agenteval-toolkit 0.1.0__py3-none-any.whl

agenteval/models.py

@@ -0,0 +1,136 @@
+# SPDX-License-Identifier: Apache-2.0
+# Copyright 2026 The AgentEval Authors
+"""Core data models: rules, assets, findings and scan results."""
+
+from __future__ import annotations
+
+from datetime import date, datetime, timezone
+from enum import Enum
+from typing import Any
+
+from pydantic import BaseModel, Field
+
+
+class Severity(str, Enum):
+    critical = "critical"
+    high = "high"
+    medium = "medium"
+    low = "low"
+    info = "info"
+
+    @property
+    def rank(self) -> int:
+        return ["info", "low", "medium", "high", "critical"].index(self.value)
+
+
+class RuleType(str, Enum):
+    automated = "automated"      # Decided automatically by the scanner from collected facts
+    attestation = "attestation"  # Requires a human-submitted attestation (e.g. an impact assessment)
+
+
+class Status(str, Enum):
+    passed = "pass"
+    failed = "fail"
+    not_applicable = "not_applicable"
+    manual = "manual"            # Awaiting human attestation
+    error = "error"
+
+
+class Source(BaseModel):
+    """Regulation/framework source a rule maps to."""
+
+    framework: str
+    reference: str
+    url: str | None = None
+    # Set to True only after legal/compliance has reviewed the mapping
+    verified: bool = False
+
+
+class Applicability(BaseModel):
+    asset_types: list[str]
+    when: dict[str, Any] | None = None  # Optional precondition; same syntax as check
+
+
+class RulePack(BaseModel):
+    """Rule pack, organized by tier (national / regional / industry)."""
+
+    id: str
+    name: str
+    name_zh: str = ""
+    tier: str  # national | regional | industry
+    jurisdiction: str
+    description: str = ""
+    description_zh: str = ""
+
+
+class Rule(BaseModel):
+    id: str
+    title: str
+    description: str
+    severity: Severity
+    type: RuleType
+    jurisdiction: str = "us-federal"
+    category: str
+    applies_to: Applicability
+    check: dict[str, Any] | None = None
+    attestation: dict[str, Any] | None = None
+    remediation: str
+    sources: list[Source]
+    effective_date: date | None = None
+    version: str = "1.0.0"
+    enabled: bool = True
+    pack: str = ""  # Set by the loader from the pack.yaml in the rule's directory
+    # Attached by the loader from i18n/<lang>.yaml: {lang: {title, description, remediation, sources: [reference...]}}
+    translations: dict[str, dict[str, Any]] = Field(default_factory=dict)
+
+
+class Asset(BaseModel):
+    """Normalized asset produced by a collector. facts holds the flat/nested facts used for rule evaluation."""
+
+    id: str
+    type: str
+    name: str
+    namespace: str | None = None
+    facts: dict[str, Any] = Field(default_factory=dict)
+
+
+class Finding(BaseModel):
+    rule_id: str
+    pack: str = ""
+    rule_title: str
+    severity: Severity
+    category: str
+    asset_id: str
+    asset_name: str
+    asset_type: str
+    status: Status
+    message: str                      # Default Chinese text (backward compatibility)
+    message_code: str = ""            # for multilingual rendering; see agenteval/i18n.py
+    message_params: dict[str, Any] = Field(default_factory=dict)
+    evidence: dict[str, Any] = Field(default_factory=dict)
+    remediation: str
+    sources: list[Source]
+    translations: dict[str, dict[str, Any]] = Field(default_factory=dict)
+
+
+class ScanSummary(BaseModel):
+    total: int = 0
+    by_status: dict[str, int] = Field(default_factory=dict)
+    failed_by_severity: dict[str, int] = Field(default_factory=dict)
+    score: float = 100.0  # passed / (passed + failed); manual items are excluded
+
+
+class ScanResult(BaseModel):
+    scan_id: str
+    source: str
+    started_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc))
+    finished_at: datetime | None = None
+    rules_version: str
+    asset_count: int
+    findings: list[Finding]
+    summary: ScanSummary
+    disclaimer: str = (
+        "This report is an evidence-based assessment of technical controls and does not constitute legal "
+        "advice or a determination of regulatory compliance. Rule-to-regulation mappings must be reviewed "
+        "by your legal/compliance team."
+    )

agenteval/reports/__init__.py

@@ -0,0 +1,86 @@
+# SPDX-License-Identifier: Apache-2.0
+# Copyright 2026 The AgentEval Authors
+"""Report output: JSON / SARIF 2.1.0 / HTML."""
+
+from __future__ import annotations
+
+import json
+from importlib import resources
+
+from jinja2 import Environment, select_autoescape
+
+from agenteval import __version__
+from agenteval.i18n import DEFAULT_LANG, UI, localize_result
+from agenteval.models import ScanResult, Severity, Status
+
+FORMATS = ("json", "sarif", "html")
+
+_SARIF_LEVEL = {
+    Severity.critical: "error", Severity.high: "error",
+    Severity.medium: "warning", Severity.low: "note", Severity.info: "note",
+}
+
+
+def to_json(result: ScanResult) -> str:
+    return result.model_dump_json(indent=2)
+
+
+def to_sarif(result: ScanResult) -> str:
+    rules_seen: dict[str, dict] = {}
+    results = []
+    for f in result.findings:
+        if f.status not in (Status.failed, Status.manual, Status.error):
+            continue
+        if f.rule_id not in rules_seen:
+            rules_seen[f.rule_id] = {
+                "id": f.rule_id,
+                "name": f.rule_id.replace("-", ""),
+                "shortDescription": {"text": f.rule_title},
+                "help": {"text": f.remediation},
+                "properties": {
+                    "severity": f.severity.value,
+                    "category": f.category,
+                    "sources": [s.model_dump() for s in f.sources],
+                },
+            }
+        results.append({
+            "ruleId": f.rule_id,
+            "level": _SARIF_LEVEL[f.severity] if f.status == Status.failed else "note",
+            "kind": "fail" if f.status == Status.failed else "review",
+            "message": {"text": f"[{f.asset_id}] {f.message}"},
+            "locations": [{"logicalLocations": [{"fullyQualifiedName": f.asset_id, "kind": f.asset_type}]}],
+            "properties": {"status": f.status.value, "evidence": f.evidence},
+        })
+    doc = {
+        "$schema": "https://json.schemastore.org/sarif-2.1.0.json",
+        "version": "2.1.0",
+        "runs": [{
+            "tool": {"driver": {
+                "name": "agenteval", "version": __version__,
+                "informationUri": "https://github.com/your-org/agenteval",
+                "rules": list(rules_seen.values()),
+            }},
+            "automationDetails": {"id": f"agenteval/{result.scan_id}"},
+            "properties": {"rulesVersion": result.rules_version, "summary": result.summary.model_dump()},
+            "results": results,
+        }],
+    }
+    return json.dumps(doc, indent=2, ensure_ascii=False, default=str)
+
+
+def to_html(result: ScanResult, lang: str = DEFAULT_LANG) -> str:
+    env = Environment(autoescape=select_autoescape(default=True, default_for_string=True))
+    env.filters["tojson_pretty"] = lambda v: json.dumps(v, ensure_ascii=False, indent=1, default=str)
+    tpl_text = resources.files("agenteval").joinpath("reports/templates/report.html.j2").read_text(encoding="utf-8")
+    return env.from_string(tpl_text).render(r=result, t=UI[lang])
+
+
+def render(result: ScanResult, fmt: str, lang: str = DEFAULT_LANG) -> str:
+    result = localize_result(result, lang)
+    if fmt == "json":
+        return to_json(result)
+    if fmt == "sarif":
+        return to_sarif(result)
+    if fmt == "html":
+        return to_html(result, lang)
+    raise ValueError(f"Unsupported format: {fmt} (choose from: {', '.join(FORMATS)})")

agenteval/reports/templates/certificate.html.j2

@@ -0,0 +1,178 @@
+<!doctype html>
+<html lang="{{ t.html_lang }}">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>{{ t.title }} · {{ c.id }}</title>
+<link rel="preconnect" href="https://fonts.googleapis.com">
+<link rel="stylesheet" href="https://fonts.googleapis.com/css2?family=IBM+Plex+Sans:wght@400;500;600&family=IBM+Plex+Mono:wght@400;500&family=Source+Serif+4:opsz,wght@8..60,600&display=swap">
+<style>
+  /* 证书为单一视觉：浅色纸面，固定配色以保证打印一致 */
+  :root {
+    --paper: #fbfcfd; --ink: #142036; --muted: #5a6779; --line: #cfd7e1; --soft: #eef2f6;
+    --accent: #145c4f; --accent-soft: #e3f0ec; --bad: #a32020; --bad-soft: #fbe9e9; --demo: #b42318;
+    --serif: "Source Serif 4", "Noto Serif CJK SC", "Songti SC", Georgia, serif;
+    --sans: "IBM Plex Sans", "Inter", "Noto Sans CJK SC", "PingFang SC", "Microsoft YaHei", system-ui, sans-serif;
+    --mono: "IBM Plex Mono", "DejaVu Sans Mono", ui-monospace, monospace;
+    color-scheme: light;
+  }
+  * { box-sizing: border-box; }
+  html, body { margin: 0; background: #e6eaef; color: var(--ink); font: 13px/1.45 var(--sans); }
+  .toolbar { max-width: 1062px; margin: 16px auto 0; padding-inline: 16px; display: flex; justify-content: flex-end; }
+  .toolbar button { font: 500 13px var(--sans); padding: 6px 12px; border-radius: 6px; border: 1px solid var(--line);
+                    background: #fff; color: var(--ink); cursor: pointer; }
+  .wrap { overflow-x: auto; padding: 16px; }
+  .sheet { position: relative; width: 1062px; margin: 0 auto; background: var(--paper); padding: 26px 34px 22px;
+           border: 1px solid var(--line); box-shadow: 0 1px 2px rgba(20, 32, 54, .08); overflow: hidden; }
+  .sheet::before { content: ""; position: absolute; inset: 10px; border: 1.5px solid var(--accent); pointer-events: none; }
+  .sheet::after { content: ""; position: absolute; inset: 14px; border: .5px solid var(--accent); opacity: .5; pointer-events: none; }
+  .top { display: flex; justify-content: space-between; align-items: baseline; gap: 16px; }
+  .issuer { font: 600 13px var(--sans); letter-spacing: .06em; text-transform: uppercase; color: var(--accent); }
+  .no { font: 500 12px var(--mono); color: var(--muted); }
+  h1 { font: 600 30px/1.15 var(--serif); text-align: center; margin: 14px 0 4px; letter-spacing: .01em; }
+  .certifies { text-align: center; color: var(--muted); margin: 6px 0 2px; }
+  .name { font: 600 26px/1.2 var(--serif); text-align: center; margin: 0; text-wrap: balance; }
+  .org { text-align: center; color: var(--muted); margin: 2px 0 0; }
+  .statement { text-align: center; max-width: 760px; margin: 8px auto 0; }
+  .facts { display: flex; justify-content: center; flex-wrap: wrap; gap: 6px 22px; margin-top: 8px; font-size: 12px; color: var(--muted); }
+  .facts b { color: var(--ink); font-weight: 500; font-family: var(--mono); font-size: 11.5px; }
+  .panels { display: grid; grid-template-columns: 1.25fr 1fr; gap: 16px; margin-top: 14px; }
+  .panel { border: 1px solid var(--line); border-radius: 8px; padding: 12px 14px; background: #fff; min-width: 0; }
+  .panel-head { display: flex; justify-content: space-between; align-items: flex-start; gap: 10px; margin-bottom: 8px; }
+  .panel-head h2 { font: 600 15px var(--sans); margin: 0; }
+  .panel-head p { margin: 1px 0 0; color: var(--muted); font-size: 11.5px; }
+  .big { font: 600 28px/1 var(--sans); color: var(--accent); white-space: nowrap; text-align: right; }
+  .big small { display: block; font: 400 11px var(--sans); color: var(--muted); margin-top: 3px; }
+  table { width: 100%; border-collapse: collapse; font-size: 12px; }
+  th, td { text-align: left; padding: 4px 6px; border-bottom: 1px solid var(--soft); vertical-align: top; }
+  th { color: var(--muted); font-weight: 500; font-size: 11px; }
+  td.num { font-variant-numeric: tabular-nums; white-space: nowrap; }
+  .pill { display: inline-block; padding: 0 7px; border-radius: 999px; font-size: 11px; font-weight: 600; white-space: nowrap; }
+  .pill.passed { background: var(--accent-soft); color: var(--accent); }
+  .pill.failed { background: var(--bad-soft); color: var(--bad); }
+  .pill.not_applicable, .pill.not_assessed { background: var(--soft); color: var(--muted); }
+  .meta-q { display: flex; flex-wrap: wrap; gap: 4px 14px; font-size: 11.5px; color: var(--muted); margin-bottom: 6px; }
+  .meta-q b { color: var(--ink); font-weight: 500; }
+  .bottom { display: grid; grid-template-columns: 1fr auto auto; gap: 18px; align-items: end; margin-top: 14px; }
+  .dates { display: flex; gap: 26px; }
+  .dates div span { display: block; color: var(--muted); font-size: 11px; }
+  .dates div b { font: 600 15px var(--sans); }
+  .evidence { font: 11px/1.5 var(--mono); color: var(--muted); margin-top: 6px; overflow-wrap: anywhere; }
+  .seal { width: 92px; height: 92px; border-radius: 50%; border: 2px solid var(--accent); display: grid; place-items: center;
+          text-align: center; color: var(--accent); font: 600 9.5px/1.25 var(--sans); letter-spacing: .04em; position: relative; }
+  .seal::before { content: ""; position: absolute; inset: 5px; border-radius: 50%; border: .75px solid var(--accent); }
+  .seal b { display: block; font-size: 22px; letter-spacing: 0; }
+  .qr { display: grid; justify-items: center; gap: 4px; width: 150px; }
+  .qr svg { width: 92px; height: 92px; background: #fff; }
+  .qr span { font-size: 10px; color: var(--muted); text-align: center; }
+  .verify-url { font: 9.5px var(--mono); color: var(--muted); overflow-wrap: anywhere; text-align: center; }
+  .sigline { margin-top: 10px; padding-top: 8px; border-top: 1px solid var(--soft); display: flex; justify-content: space-between;
+             gap: 16px; font-size: 10.5px; color: var(--muted); }
+  .sigline code { font: 10.5px var(--mono); color: var(--ink); }
+  .disclaimer { font-size: 10px; color: var(--muted); margin: 6px 0 0; }
+  .demo-band { margin-top: 10px; background: var(--bad-soft); color: var(--demo); border: 1px solid #f1c4c0; border-radius: 6px;
+               padding: 5px 10px; font-size: 11.5px; text-align: center; }
+  .demo-band b { letter-spacing: .04em; }
+  .watermark { position: absolute; inset: 0; display: grid; place-items: center; pointer-events: none; }
+  .watermark span { transform: rotate(-24deg); font: 700 150px/1 var(--sans); color: rgba(180, 35, 24, .07); letter-spacing: .12em; }
+  .stamp { position: absolute; top: 120px; right: 60px; transform: rotate(-12deg); border: 4px solid var(--bad); color: var(--bad);
+           font: 700 34px/1 var(--sans); padding: 8px 18px; border-radius: 8px; letter-spacing: .08em; background: rgba(255, 255, 255, .7); }
+  @page { size: A4 landscape; margin: 7mm; }
+  @media print {
+    html, body { background: #fff; }
+    .toolbar { display: none; }
+    .wrap { padding: 0; overflow: visible; }
+    .sheet { box-shadow: none; border: none; width: 1062px; zoom: .84; }
+  }
+</style>
+</head>
+<body>
+{% set s = c.subject %}{% set sec = c.assurance.security %}{% set q = c.assurance.quality %}
+{% if printable_button %}<div class="toolbar"><button type="button" onclick="window.print()">{{ t.print }}</button></div>{% endif %}
+<div class="wrap">
+<article class="sheet" aria-label="{{ t.title }}">
+  {% if c.demo %}<div class="watermark" aria-hidden="true"><span>DEMO</span></div>{% endif %}
+  {% if state != 'active' %}<div class="stamp">{{ t['state_' ~ state] }}</div>{% endif %}
+
+  <div class="top">
+    <div class="issuer">{{ c.issuer.name_zh if lang == 'zh' and c.issuer.name_zh else c.issuer.name }}</div>
+    <div class="no">{{ t.cert_no }} {{ c.id }}</div>
+  </div>
+
+  <h1>{{ t.title }}</h1>
+  <p class="certifies">{{ t.certifies }}</p>
+  <p class="name">{{ s.name_zh if lang == 'zh' and s.name_zh else s.name }}</p>
+  <p class="org">{{ s.organization }}{% if s.industry %} · {{ s.industry }}{% endif %}</p>
+  <p class="statement">{{ t.statement }}</p>
+  <div class="facts">
+    <span>{{ t.asset }} <b>{{ s.asset_id }}</b></span>
+    {% if s.model %}<span>{{ t.model }} <b>{{ s.model }}</b></span>{% endif %}
+    <span>{{ t.agent }} <b>{{ s.agent_name }}</b></span>
+  </div>
+
+  <div class="panels">
+    <section class="panel">
+      <div class="panel-head">
+        <div><h2>{{ t.security }}</h2><p>{{ t.security_sub }}</p></div>
+        <div class="big">{{ '%.1f' % sec.score }}%<small>{{ t.threshold }} &gt; {{ sec.threshold | int }}%</small></div>
+      </div>
+      <table>
+        <thead><tr><th>{{ t.tier }}</th><th>{{ t.pack }}</th><th>{{ t.checks }}</th><th>{{ t.status }}</th></tr></thead>
+        <tbody>
+        {% for p in sec.packs %}
+          <tr>
+            <td>{{ t[p.tier] if p.tier in ('national', 'regional', 'industry') else t.unknown }}</td>
+            <td>{{ p.name_zh if lang == 'zh' and p.name_zh else p.name }}</td>
+            <td class="num">{{ p.passed }} / {{ p.passed + p.failed + p.manual }}</td>
+            <td><span class="pill {{ p.status }}">{{ t['st_' ~ p.status] }}</span></td>
+          </tr>
+        {% endfor %}
+        </tbody>
+      </table>
+    </section>
+    <section class="panel">
+      <div class="panel-head">
+        <div><h2>{{ t.quality }}</h2><p>{{ t.quality_sub }}</p></div>
+        <div class="big">{{ '%.1f' % q.score }}<small>{{ t.grade }} {{ q.grade }} · {{ t.threshold }} &gt; {{ q.threshold | int }}</small></div>
+      </div>
+      <div class="meta-q">
+        <span>{{ t.exam }} <b>{{ q.exam }}@{{ q.exam_version }}</b></span>
+        <span>{{ t.cases }} <b>{{ q.cases_passed }}/{{ q.cases }}</b></span>
+        <span>{{ t.judge }} <b>{{ q.judge }}</b></span>
+      </div>
+      <table>
+        <thead><tr><th>{{ t.dimension }}</th><th>{{ t.score }}</th></tr></thead>
+        <tbody>
+        {% for d in q.dimensions %}
+          <tr><td>{{ d.name }} <span style="color:var(--muted)">· {{ (d.weight * 100) | round | int }}%</span></td>
+              <td class="num">{{ d.score if d.score is not none else '—' }}</td></tr>
+        {% endfor %}
+        </tbody>
+      </table>
+    </section>
+  </div>
+
+  {% if c.demo %}<div class="demo-band"><b>{{ t.demo }}</b> — {{ t.demo_reason }}</div>{% endif %}
+
+  <div class="bottom">
+    <div>
+      <div class="dates">
+        <div><span>{{ t.issued }}</span><b>{{ c.issued_at | date }}</b></div>
+        <div><span>{{ t.expires }}</span><b>{{ c.expires_at | date }}</b></div>
+      </div>
+      <div class="evidence">{{ t.evidence }}: {{ t.scan }} {{ sec.scan_id }} · {{ t.rules }} {{ sec.rules_version }} · {{ t.run }} {{ q.run_id }}</div>
+    </div>
+    <div class="seal" aria-hidden="true"><div>AGENTEVAL<b>✓</b>{{ 'ASSURED' if lang == 'en' else '认证' }}</div></div>
+    {% if qr %}<div class="qr">{{ qr }}<span>{{ t.verify }}</span><div class="verify-url">{{ c.verify_url }}</div></div>
+    {% else %}<div class="qr"><span>{{ t.verify_offline }}</span><div class="verify-url">agenteval cert verify {{ c.id }}.json --public-key issuer.pem</div></div>{% endif %}
+  </div>
+
+  <div class="sigline">
+    <span>{{ t.signature }} <code>{{ sig.key_id }}</code></span>
+    <span><code>{{ sig.value[:44] }}…</code></span>
+  </div>
+  <p class="disclaimer">{{ c.disclaimer_zh if lang == 'zh' else c.disclaimer }}</p>
+</article>
+</div>
+</body>
+</html>

agenteval/reports/templates/eval_report.html.j2

@@ -0,0 +1,191 @@
+<!doctype html>
+<html lang="{{ t.html_lang }}">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>{{ t.title }} · {{ r.agent_name }}</title>
+<style>
+  :root {
+    color-scheme: light;
+    --surface: #fcfcfb; --surface-2: #f1f1ef; --line: #e2e1dd;
+    --text-primary: #0b0b0b; --text-secondary: #52514e; --text-muted: #77756f;
+    --series-1: #2a78d6; --track: #e4ecf7;
+    --good: #1f7a3f; --good-bg: #e7f4ec; --critical: #b42318; --critical-bg: #fcebea; --warning: #8a5a00;
+  }
+  @media (prefers-color-scheme: dark) {
+    :root:where(:not([data-theme="light"])) {
+      color-scheme: dark;
+      --surface: #1a1a19; --surface-2: #232321; --line: #34332f;
+      --text-primary: #ffffff; --text-secondary: #c3c2b7; --text-muted: #9a988f;
+      --series-1: #3987e5; --track: #22324a;
+      --good: #6fd39b; --good-bg: #183324; --critical: #ff8a80; --critical-bg: #3a1d1b; --warning: #f0c060;
+    }
+  }
+  :root[data-theme="dark"] {
+    color-scheme: dark;
+    --surface: #1a1a19; --surface-2: #232321; --line: #34332f;
+    --text-primary: #ffffff; --text-secondary: #c3c2b7; --text-muted: #9a988f;
+    --series-1: #3987e5; --track: #22324a;
+    --good: #6fd39b; --good-bg: #183324; --critical: #ff8a80; --critical-bg: #3a1d1b; --warning: #f0c060;
+  }
+  * { box-sizing: border-box; }
+  body { margin: 0; background: var(--surface); color: var(--text-primary);
+         font: 14px/1.55 system-ui, -apple-system, "Segoe UI", "PingFang SC", sans-serif; }
+  main { max-width: 1080px; margin: 0 auto; padding: 28px 16px 56px; }
+  h1 { font-size: 22px; margin: 0 0 4px; }
+  h2 { font-size: 16px; margin: 32px 0 12px; }
+  .meta { color: var(--text-secondary); font-size: 13px; }
+  .meta span + span::before { content: " · "; color: var(--text-muted); }
+  .hero { display: grid; grid-template-columns: minmax(200px, 260px) 1fr; gap: 16px; margin-top: 20px; }
+  @media (max-width: 640px) { .hero { grid-template-columns: 1fr; } }
+  .card { background: var(--surface-2); border: 1px solid var(--line); border-radius: 10px; padding: 16px; }
+  .score { font-size: 56px; font-weight: 650; line-height: 1; }
+  .score small { font-size: 18px; color: var(--text-secondary); font-weight: 500; }
+  .badges { display: flex; gap: 8px; flex-wrap: wrap; margin-top: 12px; }
+  .badge { display: inline-flex; align-items: center; gap: 4px; padding: 2px 10px; border-radius: 999px;
+           font-weight: 600; font-size: 13px; border: 1px solid var(--line); color: var(--text-primary); }
+  .badge.good { background: var(--good-bg); color: var(--good); border-color: transparent; }
+  .badge.bad { background: var(--critical-bg); color: var(--critical); border-color: transparent; }
+  .kpis { display: grid; grid-template-columns: repeat(auto-fit, minmax(140px, 1fr)); gap: 12px; }
+  .kpi b { display: block; font-size: 22px; }
+  .kpi { color: var(--text-secondary); font-size: 13px; }
+  .kpi b { color: var(--text-primary); }
+  .bars { display: grid; gap: 10px; }
+  .bar-row { display: grid; grid-template-columns: minmax(120px, 200px) 1fr 52px; align-items: center; gap: 12px; }
+  .bar-label { font-size: 13px; }
+  .bar-label small { display: block; color: var(--text-muted); font-size: 12px; }
+  .track { height: 14px; background: var(--track); border-radius: 0 4px 4px 0; position: relative; }
+  .fill { height: 14px; background: var(--series-1); border-radius: 0 4px 4px 0; }
+  .pass-mark { position: absolute; top: -4px; bottom: -4px; width: 1px; background: var(--text-muted); }
+  .bar-val { font-variant-numeric: tabular-nums; text-align: right; font-weight: 600; }
+  .note { border-left: 3px solid var(--warning); padding: 8px 12px; background: var(--surface-2); font-size: 13px;
+          color: var(--text-secondary); margin-top: 16px; }
+  .wrap { overflow-x: auto; }
+  table { width: 100%; border-collapse: collapse; }
+  th, td { text-align: left; padding: 8px; border-bottom: 1px solid var(--line); vertical-align: top; }
+  th { font-size: 12px; color: var(--text-secondary); font-weight: 600; }
+  td.num { font-variant-numeric: tabular-nums; }
+  .ok { color: var(--good); font-weight: 600; white-space: nowrap; }
+  .ko { color: var(--critical); font-weight: 600; white-space: nowrap; }
+  .ko-msg { color: var(--critical); font-size: 12px; margin-top: 2px; }
+  details { margin-top: 6px; }
+  details summary { cursor: pointer; color: var(--text-secondary); }
+  .msg { margin: 6px 0; padding: 8px 10px; border-radius: 8px; background: var(--surface); border: 1px solid var(--line);
+         white-space: pre-wrap; word-break: break-word; }
+  .msg .who { font-size: 12px; color: var(--text-muted); display: block; }
+  .tool { font-family: ui-monospace, SFMono-Regular, Menlo, monospace; font-size: 12px; color: var(--text-secondary); }
+  .small { font-size: 12px; color: var(--text-secondary); }
+  table.cases { table-layout: fixed; min-width: 720px; }
+  table.cases td { overflow-wrap: anywhere; }
+  tr.case-row td { border-bottom: none; padding-bottom: 2px; }
+  tr.case-detail td { padding-top: 0; }
+  table.inner { table-layout: fixed; background: var(--surface); border: 1px solid var(--line); border-radius: 8px; margin: 6px 0; }
+  table.inner td { border-bottom: 1px solid var(--line); word-break: break-word; }
+  td.code { font-family: ui-monospace, SFMono-Regular, Menlo, monospace; }
+  footer { margin-top: 40px; color: var(--text-muted); font-size: 12px; border-top: 1px solid var(--line); padding-top: 12px; }
+</style>
+</head>
+<body>
+<main>
+  <h1>{{ t.title }}</h1>
+  <div class="meta">
+    <span>{{ t.paper }}: {{ r.paper_title }} ({{ r.paper_id }}@{{ r.paper_version }})</span>
+    <span>{{ t.agent }}: {{ r.agent_name }}</span>
+    <span>{{ t.judge }}: {{ r.judge }}</span>
+    <span>{{ t.run }}: {{ r.run_id }}</span>
+    <span>{{ r.started_at.strftime('%Y-%m-%d %H:%M UTC') }}</span>
+  </div>
+
+  <div class="hero">
+    <div class="card">
+      <div class="small">{{ t.overall }}</div>
+      <div class="score">{{ '%.1f' % r.overall_score }}<small> / 100</small></div>
+      <div class="badges">
+        <span class="badge">{{ t.grade }} {{ r.grade }}</span>
+        {% if r.passed %}<span class="badge good">✓ {{ t.passed }}</span>
+        {% else %}<span class="badge bad">✕ {{ t.not_passed }}</span>{% endif %}
+      </div>
+      <div class="small" style="margin-top:8px">{{ t.passing }}: {{ r.passing_score }}</div>
+    </div>
+    <div class="card kpis">
+      <div class="kpi"><b>{{ r.stats.cases_passed }} / {{ r.stats.cases }}</b>{{ t.cases }}</div>
+      <div class="kpi"><b>{{ r.stats.pass_rate }}%</b>{{ t.case_pass_rate }}</div>
+      <div class="kpi"><b>{{ r.stats.critical_failures }}</b>{{ t.critical }}</div>
+      <div class="kpi"><b>{{ r.stats.latency_p50_ms | int if r.stats.latency_p50_ms is not none else '—' }} / {{ r.stats.latency_p95_ms | int if r.stats.latency_p95_ms is not none else '—' }} ms</b>{{ t.latency }}</div>
+    </div>
+  </div>
+
+  {% if r.judge == 'heuristic' %}<p class="note">{{ t.heuristic_note }}</p>{% endif %}
+
+  <h2>{{ t.dimensions }}</h2>
+  <div class="card bars" role="table" aria-label="{{ t.dimensions }}">
+    {% for d in r.dimensions %}
+    <div class="bar-row" role="row" title="{{ d.name }}: {{ d.score if d.score is not none else t.na }} ({{ t.weight }} {{ (d.weight * 100) | round | int }}%, {{ d.items }} {{ t.items_label }})">
+      <div class="bar-label" role="cell">{{ d.name }}<small>{{ t.weight }} {{ (d.weight * 100) | round | int }}% · {{ d.items }} {{ t.items_label }}</small></div>
+      <div class="track" role="cell">
+        {% if d.score is not none %}<div class="fill" style="width: {{ d.score }}%"></div>{% endif %}
+        <div class="pass-mark" style="left: {{ r.passing_score }}%" aria-hidden="true"></div>
+      </div>
+      <div class="bar-val" role="cell">{{ d.score if d.score is not none else '—' }}</div>
+    </div>
+    {% endfor %}
+  </div>
+
+  <h2>{{ t.case_results }}</h2>
+  <div class="wrap">
+  <table class="cases">
+    <colgroup><col style="width:76px"><col><col style="width:140px"><col style="width:84px"><col style="width:64px"><col style="width:96px"></colgroup>
+    <thead><tr><th>ID</th><th>{{ t.cases }}</th><th>{{ t.category }}</th><th>{{ t.difficulty }}</th><th>{{ t.score }}</th><th>{{ t.status }}</th></tr></thead>
+    <tbody>
+    {% for c in r.cases %}
+    <tr class="case-row">
+      <td><b>{{ c.id }}</b></td>
+      <td>{{ c.title }}
+        {% if c.error %}<div class="ko-msg">{{ t.error }}: {{ c.error }}</div>{% endif %}
+        {% if c.critical_failures %}<div class="ko-msg">✕ {{ t.critical }}: {{ c.critical_failures | join(', ') }}</div>{% endif %}
+      </td>
+      <td class="small">{{ c.category }}</td>
+      <td class="small">{{ c.difficulty }}</td>
+      <td class="num"><b>{{ c.score }}</b></td>
+      <td class="{{ 'ok' if c.passed else 'ko' }}">{{ ('✓ ' ~ t.case_pass) if c.passed else ('✕ ' ~ t.case_fail) }}</td>
+    </tr>
+    <tr class="case-detail"><td></td><td colspan="5">
+        <details><summary>{{ t.transcript }}</summary>
+          {% for m in c.transcript %}
+          <div class="msg"><span class="who">{{ t.user if m.role == 'user' else t.agent_role }}{% if m.latency_ms is not none %} · {{ m.latency_ms | int }} ms{% endif %}</span>{{ m.content }}{% if m.tool_calls %}
+<span class="tool">{{ t.tools }}: {% for tc in m.tool_calls %}{{ tc.name }}({{ tc.arguments | tojson }}){% if not loop.last %}, {% endif %}{% endfor %}</span>{% endif %}</div>
+          {% endfor %}
+        </details>
+        {% if c.assertions %}
+        <details><summary>{{ t.assertions }} ({{ c.assertions | selectattr('passed') | list | length }}/{{ c.assertions | length }})</summary>
+          <table class="inner">
+            {% for a in c.assertions %}
+            <tr><td class="{{ 'ok' if a.passed else 'ko' }}" style="width:28px">{{ '✓' if a.passed else '✕' }}</td>
+                <td class="small" style="width:170px">T{{ a.turn }} · {{ a.type }}{% if a.critical %} · critical{% endif %}<br>{{ a.dimension }}</td>
+                <td class="small code">{{ a.value | tojson }}</td>
+                <td class="small" style="width:30%">{{ a.detail }}</td></tr>
+            {% endfor %}
+          </table>
+        </details>
+        {% endif %}
+        {% if c.rubric %}
+        <details><summary>{{ t.rubric }}</summary>
+          <table class="inner">
+            {% for g in c.rubric %}
+            <tr><td class="num" style="width:48px"><b>{{ g.score if g.score is not none else '—' }}</b>/5</td>
+                <td class="small" style="width:140px">{{ g.dimension }}</td>
+                <td class="small">{{ g.criterion }}<br><i>{{ g.rationale }}</i></td></tr>
+            {% endfor %}
+          </table>
+        </details>
+        {% endif %}
+    </td></tr>
+    {% endfor %}
+    </tbody>
+  </table>
+  </div>
+
+  <footer>{{ t.disclaimer }}</footer>
+</main>
+</body>
+</html>