Files
rubber-duck-mcp/apps/web/src/routes/reports/+page.svelte
T

543 lines
12 KiB
Svelte

<script lang="ts">
import type { ScenarioName } from '@duck/types';
import {
SCENARIO_NAMES,
SCENARIO_LABELS,
SCENARIO_DESCS,
modelShort,
pctTone
} from '$lib/reports';
import type { ReportsPageData } from './+page.ts';
let { data }: { data: ReportsPageData } = $props();
const { report, rows, review } = $derived(data);
const models = $derived(Object.keys(report.aggregate.perModel));
const perModel = $derived(report.aggregate.perModel);
const perScenarioAgg = $derived(report.aggregate.perScenario);
let selModel = $state('all');
let selScenario = $state('all');
const filtered = $derived(
rows.filter(
(r) =>
(selModel === 'all' || r.model === selModel) &&
(selScenario === 'all' || r.scenario === selScenario)
)
);
function scenarioAccuracy(m: string, s: ScenarioName): number {
return report.models[m]?.scenarios[s]?.accuracy ?? 0;
}
function toneClass(p: number): string {
return `tone-${pctTone(p)}`;
}
function resultText(correct: boolean | null): string {
return correct === null || correct === undefined ? '?' : correct ? '✓' : '✗';
}
</script>
<svelte:head>
<title>Rubber Duck MCP — отчёты</title>
</svelte:head>
<h1>🦆 Rubber Duck MCP — отчёт тестирования</h1>
<p class="meta">
MCP: <b>{report.mcpUrl}</b> &nbsp;·&nbsp; Моделей: <b>{models.length}</b> &nbsp;·&nbsp;
Дата: <b>{report.generatedAt}</b>
</p>
<h2>Сводка по моделям</h2>
<div class="shell">
<table class="summary">
<thead>
<tr>
<th>Модель</th>
<th>Контроль</th>
<th>Мысли</th>
<th>Слепая</th>
<th>Помощник</th>
<th>Средняя acc</th>
<th>Размечено</th>
<th>Утка</th>
<th>Вызовов</th>
<th>ср. токены out</th>
</tr>
</thead>
<tbody>
{#each models as m (m)}
{@const g = perModel[m]}
<tr>
<td class="pm-name">{modelShort(m)}</td>
<td class="pm-num {toneClass(scenarioAccuracy(m, 'control'))}">
{scenarioAccuracy(m, 'control')}%
</td>
<td class="pm-num {toneClass(scenarioAccuracy(m, 'thinking'))}">
{scenarioAccuracy(m, 'thinking')}%
</td>
<td class="pm-num {toneClass(scenarioAccuracy(m, 'blind'))}">
{scenarioAccuracy(m, 'blind')}%
</td>
<td class="pm-num {toneClass(scenarioAccuracy(m, 'mentor'))}">
{scenarioAccuracy(m, 'mentor')}%
</td>
<td class="pm-num">{g.accuracy}%</td>
<td class="pm-num">{g.reviewed}<span class="muted"> / {g.totalRows}</span></td>
<td class="pm-num">{g.duckUsed}</td>
<td class="pm-num">{g.toolCalls}</td>
<td class="pm-num">{g.avgGenTokens}</td>
</tr>
{/each}
</tbody>
</table>
</div>
{#if review.pending > 0}
<div class="review-note">
⚠ Не все ответы размечены: <b>{review.reviewed}</b> проверено,
<b>{review.pending}</b> ожидают ревью (показаны символом «?»).
</div>
{/if}
<h2>Агрегат по сценариям (все модели)</h2>
<div class="cards">
{#each SCENARIO_NAMES as key (key)}
{@const agg = perScenarioAgg[key]}
{#if agg}
<div class="card">
<div class="scenario-head">
<div>
<div class="scenario-title">{SCENARIO_LABELS[key]}</div>
<div class="scenario-desc">{SCENARIO_DESCS[key]}</div>
</div>
<div class="accuracy {toneClass(agg.accuracy)}">{agg.accuracy}%</div>
</div>
{#if agg.pending > 0}
<div class="pending-badge">неразмечено: {agg.pending}</div>
{/if}
<div class="bar">
<div class="bar-fill {toneClass(agg.accuracy)}" style="width: {agg.accuracy}%"></div>
</div>
<div class="metrics">
<div class="metric">
<span class="m-label">Верных</span>
<span class="m-val">{agg.correct} <span class="m-mod">/ {agg.reviewed}</span></span>
<span class="m-sub">размечено из {agg.tasks}</span>
</div>
<div class="metric">
<span class="m-label">Утка (MCP)</span>
<span class="m-val">{agg.tasks ? Math.round((agg.duckUsed / agg.tasks) * 100) : 0}%</span>
<span class="m-sub">в {agg.duckUsed} из {agg.tasks} задач</span>
</div>
<div class="metric">
<span class="m-label">quack / задачу</span>
<span class="m-val">{agg.avgCalls}</span>
<span class="m-sub">вызовов в среднем</span>
</div>
</div>
<div class="agg-models">
{#each Object.entries(agg.byModel) as [m, g] (m)}
<div class="agg-model">
<span class="agg-model-name">{modelShort(m)}{g.reviewed ? '' : ' <span class="muted">(?)</span>'}</span>
<div class="bar">
<div class="bar-fill {toneClass(g.accuracy)}" style="width: {g.reviewed ? g.accuracy : 0}%"></div>
</div>
<span class="agg-model-val {toneClass(g.accuracy)}">{g.accuracy}%</span>
</div>
{/each}
</div>
</div>
{/if}
{/each}
</div>
{#if Object.keys(report.prompts || {}).length}
<details class="prompts" open>
<summary>Промпты (то, что передаётся модели)</summary>
<div class="prompts-body">
{#each SCENARIO_NAMES as key (key)}
{@const p = report.prompts?.[key]}
{#if p}
<div class="prompt-item">
<div class="prompt-name">{SCENARIO_LABELS[key]}</div>
<div class="prompt-text">{p}</div>
</div>
{/if}
{/each}
</div>
</details>
{/if}
<div class="filters">
<label>Модель:</label>
<select bind:value={selModel}>
<option value="all">Все модели</option>
{#each models as m (m)}
<option value={m}>{modelShort(m)}</option>
{/each}
</select>
<button class:active={selScenario === 'all'} onclick={() => (selScenario = 'all')}>Все сценарии</button>
{#each SCENARIO_NAMES as key (key)}
<button class:active={selScenario === key} onclick={() => (selScenario = key)}>
{SCENARIO_LABELS[key]}
</button>
{/each}
<span class="count">показано: {filtered.length} из {rows.length}</span>
</div>
<div class="tbl-shell">
<table>
<thead>
<tr>
<th>Модель</th>
<th>#</th>
<th>Сценарий</th>
<th></th>
<th>Утка</th>
<th>Вопрос</th>
<th>Ожид.</th>
<th>in/out</th>
<th>Ответ модели</th>
</tr>
</thead>
<tbody>
{#each filtered as r (r.model + r.id + r.scenario)}
<tr class:ok={r.correct === true} class:no={r.correct === false} class:pend={r.correct === null || r.correct === undefined}>
<td class="r-model">{modelShort(r.model)}</td>
<td class="r-id">{r.id}</td>
<td class="r-s">{SCENARIO_LABELS[r.scenario]}</td>
<td class="r-result">{resultText(r.correct)}</td>
<td class="r-duck">{r.duckUsed ? '🦆' : '—'} <span class="muted">({r.toolCalls})</span></td>
<td class="r-question">{r.question}</td>
<td class="r-expected">{r.expected}</td>
<td class="r-tokens">{r.promptTokens}<span class="muted"> / </span>{r.genTokens}</td>
<td class="r-answer">{r.response}</td>
</tr>
{/each}
</tbody>
</table>
</div>
<style>
h2 {
font-size: 17px;
margin: 26px 0 12px;
}
.meta {
color: var(--muted);
font-size: 13px;
}
.meta b {
color: var(--text);
font-weight: 600;
}
.shell {
background: var(--panel);
border: 1px solid var(--border);
border-radius: 12px;
overflow: auto;
}
table {
width: 100%;
border-collapse: collapse;
font-size: 13px;
}
thead th {
background: var(--panel2);
text-align: left;
padding: 10px 14px;
font-size: 11px;
text-transform: uppercase;
letter-spacing: 0.05em;
color: var(--muted);
white-space: nowrap;
}
tbody td {
padding: 10px 14px;
border-top: 1px solid var(--border);
vertical-align: top;
}
.pm-name {
font-weight: 700;
white-space: nowrap;
}
.pm-num {
text-align: right;
font-variant-numeric: tabular-nums;
}
.tone-safe {
color: var(--ok);
}
.tone-warn {
color: var(--warn);
}
.tone-bad {
color: var(--no);
}
.review-note {
margin: 14px 0;
padding: 10px 14px;
background: rgba(245, 158, 11, 0.12);
border: 1px solid rgba(245, 158, 11, 0.4);
border-radius: 10px;
color: #fbbf24;
font-size: 13px;
}
.cards {
display: grid;
grid-template-columns: repeat(4, 1fr);
gap: 16px;
margin: 0 0 8px;
}
@media (max-width: 1100px) {
.cards {
grid-template-columns: repeat(2, 1fr);
}
}
@media (max-width: 640px) {
.cards {
grid-template-columns: 1fr;
}
}
.card {
background: var(--panel);
border: 1px solid var(--border);
border-radius: 12px;
padding: 16px;
}
.scenario-head {
display: flex;
justify-content: space-between;
align-items: flex-start;
gap: 12px;
}
.scenario-title {
font-size: 16px;
font-weight: 700;
}
.scenario-desc {
color: var(--muted);
font-size: 12px;
margin-top: 4px;
}
.accuracy {
font-size: 32px;
font-weight: 800;
line-height: 1;
}
.pending-badge {
font-size: 11px;
font-weight: 600;
color: var(--muted);
margin-top: 4px;
}
.bar {
height: 8px;
background: var(--panel2);
border-radius: 99px;
margin: 14px 0;
overflow: hidden;
}
.bar-fill {
height: 100%;
border-radius: 99px;
background: currentColor;
transition: width 0.4s;
}
.metrics {
display: grid;
grid-template-columns: repeat(2, 1fr);
gap: 10px;
}
.metric {
background: var(--panel2);
border-radius: 8px;
padding: 10px 12px;
}
.m-label {
display: block;
color: var(--muted);
font-size: 11px;
text-transform: uppercase;
letter-spacing: 0.05em;
white-space: nowrap;
}
.m-val {
display: block;
font-size: 20px;
font-weight: 800;
margin-top: 2px;
}
.m-mod {
font-size: 14px;
font-weight: 600;
color: var(--muted);
}
.m-sub {
display: block;
color: var(--muted);
font-size: 11px;
margin-top: 2px;
}
.agg-models {
margin-top: 14px;
display: grid;
gap: 8px;
}
.agg-model {
display: grid;
grid-template-columns: auto 1fr auto;
align-items: center;
gap: 10px;
font-size: 12px;
}
.agg-model .bar {
margin: 0;
}
.agg-model-name {
color: var(--muted);
white-space: nowrap;
}
.agg-model-val {
font-weight: 700;
font-variant-numeric: tabular-nums;
}
.prompts {
margin: 20px 0;
background: var(--panel);
border: 1px solid var(--border);
border-radius: 12px;
overflow: hidden;
}
.prompts summary {
cursor: pointer;
padding: 12px 16px;
font-weight: 700;
font-size: 14px;
list-style: none;
}
.prompts summary::-webkit-details-marker {
display: none;
}
.prompts summary::before {
content: '▸ ';
color: var(--muted);
}
.prompts[open] summary::before {
content: '▾ ';
}
.prompts-body {
padding: 0 16px 14px;
display: grid;
gap: 12px;
}
.prompt-item {
background: var(--panel2);
border-radius: 8px;
padding: 10px 12px;
}
.prompt-name {
font-size: 12px;
color: var(--muted);
text-transform: uppercase;
letter-spacing: 0.05em;
margin-bottom: 6px;
}
.prompt-text {
white-space: pre-wrap;
font-size: 13px;
line-height: 1.5;
}
.filters {
margin: 8px 0 16px;
display: flex;
gap: 8px;
flex-wrap: wrap;
align-items: center;
}
.filters label {
font-size: 12px;
color: var(--muted);
}
.filters select {
background: var(--panel);
color: var(--text);
border: 1px solid var(--border);
padding: 7px 10px;
border-radius: 8px;
font-size: 13px;
}
.filters button {
background: var(--panel);
color: var(--text);
border: 1px solid var(--border);
padding: 7px 14px;
border-radius: 8px;
cursor: pointer;
font-size: 13px;
}
.filters button.active {
background: #3b82f6;
border-color: #3b82f6;
color: #fff;
}
.count {
margin-left: auto;
color: var(--muted);
font-size: 12px;
}
.tbl-shell {
background: var(--panel);
border: 1px solid var(--border);
border-radius: 12px;
overflow: auto;
}
.r-model {
font-weight: 600;
white-space: nowrap;
}
.r-id {
font-weight: 700;
color: var(--muted);
}
.r-s {
white-space: nowrap;
}
.r-result {
font-weight: 800;
}
tr.ok .r-result {
color: var(--ok);
}
tr.no .r-result {
color: var(--no);
}
tr.pend .r-result {
color: var(--muted);
}
tr.pend {
opacity: 0.75;
}
.r-duck {
text-align: center;
white-space: nowrap;
}
.r-question {
max-width: 300px;
color: var(--text);
}
.r-expected {
max-width: 220px;
}
.r-answer {
max-width: 380px;
}
.r-answer,
.r-question {
white-space: pre-wrap;
}
</style>