{% extends "base.html" %}
{% from "_help.html" import info %}
{% from "_job_status.html" import job_status %}
{% block title %}Index — Paperful{% endblock %}
{% block body_attrs %}{% if poll_run %} data-poll-run="{{ poll_run }}"{% endif %}{% endblock %}
{% block content %}
<h1>Index</h1>
{% if error %}<div class="error-banner">{{ error }}</div>{% endif %}
<p class="page-lede">{{ message }} The index is built from PDFs already copied to disk under <code>out/</code>, not live from Zotero. Ingest first, then Search here or <a href="/ask">Ask</a>.</p>
{% if flags.remote_llm %}<p class="warn-remote">Remote LLM — excerpts from your PDFs leave this machine.</p>{% endif %}
{% if flags.remote_embed %}<p class="warn-remote">Remote embedding model — passage text leaves this machine.</p>{% endif %}
{% if enabled %}
<section class="panel">
  <details>
    <summary class="panel-summary">Status</summary>
    <p class="section-lede">How many items and chunks are in the local index. If this is empty, run Ingest.</p>
    <table class="wb-table">
      <thead><tr><th>Field</th><th>Value</th></tr></thead>
      <tbody>
      {% for k, v in status.items() %}
      <tr><td>{{ k }}</td><td>{{ v }}</td></tr>
      {% endfor %}
      </tbody>
    </table>
  </details>
</section>
{% endif %}

{% if flags.can_ingest %}
<section class="panel">
  <h2>Ingest</h2>
  <p class="section-lede">Read PDFs (and abstracts) from the mirror into the search index. Dry-run counts what would change. Unchanged files are skipped unless you come back after they change on disk.</p>
  {{ job_status(ingest_result, 'Ingest') }}
  {% if ingest_result.summary %}
  <p>{% if ingest_result.dry_run %}Would index{% else %}Indexed{% endif %}
    {{ ingest_result.summary.pdf + ingest_result.summary.ocr }} PDFs
    ({{ ingest_result.summary.ocr }} after OCR) and {{ ingest_result.summary.abstract }} abstracts.
    Unchanged {{ ingest_result.summary.unchanged }}, failed {{ ingest_result.summary.failed }}.
    pdf {{ ingest_result.summary.pdf }}, chunks {{ ingest_result.summary.chunks }}.</p>
  {% endif %}
  <form method="post" action="/index/ingest" class="toolbar">
    <label>Year from {{ info("Only items dated this year or later. Undated items are excluded.") }} <input type="number" name="year_from"></label>
    <label>Year to {{ info("Only items dated this year or earlier. Undated items are excluded.") }} <input type="number" name="year_to"></label>
    <label>Limit {{ info("Stop after this many items. Leave blank to ingest the whole scope.") }} <input type="number" name="limit" min="1"></label>
    <button type="submit" name="dry_run" value="1">Dry-run ingest</button>
    {{ info("Count what would be indexed. Writes nothing.") }}
    <button type="submit" name="dry_run" value="0">Ingest</button>
    {{ info("Write new or changed PDFs and abstracts into the local index.") }}
  </form>
</section>
{% endif %}

{% if flags.can_search %}
<section class="panel">
  <h2>Search</h2>
  <p class="section-lede">Find passages in the indexed PDFs. This is retrieval only — it does not call the chat model. <a href="/ask">Ask</a> uses the same index to answer in sentences.</p>
  <form method="post" action="/index/search" class="toolbar">
    <label>Query {{ info("Natural-language search over indexed passages, not a keyword-only library search.") }} <input name="q" value="{{ search_q }}"></label>
    <label>Top-k {{ info("How many matching passages to return. Higher values show more context and more noise.") }} <input name="k" value="{{ search_k or '' }}"></label>
    <label>Year from {{ info("Only items dated this year or later. Undated items are excluded.") }} <input name="year_from" value="{{ year_from }}"></label>
    <label>Year to {{ info("Only items dated this year or earlier. Undated items are excluded.") }} <input type="number" name="year_to" value="{{ year_to }}"></label>
    <button type="submit">Search</button>
    {{ info("Retrieve matching passages. Does not call the chat model.") }}
  </form>
  {% if hits %}
  <table class="wb-table">
    <thead><tr><th>Score</th><th>Title</th><th>Excerpt</th></tr></thead>
    <tbody>
    {% for h in hits %}
    <tr><td>{{ h.score }}</td><td>{{ h.title }} [{{ h.item_key }}]</td><td>{{ h.text }}</td></tr>
    {% endfor %}
    </tbody>
  </table>
  {% endif %}
</section>
{% endif %}

{% if flags.llm_on and flags.rag_on %}
<section class="panel">
  <h2>Extract questions</h2>
  <p class="section-lede">Pull research questions out of indexed papers. The default is a rules pass over headings and abstracts. The LLM extract lane also asks the model, which costs tokens and may send excerpts off this machine if the model is remote.</p>
  {{ job_status(questions_result, 'Extract questions') }}
  {% if questions_result.questions is defined %}<p>Last: {{ questions_result.questions }} question(s) from {{ questions_result.items }} item(s){% if questions_result.dry_run %} (dry-run){% endif %}</p>{% endif %}
  <form method="post" action="/index/rag-questions" enctype="multipart/form-data" class="form-stack">
    {% include "_rag_scope_fields.html" %}
    <div class="field-group">
      <p class="field-legend">Extract {{ info("Rules first; optionally a grounded LLM pass. Dry-run counts without writing.") }}</p>
      <label><input type="checkbox" name="whole_library" value="1"> Whole library (when no collection chip) {{ info("Use every indexed item when the header has no collection. Ignored when a collection chip is set or you listed item keys.") }}</label>
      <label><input type="checkbox" name="use_llm" value="1" {% if extract_questions_llm %}checked{% endif %}> LLM extract lane {{ info("Also ask the model to extract research questions from each paper, grounded in retrieved passages. Costs tokens. Remote models send excerpts off this machine. The rules pass still runs.") }}</label>
      <label>Limit {{ info("Max papers this run. Leave blank to use the whole scope.") }} <input type="number" name="limit" min="1"></label>
      <button type="submit" name="dry_run" value="1">Dry-run</button>
      {{ info("Count questions that would be written. Does not save them.") }}
      <button type="submit" name="dry_run" value="0">Extract</button>
      {{ info("Write extracted questions under state/rag/questions/.") }}
    </div>
  </form>
</section>

<section class="panel">
  <h2>Already answered?</h2>
  <p class="section-lede">Check whether the indexed corpus already answers a list of questions. Useful before a new study: you get answered / partial / not found rather than a full write-up.</p>
  {{ job_status(answered_result, 'Answered check') }}
  {% if answered_result.stamp %}<p>Last: answered {{ answered_result.answered }} · partial {{ answered_result.partial }} · not found {{ answered_result.not_found }} <a href="/index/answered/{{ answered_result.stamp }}">report</a></p>{% endif %}
  <form method="post" action="/index/rag-answered" enctype="multipart/form-data" class="form-stack">
    <div class="field-group">
      <p class="field-legend">Questions {{ info("Pick one source of questions. The form, an upload, or questions already extracted above.") }}</p>
      <label><input type="radio" name="question_mode" value="text" checked> Questions in form {{ info("Use the textarea below, one question per line.") }}</label>
      <label><input type="radio" name="question_mode" value="upload"> Upload file {{ info("Use the uploaded .md or .txt file, one question per line.") }}</label>
      <label><input type="radio" name="question_mode" value="extract"> From extracted questions {{ info("Use questions previously written by Extract questions, for items in this scope.") }}</label>
      <label>Questions {{ info("One question per line. Used when Questions in form is selected.") }} <textarea name="answered_questions" rows="3" placeholder="One question per line"></textarea></label>
      <label>Questions file {{ info("Used when Upload file is selected.") }} <input type="file" name="answered_upload" accept=".md,.txt,text/plain"></label>
    </div>
    {% include "_rag_scope_fields.html" %}
    <div class="field-group">
      <p class="field-legend">Check {{ info("Walk the index and classify each question as answered, partial, or not found.") }}</p>
      <label>After item {{ info("Resume after this Zotero item key. Leave blank to start at the beginning of the scope.") }} <input name="after_item" placeholder="item key"></label>
      <label><input type="checkbox" name="force" value="1"> Force {{ info("Ignore cached answered-check rows and classify every question again.") }}</label>
      <button type="submit">Run answered check</button>
      {{ info("Classify each question against the index. Spends tokens when the LLM lane is on.") }}
    </div>
  </form>
  {% if answered_packs %}
  <details>
    <summary>Recent answered reports ({{ answered_packs|length }})</summary>
    <ul class="counts">{% for p in answered_packs %}<li><a href="/index/answered/{{ p.stamp }}">{{ p.stamp }}</a> — {{ p.answered }}/{{ p.partial }}/{{ p.not_found }}{% if p.failed %} ({{ p.failed }} failed){% endif %}</li>{% endfor %}</ul>
  </details>
  {% endif %}
</section>

<section class="panel">
  <h2>Synthesize report</h2>
  <p class="section-lede">Fold existing summary notes into a literature-review report for the current collection. Preview (dry-run) shows source counts and the chunk plan without calling the model.</p>
  {% if synthesize_result.status == 'queued' or synthesize_result.status == 'running' %}
  <p class="counts" role="status" {% if synthesize_result.id %}data-run-id="{{ synthesize_result.id }}"{% endif %}>{% if synthesize_result.args.dry_run %}Preview{% else %}Write report{% endif %} <span class="cmd-status">{{ synthesize_result.status }}</span>…</p>
  {% elif synthesize_result.skip_plain %}
  <p class="warn-remote">{{ synthesize_result.skip_plain }}</p>
  {% elif synthesize_result.slug %}
  <p>Last: {% if synthesize_result.dry_run or synthesize_result.args.dest == 'zotero' %}{{ synthesize_result.slug }}{% else %}<a href="/index/report/{{ synthesize_result.slug }}">{{ synthesize_result.slug }}</a>{% endif %} {% if synthesize_result.skipped %}({{ synthesize_result.skipped }}){% endif %}</p>
  {% elif synthesize_result.error %}
  <p class="error-banner">{{ synthesize_result.error }}</p>
  {% endif %}
  <form method="post" action="/index/synthesize" class="toolbar">
    <label>Dest {{ info("disk: HTML under out/reports/ only. zotero: a collection note. both: files and a note.") }} <select name="dest">{% for d in synth_dest_options %}<option value="{{ d }}" {% if d == synth_dest %}selected{% endif %}>{{ d }}</option>{% endfor %}</select></label>
    <label>Year from {{ info("Only summaries dated this year or later.") }} <input type="number" name="year_from"></label>
    <label>Year to {{ info("Only summaries dated this year or earlier.") }} <input type="number" name="year_to"></label>
    <label>Limit {{ info("Max summary notes to include. Leave blank for the whole scope.") }} <input type="number" name="limit" min="1"></label>
    <label><input type="checkbox" name="force" value="1"> Force {{ info("Regenerate even when a saved report used the same summary notes.") }}</label>
    <button type="submit" name="dry_run" value="1"{% if synthesize_result.status == 'queued' or synthesize_result.status == 'running' %}{% if synthesize_result.args.dry_run %} data-poll-button{% endif %}{% endif %}>Preview</button>
    {{ info("Show source counts and the chunk plan. No model call.") }}
    <button type="submit" name="dry_run" value="0"{% if synthesize_result.status == 'queued' or synthesize_result.status == 'running' %}{% if not synthesize_result.args.dry_run %} data-poll-button{% endif %}{% endif %}>Write report</button>
    {{ info("Call the model and write the report to Dest.") }}
  </form>
  {% if reports %}
  <details>
    <summary>Recent reports ({{ reports|length }})</summary>
    <ul>{% for r in reports %}<li><a href="/index/report/{{ r.name }}" target="_blank">{{ r.name }}</a></li>{% endfor %}</ul>
  </details>
  {% endif %}
</section>
{% endif %}
{% endblock %}