of equal remainders, the one listed first.
+ int remainder(int i) => ran[i].model.attempts * 100 % total;
+ final byRemainder = [for (var i = 0; i < ran.length; i++) i]
+ ..sort((a, b) {
+ final larger = remainder(b) - remainder(a);
+ return larger != 0 ? larger : a - b;
+ });
+ for (final i in byRemainder.take(short)) {
+ percents[i]++;
+ }
+ return [
+ for (var i = 0; i < ran.length; i++)
+ LeaderboardModelShare(
+ name: ran[i].name,
+ // Of two engines alike, the one listed first.
+ engine: ran[i].model.engines.entries
+ .reduce((a, b) => b.value > a.value ? b : a)
+ .key,
+ attempts: ran[i].model.attempts,
+ total: total,
+ percent: percents[i],
+ reviewed: ran[i].model.reviewed,
+ passed: ran[i].model.passed,
+ ),
];
- shares.sort((a, b) {
- final share = b.passed * a.reviewed - a.passed * b.reviewed;
- if (share != 0) return share;
- final reviewed = b.reviewed - a.reviewed;
- return reviewed != 0 ? reviewed : a.name.compareTo(b.name);
- });
- return shares;
}
-/// The left sidebar's Models, after Machines (Dee, 2026-10-29): the models
-/// whose work reviewers approved most often on first review in the week the
-/// right pane shows, each as a bare percentage, the counts in words on hover
-/// and to a screen reader. The heading shows them all. Nothing shows without
-/// a board of [projectId], or when no model has enough first reviews.
+/// The left sidebar's Models, after Machines (Dee, 2026-10-29): each model's
+/// share of the tasks of the week the right pane shows, behind its engine's
+/// icon, the counts and its first-review pass rate in words on hover and to a
+/// screen reader. Behind the first [leaderboardModelsShown] the rest are one
+/// Other row, so the rows shown add up to 100%; the heading names them all.
+/// Nothing shows without a board of [projectId], or when no model ran a task
+/// that week.
class LeaderboardModelsSection extends StatefulWidget {
const LeaderboardModelsSection({
super.key,
@@ -117,7 +170,7 @@ class _LeaderboardModelsSectionState extends State<LeaderboardModelsSection> {
crossAxisAlignment: CrossAxisAlignment.stretch,
children: [
SuperTabSectionHeading(
- 'Models',
+ 'Models · $week',
key: const ValueKey('super-models-heading'),
endInset: 0,
startInset: SuperTabSectionHeading.indentFor(0),
@@ -127,6 +180,16 @@ class _LeaderboardModelsSectionState extends State<LeaderboardModelsSection> {
for (final share
in _all ? shares : shares.take(leaderboardModelsShown))
_row(share, week, style, column),
+ if (more && !_all)
+ _row(
+ LeaderboardModelShare.other(
+ shares.sublist(leaderboardModelsShown),
+ ),
+ week,
+ style,
+ column,
+ key: const ValueKey('super-models-other'),
+ ),
],
);
},
@@ -136,34 +199,49 @@ class _LeaderboardModelsSectionState extends State<LeaderboardModelsSection> {
LeaderboardModelShare share,
String week,
TextStyle style,
- double column,
- ) {
+ double column, {
+ Key? key,
+ }) {
+ final reviews = share.reviewed == 0
+ ? 'no first review yet'
+ : '${(share.passed * 100 / share.reviewed).round()}% passed first review';
final words =
- '${share.passed} of ${share.reviewed} approved on first review $week';
+ '${share.name}: ${share.attempts} of ${share.total} tasks $week · $reviews';
return Tooltip(
- key: ValueKey('super-model-${share.name}'),
+ key: key ?? ValueKey('super-model-${share.name}'),
message: words,
excludeFromSemantics: true,
child: Semantics(
container: true,
- label: '${share.name}: $words',
+ label: words,
excludeSemantics: true,
child: Padding(
padding: const EdgeInsets.only(bottom: grid.AppDesktop.listRowGap),
child: ConstrainedBox(
constraints: const BoxConstraints(minHeight: _rowHeight),
child: Padding(
- // The sidebar's rows: names after the icon column, the number
+ // The sidebar's rows: the engine in the icon column, the number
// where their badges end.
padding: EdgeInsets.fromLTRB(
- SuperTabSectionHeading.indentFor(0) +
- SuperTabSectionHeading.actionWidth,
+ SuperTabSectionHeading.indentFor(0),
_gap,
_gap + _gap / 2,
_gap,
),
child: Row(
children: [
+ SizedBox(
+ width: SuperTabSectionHeading.actionWidth,
+ // Other has no engine: its icon column stays empty.
+ child: share.engine == null
+ ? null
+ : Center(
+ child: EngineMark(
+ engine: share.engine,
+ size: AppIcons.inlineSize,
+ ),
+ ),
+ ),
Expanded(
child: Text(
share.name,
diff --git a/desktop/test/super_tab_leaderboard_shots_test.dart b/desktop/test/super_tab_leaderboard_shots_test.dart
index 3147846b2..817f5e025 100644
--- a/desktop/test/super_tab_leaderboard_shots_test.dart
+++ b/desktop/test/super_tab_leaderboard_shots_test.dart
@@ -1,5 +1,6 @@
import 'dart:io';
+import 'package:flutter/gestures.dart';
import 'package:flutter/material.dart';
import 'package:flutter_test/flutter_test.dart';
import 'package:harness/shared/theme/app_theme.dart' as grid;
@@ -76,7 +77,24 @@ void main() {
swarm: () => modelsSwarm(membersSwarm()),
size: const Size(1300, 1000),
);
+ // Codex's mark is a picture, which a test decodes only when asked.
+ await tester.runAsync(
+ () => precacheImage(
+ const AssetImage('assets/engine-icons/codex.png'),
+ tester.element(find.byKey(const ValueKey('super-models-heading'))),
+ ),
+ );
+ await tester.pumpAndSettle();
await captureWindow(tester, shots, 'members-$name');
+ // A Models row's hover line.
+ final mouse = await tester.createGesture(kind: PointerDeviceKind.mouse);
+ await mouse.addPointer(location: Offset.zero);
+ await mouse.moveTo(
+ tester.getCenter(find.byKey(const ValueKey('super-model-Opus 5.5'))),
+ );
+ await tester.pumpAndSettle(const Duration(seconds: 2));
+ await captureWindow(tester, shots, 'models-hover-$name');
+ await mouse.removePointer();
await tester.pumpWidget(const SizedBox());
}
}, skip: shots == null);
diff --git a/desktop/test/super_tab_members_list_test.dart b/desktop/test/super_tab_members_list_test.dart
index a6d437e37..48614e354 100644
--- a/desktop/test/super_tab_members_list_test.dart
+++ b/desktop/test/super_tab_members_list_test.dart
@@ -244,7 +244,7 @@ void main() {
);
await tester.pumpAndSettle();
expect(tester.takeException(), isNull);
- for (final number in ['21', '92%']) {
+ for (final number in ['21', '44%']) {
expect(find.text(number), findsOneWidget, reason: number);
final paragraph = tester.renderObject<RenderParagraph>(find.text(number));
expect(paragraph.didExceedMaxLines, isFalse, reason: number);
diff --git a/desktop/test/super_tab_sidebar_models_test.dart b/desktop/test/super_tab_sidebar_models_test.dart
index a4a3d90e8..53b65434d 100644
--- a/desktop/test/super_tab_sidebar_models_test.dart
+++ b/desktop/test/super_tab_sidebar_models_test.dart
@@ -3,14 +3,15 @@ import 'package:flutter_test/flutter_test.dart';
import 'package:harness/shared/theme/app_theme.dart' as grid;
import 'package:harness/super_tabs/controller.dart';
import 'package:harness/super_tabs/view.dart';
+import 'package:harness/widgets/engine_identity.dart';
import 'support/humans_fixture.dart';
import 'support/super_tab_fixture.dart';
-/// Models in the left sidebar, after Machines (Dee, 2026-10-29): the models
-/// whose work reviewers approved most often on first review in the week the
-/// right pane shows, as a bare percentage, the counts in words on hover and
-/// for a screen reader. A model needs 3 first reviews to be listed.
+/// Models in the left sidebar, after Machines (Dee, 2026-10-29): each model's
+/// share of the tasks of the week the right pane shows, rounded so the rows
+/// add up to 100%, behind its engine's icon; the counts and its first-review
+/// pass rate in words on hover and for a screen reader.
void main() {
Future<SuperTabController> show(
WidgetTester tester,
@@ -48,23 +49,32 @@ void main() {
}
final heading = find.byKey(const ValueKey('super-models-heading'));
+ final headingText = find.descendant(
+ of: heading,
+ matching: find.text('Models · this week'),
+ );
Finder row(String name) => find.byKey(ValueKey('super-model-$name'));
Finder inRow(String name, Finder finder) =>
find.descendant(of: row(name), matching: finder);
+ Finder mark(String name, String engine) => inRow(
+ name,
+ find.byWidgetPredicate((w) => w is EngineMark && w.engine == engine),
+ );
testWidgets(
- 'lists the top three by first-review approval after Machines, as a bare percentage',
+ 'lists the top three by share of the week\'s tasks after Machines',
(tester) async {
final controller = await show(tester, modelsSwarm());
- expect(heading, findsOneWidget);
+ expect(headingText, findsOneWidget);
expect(
tester.getTopLeft(heading).dy,
greaterThan(tester.getTopLeft(find.text('Machines')).dy),
);
- expect(inRow('Opus 5.5', find.text('92%')), findsOneWidget);
- expect(inRow('GPT-6.1 Sol', find.text('81%')), findsOneWidget);
- expect(inRow('Sonnet 5.5', find.text('70%')), findsOneWidget);
- // Fourth, so not shown until asked for; too few reviews, never.
+ // 32 tasks: 14, 9, 6 and 3 of them, 43.75, 28.1, 18.75 and 9.4%.
+ expect(inRow('Opus 5.5', find.text('44%')), findsOneWidget);
+ expect(inRow('GPT-6.1 Sol', find.text('28%')), findsOneWidget);
+ expect(inRow('Sonnet 5.5', find.text('19%')), findsOneWidget);
+ // Fourth, so not shown until asked for; no task of its own, never.
expect(row('Haiku 5.5'), findsNothing);
expect(row('GPT-6.1 Luna'), findsNothing);
final top = tester.getTopLeft(row('Opus 5.5')).dy;
@@ -75,98 +85,244 @@ void main() {
expect(find.textContaining('Opus 5.5 ·'), findsNothing);
// The percentages line up at their trailing edge.
expect(
- tester.getTopRight(find.text('92%')).dx,
- tester.getTopRight(find.text('70%')).dx,
+ tester.getTopRight(find.text('44%')).dx,
+ tester.getTopRight(find.text('19%')).dx,
);
await close(tester, controller);
},
);
- testWidgets('says the counts in words on hover and to a screen reader', (
+ testWidgets('puts its engine\'s icon before each model\'s name', (
tester,
) async {
- final semantics = tester.ensureSemantics();
final controller = await show(tester, modelsSwarm());
+ for (final (name, engine) in [
+ ('Opus 5.5', 'claude'),
+ ('GPT-6.1 Sol', 'codex'),
+ ('Sonnet 5.5', 'claude'),
+ ]) {
+ expect(mark(name, engine), findsOneWidget, reason: name);
+ expect(
+ tester.getTopRight(mark(name, engine)).dx,
+ lessThan(tester.getTopLeft(inRow(name, find.text(name))).dx),
+ reason: name,
+ );
+ }
+ // In the sidebar's icon column: under the channel marks' column.
expect(
- find.byTooltip('11 of 12 approved on first review this week'),
- findsOneWidget,
- );
- expect(
- find.bySemanticsLabel(
- 'Opus 5.5: 11 of 12 approved on first review this week',
- ),
- findsOneWidget,
+ tester.getCenter(mark('Opus 5.5', 'claude')).dx,
+ closeTo(tester.getCenter(find.text('#').first).dx, 1),
);
- semantics.dispose();
await close(tester, controller);
});
+ testWidgets(
+ 'says the tasks and first-review pass rate on hover and to a screen reader',
+ (tester) async {
+ final semantics = tester.ensureSemantics();
+ final controller = await show(tester, modelsSwarm());
+ const opus =
+ 'Opus 5.5: 14 of 32 tasks this week · 92% passed first review';
+ expect(find.byTooltip(opus), findsOneWidget);
+ expect(find.bySemanticsLabel(opus), findsOneWidget);
+ expect(
+ find.byTooltip(
+ 'GPT-6.1 Sol: 9 of 32 tasks this week · 75% passed first review',
+ ),
+ findsOneWidget,
+ );
+ semantics.dispose();
+ await close(tester, controller);
+ },
+ );
+
testWidgets('shows the whole list from its heading, and back', (
tester,
) async {
final controller = await show(tester, modelsSwarm());
- await tester.tap(
- find.descendant(of: heading, matching: find.text('Models')),
- );
+ await tester.tap(headingText);
await tester.pumpAndSettle();
- expect(inRow('Haiku 5.5', find.text('50%')), findsOneWidget);
- expect(row('GPT-6.1 Luna'), findsNothing);
- await tester.tap(
- find.descendant(of: heading, matching: find.text('Models')),
+ expect(inRow('Haiku 5.5', find.text('9%')), findsOneWidget);
+ expect(mark('Haiku 5.5', 'claude'), findsOneWidget);
+ // Not reviewed yet: says so rather than a pass rate.
+ expect(
+ find.byTooltip(
+ 'Haiku 5.5: 3 of 32 tasks this week · no first review yet',
+ ),
+ findsOneWidget,
);
+ expect(row('GPT-6.1 Luna'), findsNothing);
+ await tester.tap(headingText);
await tester.pumpAndSettle();
expect(row('Haiku 5.5'), findsNothing);
await close(tester, controller);
});
+ testWidgets(
+ 'adds the models behind the heading as one Other row, so the shown rows add up to 100%',
+ (tester) async {
+ final semantics = tester.ensureSemantics();
+ final controller = await show(tester, modelsSwarm());
+ final other = find.byKey(const ValueKey('super-models-other'));
+ expect(other, findsOneWidget);
+ // Haiku's 3 of 32 tasks, 9%: 44 + 28 + 19 + 9 = 100.
+ expect(
+ find.descendant(of: other, matching: find.text('Other')),
+ findsOneWidget,
+ );
+ expect(
+ find.descendant(of: other, matching: find.text('9%')),
+ findsOneWidget,
+ );
+ final shown = [
+ for (final name in ['Opus 5.5', 'GPT-6.1 Sol', 'Sonnet 5.5'])
+ inRow(name, find.textContaining('%')),
+ find.descendant(of: other, matching: find.textContaining('%')),
+ ];
+ expect(
+ shown
+ .map(
+ (f) =>
+ int.parse(tester.widget<Text>(f).data!.replaceAll('%', '')),
+ )
+ .fold<int>(0, (sum, percent) => sum + percent),
+ 100,
+ );
+ const words = 'Other: 3 of 32 tasks this week · no first review yet';
+ expect(find.byTooltip(words), findsOneWidget);
+ expect(find.bySemanticsLabel(words), findsOneWidget);
+ // The whole list names them instead.
+ await tester.tap(headingText);
+ await tester.pumpAndSettle();
+ expect(other, findsNothing);
+ expect(inRow('Haiku 5.5', find.text('9%')), findsOneWidget);
+ semantics.dispose();
+ await close(tester, controller);
+ },
+ );
+
+ testWidgets('has no Other row when every model is shown', (tester) async {
+ final server = modelsSwarm();
+ server.leaderboard!['byModel'] = [
+ modelScore('claude', 'claude-opus-5-5', 'high', attempts: 3),
+ modelScore('codex', 'gpt-6.1-sol', null, attempts: 1),
+ ];
+ final controller = await show(tester, server);
+ expect(find.byKey(const ValueKey('super-models-other')), findsNothing);
+ expect(inRow('Opus 5.5', find.text('75%')), findsOneWidget);
+ expect(inRow('GPT-6.1 Sol', find.text('25%')), findsOneWidget);
+ await close(tester, controller);
+ });
+
+ testWidgets('rounds the shares so all rows add up to 100%', (tester) async {
+ final server = modelsSwarm();
+ // A third each: plain rounding says 33% three times, 99%.
+ server.leaderboard!['byModel'] = [
+ modelScore('claude', 'claude-opus-5-5', 'high', attempts: 1),
+ modelScore('codex', 'gpt-6.1-sol', null, attempts: 1),
+ modelScore('claude', 'claude-sonnet-5-5', 'high', attempts: 1),
+ ];
+ final controller = await show(tester, server);
+ // Of three alike, by name.
+ expect(inRow('GPT-6.1 Sol', find.text('34%')), findsOneWidget);
+ expect(inRow('Opus 5.5', find.text('33%')), findsOneWidget);
+ expect(inRow('Sonnet 5.5', find.text('33%')), findsOneWidget);
+ expect(
+ tester.getTopLeft(row('GPT-6.1 Sol')).dy,
+ lessThan(tester.getTopLeft(row('Opus 5.5')).dy),
+ );
+ await close(tester, controller);
+ });
+
testWidgets('follows the week the right pane shows', (tester) async {
final server = modelsSwarm();
final controller = await show(tester, server);
server.leaderboard!['byModel'] = [
- modelScore('claude', 'claude-sonnet-5-5', 'high', reviewed: 4, passed: 4),
+ modelScore(
+ 'claude',
+ 'claude-sonnet-5-5',
+ 'high',
+ attempts: 4,
+ reviewed: 4,
+ passed: 4,
+ ),
];
await tester.tap(find.byKey(const ValueKey('leaderboard-week')));
await tester.pumpAndSettle();
await tester.tap(find.text('Last week').last);
await tester.pumpAndSettle();
expect(server.leaderboardReads.last, contains('&week='));
+ expect(
+ find.descendant(of: heading, matching: find.text('Models · last week')),
+ findsOneWidget,
+ );
expect(row('Opus 5.5'), findsNothing);
expect(inRow('Sonnet 5.5', find.text('100%')), findsOneWidget);
expect(
- find.byTooltip('4 of 4 approved on first review last week'),
+ find.byTooltip(
+ 'Sonnet 5.5: 4 of 4 tasks last week · 100% passed first review',
+ ),
findsOneWidget,
);
await close(tester, controller);
});
testWidgets(
- 'lists a model run through two engines once, its reviews together',
+ 'lists a model run through two engines once, behind the engine that ran most of it',
(tester) async {
// Review of e4b70e9: one grid model through Claude Code and Codex drew
// two rows with one key ("Duplicate keys found").
final server = modelsSwarm();
server.leaderboard!['byModel'] = [
- modelScore('claude', 'qwen3', 'high', reviewed: 4, passed: 3),
- modelScore('codex', 'qwen3', null, reviewed: 4, passed: 4),
+ modelScore(
+ 'claude',
+ 'qwen3',
+ 'high',
+ attempts: 3,
+ reviewed: 4,
+ passed: 3,
+ ),
+ modelScore('codex', 'qwen3', null, attempts: 5, reviewed: 4, passed: 4),
];
final controller = await show(tester, server);
expect(tester.takeException(), isNull);
expect(row('qwen3'), findsOneWidget);
- expect(inRow('qwen3', find.text('88%')), findsOneWidget);
+ expect(inRow('qwen3', find.text('100%')), findsOneWidget);
+ expect(mark('qwen3', 'codex'), findsOneWidget);
expect(
- find.byTooltip('7 of 8 approved on first review this week'),
+ find.byTooltip(
+ 'qwen3: 8 of 8 tasks this week · 88% passed first review',
+ ),
findsOneWidget,
);
await close(tester, controller);
},
);
- testWidgets('hides Models when no model has three first reviews', (
+ testWidgets('hides Models when no model ran a task this week', (
tester,
) async {
final server = modelsSwarm();
server.leaderboard!['byModel'] = [
- modelScore('claude', 'claude-opus-5-5', 'high', reviewed: 2, passed: 2),
+ modelScore('codex', 'gpt-6.1-luna', null, attempts: 0, reviews: 3),
+ ];
+ final controller = await show(tester, server);
+ expect(heading, findsNothing);
+ await close(tester, controller);
+ });
+
+ testWidgets('hides Models from a server before task counts', (tester) async {
+ // Its rows carry first reviews but no tasks: no share to show.
+ final server = modelsSwarm();
+ server.leaderboard!['byModel'] = [
+ modelScore(
+ 'claude',
+ 'claude-opus-5-5',
+ 'high',
+ attempts: null,
+ reviewed: 5,
+ passed: 5,
+ ),
];
final controller = await show(tester, server);
expect(heading, findsNothing);
@@ -183,7 +339,7 @@ void main() {
await controller.refresh();
await tester.pumpAndSettle();
expect(heading, findsNothing);
- expect(find.text('92%'), findsNothing);
+ expect(find.text('44%'), findsNothing);
await close(tester, controller);
});
diff --git a/desktop/test/support/humans_fixture.dart b/desktop/test/support/humans_fixture.dart
index 07ae0d4a5..fa3fa9d22 100644
--- a/desktop/test/support/humans_fixture.dart
+++ b/desktop/test/support/humans_fixture.dart
@@ -186,13 +186,17 @@ ConversationServer membersSwarm() {
return server;
}
-/// One model's week on the board: its first reviews and how many passed.
+/// One model's week on the board: the tasks it ran, its first reviews and
+/// how many passed. A board from a server before task counts has no
+/// `attempts` ([attempts] null).
Map<String, dynamic> modelScore(
String engine,
String? model,
String? effort, {
- required int reviewed,
- required int passed,
+ required int? attempts,
+ int reviewed = 0,
+ int passed = 0,
+ int reviews = 0,
}) => {
'engine': engine,
'model': model,
@@ -200,29 +204,54 @@ Map<String, dynamic> modelScore(
'grid': false,
'points': 0,
'rank': 1,
+ 'attempts': ?attempts,
'shipped': {'count': 0, 'medianAttemptMs': null, 'medianRequestMs': null},
'firstReview': {'reviewed': reviewed, 'passed': passed},
- 'reviews': 0,
+ 'reviews': reviews,
};
-/// The week's models (Dee, 2026-10-29): Opus at two efforts, 11 of 12 together.
+/// The week's models (Dee, 2026-10-29): 32 tasks, Opus at two efforts 14 of
+/// them together.
ConversationServer modelsSwarm([ConversationServer? base]) {
final server = base ?? humansSwarm();
server.leaderboard!['byModel'] = [
- // One model at two efforts is one row: 11 of 12.
- modelScore('claude', 'claude-opus-5-5', 'high', reviewed: 10, passed: 9),
- modelScore('claude', 'claude-opus-5-5', 'xhigh', reviewed: 2, passed: 2),
- modelScore('codex', 'gpt-6.1-sol', null, reviewed: 16, passed: 13),
+ // One model at two efforts is one row: 14 tasks, 11 of 12 passed.
+ modelScore(
+ 'claude',
+ 'claude-opus-5-5',
+ 'high',
+ attempts: 12,
+ reviewed: 10,
+ passed: 9,
+ ),
+ modelScore(
+ 'claude',
+ 'claude-opus-5-5',
+ 'xhigh',
+ attempts: 2,
+ reviewed: 2,
+ passed: 2,
+ ),
+ modelScore(
+ 'codex',
+ 'gpt-6.1-sol',
+ null,
+ attempts: 9,
+ reviewed: 8,
+ passed: 6,
+ ),
modelScore(
'claude',
'claude-sonnet-5-5',
'medium',
- reviewed: 10,
- passed: 7,
+ attempts: 6,
+ reviewed: 5,
+ passed: 4,
),
- modelScore('claude', 'claude-haiku-5-5', 'low', reviewed: 4, passed: 2),
- // Too few first reviews to say anything, however good.
- modelScore('codex', 'gpt-6.1-luna', null, reviewed: 2, passed: 2),
+ // Fourth, and no first review yet.
+ modelScore('claude', 'claude-haiku-5-5', 'low', attempts: 3),
+ // Only reviewed others' work: no task of its own.
+ modelScore('codex', 'gpt-6.1-luna', null, attempts: 0, reviews: 5),
];
return server;
}