{# FI-3: top-aligned-papers fragment — lazy-loaded into #insights-top-papers.

   Context:
     learning — the _learning_state() dict. Relevant keys:
                available  — bool; False on first run / a bad interest vector.
                top_papers — list of {doi, title, cosine} dicts (may be empty).

   Three states:
     1. available + papers  → a small table (title → /corpus/{doi}, cosine).
     2. available + empty   → "no aligned papers" notice (vector needs embeddings).
     3. not available       → "engine hasn't learned" notice.

   DOI embedded literally in the /corpus/{doi:path} link (verbatim, no urlencode
   — matches the corpus detail route). All content via Jinja autoescape only —
   never |safe. Em-dash placeholders use the block form only.
#}
<div class="card" id="insights-top-papers-card">
  <h2>Top aligned papers</h2>
  {% if learning.available %}
    {% if learning.top_papers %}
  <p class="field-note">
    Library papers most aligned with your interest vector (cosine similarity).
  </p>
  <table class="papers-table">
    <thead>
      <tr><th>Paper</th><th>Cosine</th></tr>
    </thead>
    <tbody>
      {% for p in learning.top_papers %}
      <tr>
        <td>
          {% if p.doi %}
          <a href="/corpus/{{ p.doi }}">{% if p.title %}{{ p.title }}{% else %}{{ p.doi }}{% endif %}</a>
          {% elif p.title %}{{ p.title }}
          {% else %}&mdash;{% endif %}
        </td>
        <td><span class="score">{{ "%.2f" | format(p.cosine | float) }}</span></td>
      </tr>
      {% endfor %}
    </tbody>
  </table>
    {% else %}
  <p class="field-note">
    No aligned papers yet — the interest vector needs library embeddings. Run
    <code>lit-monitor brain-build</code> to embed the corpus.
  </p>
    {% endif %}
  {% else %}
  <p class="field-note">
    The engine hasn't learned anything yet — give feedback on discoveries to
    build an interest vector.
  </p>
  {% endif %}
</div>