{% extends "base.html" %}
{% block content %}
<h1>TES Baseline Status</h1>

<div class="caveat">
  Self-baseline verdicts are <strong>relative to YOUR OWN lean waste-free sessions of each type</strong> —
  not an absolute efficiency standard. "within_band" means comparable to your leaner sessions of that type,
  not that your work is universally efficient.
</div>

{% if headline %}
<h2>Coverage — Content Sessions</h2>
<div class="metric-box">
  <table style="width:auto; margin:0;">
    <tr>
      <td style="padding-right:2rem;"><strong>Sessions with content (turn_count &gt; 0)</strong></td>
      <td><strong>{{ headline.content_sessions }}</strong></td>
    </tr>
    <tr>
      <td><strong>Scored vs your lean runs (self-baseline)</strong></td>
      <td><span class="metric-highlight">{{ headline.content_self_scored }} ({{ headline.content_self_pct }}%)</span></td>
    </tr>
    <tr>
      <td>Out of scope — below scope floor</td>
      <td>{{ headline.content_sessions - headline.content_self_scored }}</td>
    </tr>
    <tr>
      <td>Pre-self-baseline (B2 corpus)</td>
      <td><span class="metric-old">{{ headline.b2_content_scored }} of {{ headline.b2_content_total }} ({{ headline.b2_content_scored_pct }}%)</span></td>
    </tr>
    <tr style="border-top:1px solid #ddd;">
      <td style="padding-top:0.5rem;"><em>Empty/stub sessions (no work product)</em></td>
      <td style="padding-top:0.5rem;"><em>{{ headline.empty_stubs }} ({{ headline.empty_stubs_pct }}% of {{ headline.total }} total)</em></td>
    </tr>
  </table>
</div>
{% endif %}

{% if status_rows %}
<h2>Per-Type Status</h2>
<table>
  <thead>
    <tr>
      <th>Task Type</th>
      <th>Total</th>
      <th>In-Scope</th>
      <th>Waste-Free</th>
      <th>Lean-N</th>
      <th>Scope Floor</th>
      <th>Status</th>
      <th>Baseline Band</th>
    </tr>
  </thead>
  <tbody>
    {% for row in status_rows %}
    <tr>
      <td><strong>{{ row.task_type }}</strong></td>
      <td>{{ row.total }}</td>
      <td>{{ row.in_scope }}</td>
      <td>{{ row.waste_free_n }}</td>
      <td>{{ row.lean_n }}</td>
      <td>{{ row.scope_floor }} turns</td>
      <td>
        {% if row.source == "self" %}
          <span class="tag-self">ACTIVE (self)</span>
        {% elif row.source == "building" %}
          <span class="tag-building">building — {{ row.sessions_needed }} more lean sessions needed</span>
        {% else %}
          <span class="tag-corpus">corpus fallback</span>
        {% endif %}
      </td>
      <td>
        {% if row.p25 is not none and row.p75 is not none %}
          {{ "{:,}".format(row.p25) }} – {{ "{:,}".format(row.p75) }}
          (median {{ "{:,}".format(row.median) }})
        {% else %}
          <span class="tag-unavailable">—</span>
        {% endif %}
      </td>
    </tr>
    {% endfor %}
  </tbody>
</table>

<h2>Domain of Validity per Type</h2>
{% for row in status_rows %}
<h3 style="font-size:0.9rem; margin-top:1rem;">{{ row.task_type }}</h3>
<div class="caveat">{{ row.domain_of_validity }}</div>
{% endfor %}

{% if waste_by_type %}
<h2>Waste Concentration</h2>
{% set total_events = waste_by_type | sum(attribute='total_events') %}
{% if total_events > 0 %}
<p class="caveat">
  Detectable waste (observable-invariant only: retry loops + redundant reads) is concentrated in
  <strong>ml-eval, infra-deploy, and research-recon</strong> — the infrastructure/ML-ops types where
  identical-failure retries and redundant file reads occur.
  {% for row in waste_by_type if row.total_events > 0 %}
    <strong>{{ row.task_type }}</strong>: {{ row.total_events }} event(s) in {{ row.with_waste }} of {{ row.total }} sessions;
  {% endfor %}
  feature-build ({{ waste_by_type | selectattr('task_type', 'eq', 'feature-build') | map(attribute='total') | first | default(0) }} sessions): zero waste events.
  Total: {{ total_events }} events across all types.
  See <a href="https://github.com/gaurav-gandhi-2411/token-efficiency-scorer/blob/master/research/11-generalization.md">report 11</a> for corpus characterisation.
</p>
{% else %}
<p class="caveat">
  No waste events in store — run <code>tes backfill-waste</code> to detect and record waste across all sessions.
  Expected concentration (from inventory): ml-eval &gt; infra-deploy &gt; research-recon; zero in feature-build.
</p>
{% endif %}
{% endif %}

{% else %}
<p class="empty">No baseline data available. Run <code>tes score</code> to accumulate sessions.</p>
{% endif %}

{% endblock %}