ResearchIT / scripts /hf_comparison.html
siddhm11
Claude Opus 5.5
feat(discovery): CLI for collection, baseline capture and comparison
245fdd3
Raw History Blame Contribute Delete
3.04 kB
<!doctype html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>ResearchIT · HF comparison</title><style>
body{font:16px/1.6 system-ui;margin:0;color:#173b32;background:#f5f6ee}header{background:#173b32;color:white;padding:32px}main{padding:24px;max-width:1250px;margin:auto}h1{margin:8px 0}h2{font-size:21px}.notice{padding:18px;background:#fff0cf;border-left:4px solid #bd8a25;margin-bottom:22px}.columns{display:grid;grid-template-columns:repeat(3,1fr);gap:18px}.card{padding:17px;background:white;border:1px solid #d9e1d5;border-radius:8px;margin-bottom:12px}small{color:#54695d}.badge{font-size:11px;padding:3px 7px;background:#dfeee0;border-radius:4px}a{color:#27694a}details{padding:18px;background:white;margin-top:20px}pre{white-space:pre-wrap;overflow-wrap:anywhere}footer{font-size:13px;padding:24px}@media(max-width:760px){.columns{grid-template-columns:1fr}}
</style></head><body><header><small style="color:#d4e7dc">PRIVATE COMPARISON · SERVING FEED UNCHANGED</small><h1>{{ report.get('profile','Reader') }}: what would change?</h1><p>{{ report.get('as_of','') }}</p></header><main>
<div class="notice"><strong>{{ 'Synthetic demonstration — no real relevance result.' if report.get('synthetic') else 'Unjudged experiment — not evidence of improved relevance.' }}</strong><br>{{ report.get('method','') }}<br>Up to {{ report.source_cap }} HF replacements, within the same assigned interest. Threshold and source cap are experimental.</div>
<div class="columns">{% for key,label in [('baseline','Current baseline'),('hf_only','HF candidates only'),('combined','Combined experiment')] %}<section><h2>{{ label }}</h2>{% for p in report[key] %}<article class="card"><span class="badge">{{ p.origin }} · Interest {{ p.interest+1 }}</span><h3>{{ loop.index }}. {{ p.title }}</h3><p>{{ p.reason }}</p><small>Topic similarity {{ '%.3f'|format(p.similarity) }} · Score {{ '%.3f'|format(p.score) }}</small><p>{% if report.get("synthetic") %}<small>Simulated ID: {{ p.arxiv_id }}</small>{% else %}<a href="https://arxiv.org/abs/{{ p.arxiv_id|urlencode }}" rel="noopener" target="_blank">{{ p.arxiv_id }}</a>{% endif %}</p></article>{% else %}<p>No eligible candidates. This is not a successful relevance result.</p>{% endfor %}</section>{% endfor %}</div>
<details open><summary>Exactly what changed</summary>{% for c in report.changes %}<p>{{ c.removed }} → <strong>{{ c.added }}</strong> · Interest {{ c.interest+1 }}<br>{{ c.reason }}</p>{% else %}<p>No replacements. Source failure, stale data, missing embeddings or poor topic fit can all lead to the baseline remaining unchanged.</p>{% endfor %}</details><details><summary>Rejected candidates and source health</summary><pre>{{ report.rejected|tojson(indent=2) }}</pre><pre>{{ report.get('store',{})|tojson(indent=2) }}</pre></details></main><footer>Local shadow report. No impressions, clicks or user-profile updates were recorded. Human review is required before a quality claim.</footer></body></html>