"""HTML renderer: a self-contained dashboard page (no server, no deps).
Generates one index.html with sortable metric tables, per-benchmark bars,
per-sample drill-down and error browser. Consumed by CLI --out-dir.
"""
import html
import json
from typing import Dict, List, Union
from ...eval.record import EvalReport
from .. import register_renderer
@register_renderer('html')
def html_dashboard(target: Union[EvalReport, List[EvalReport]], opts: Dict) -> str:
reports = target if isinstance(target, list) else [target]
reports = [r for r in reports if isinstance(r, EvalReport)]
cards = []
for rep in reports:
cards.append(_card(rep))
rows_json = json.dumps([_row(r) for r in reports], ensure_ascii=False)
return _PAGE.replace('__ROWS__', rows_json).replace('__CARDS__', '\n'.join(cards))
def _row(rep: EvalReport) -> Dict:
primary = next((k for k in rep.metrics if k != 'extraction_failure_rate'), '')
return {
'dataset': rep.dataset, 'recipe': rep.recipe, 'model': rep.model,
'n': rep.num_samples, 'primary': primary,
'value': rep.metrics.get(primary, 0.0),
'metrics': {k: v for k, v in rep.metrics.items()},
'groups': {k: v for k, v in rep.metric_groups.items() if isinstance(v, dict)},
'extract_fail': rep.num_failed_extractions,
'samples': [
{'id': s.sample_id, 'ok': s.extraction_ok, 'extracted': s.extracted_prediction[:200],
'raw': s.raw_prediction[:300], 'target': str(s.target)[:120],
'scores': s.scores, 'error': s.error[:200]}
for s in rep.samples[:200]
],
}
def _card(rep: EvalReport) -> str:
primary = next((k for k in rep.metrics if k != 'extraction_failure_rate'), '')
value = rep.metrics.get(primary, 0.0)
pct = f'{value * 100:.1f}%'
warn = (f'extraction failures: {rep.num_failed_extractions}'
f'/{rep.num_samples}') if rep.num_failed_extractions else ''
return (f'
{html.escape(rep.dataset)}
'
f'
{pct}
{html.escape(primary)} · n={rep.num_samples}'
f' · {html.escape(rep.model or "?")}
{warn}
')
_PAGE = """
EvalHarness Dashboard
EvalHarness Dashboard generated by evalharness viz