app/components/reference.py

"""Reference events panel + prediction details."""

from __future__ import annotations

import html

import streamlit as st

from app.styles import TOKENS, domain_color, severity_color
from src.models.schemas import PredictionResult


def _render_one_reference(entry: dict) -> str:
    """Render one reference row as HTML.

    ``entry`` follows PredictionResult.reference_info shape:
    ``{event_id, similarity[, country, date, num_cascade_nodes, top_domains]}``.
    Optional fields are omitted from the rendered secondary line when missing
    (fallback mode uses only event_id + similarity).
    """
    event_id = entry.get("event_id", "?")
    similarity = entry.get("similarity") or 0
    pct = int(round(similarity * 100))

    # Secondary metadata line: country · date · num_cascade_nodes
    meta_bits: list[str] = []
    country = entry.get("country")
    date_ = entry.get("date")
    if country and date_:
        meta_bits.append(f"{html.escape(str(country))} · {html.escape(str(date_))}")
    elif country:
        meta_bits.append(html.escape(str(country)))
    elif date_:
        meta_bits.append(html.escape(str(date_)))
    n_nodes = entry.get("num_cascade_nodes")
    if isinstance(n_nodes, int):
        meta_bits.append(f"{n_nodes} cascade nodes")
    meta_line = (
        f'<div class="node-meta" style="margin:2px 0 6px 0">{ " · ".join(meta_bits) }</div>'
        if meta_bits else ""
    )

    # top_domains pills (reuse the DAG legend style)
    top_domains = entry.get("top_domains") or []
    pills = ""
    if top_domains:
        pill_html = "".join(
            f'<span class="domain-pill" style="font-size:10px;padding:2px 8px">'
            f'<span class="swatch" style="background:{domain_color(d)}"></span>'
            f'{html.escape(str(d))}</span>'
            for d in top_domains
        )
        pills = (
            f'<div style="display:flex;flex-wrap:wrap;gap:6px;'
            f'margin:0 0 6px 0">{pill_html}</div>'
        )

    return (
        f'<div class="ref-row" style="flex-direction:column;'
        f'align-items:stretch;gap:4px">'
        f'  <div style="display:flex;align-items:center;gap:14px">'
        f'    <span class="ref-id">{html.escape(event_id)}</span>'
        f'    <span class="ref-meter">'
        f'      <span class="ref-meter-fill" style="width:{pct}%"></span>'
        f'    </span>'
        f'    <span class="ref-pct">{pct}%</span>'
        f'  </div>'
        f'  {meta_line}'
        f'  {pills}'
        f'</div>'
    )


def render_reference_events(result: PredictionResult) -> None:
    """Historical analogues with similarity meters + enriched provenance.

    Consumes ``result.reference_info`` when populated (post-v2 schema). Falls
    back to event_id + similarity pairs when the chain index was unavailable
    at predict time, so older callers/artifacts still render correctly.
    """
    if not result.reference_event_ids and not result.reference_info:
        st.markdown(
            '<div class="node-meta">No historical analogues retrieved.</div>',
            unsafe_allow_html=True,
        )
        return

    # Prefer enriched info when present; otherwise synthesize minimal entries
    # from the parallel lists so rendering stays uniform.
    if result.reference_info:
        entries = result.reference_info
    else:
        entries = [
            {"event_id": eid,
             "similarity": result.similarity_scores[i]
                           if i < len(result.similarity_scores) else 0}
            for i, eid in enumerate(result.reference_event_ids)
        ]

    rows = "".join(_render_one_reference(e) for e in entries)

    st.markdown(
        f'<div class="panel" style="padding:6px 16px 4px 16px">{rows}</div>',
        unsafe_allow_html=True,
    )
    st.markdown(
        '<div class="node-meta" style="margin-top:8px">'
        'RAG · MiniLM-L6 · cosine similarity'
        '</div>',
        unsafe_allow_html=True,
    )


def render_prediction_details(result: PredictionResult) -> None:
    """Per-node detail cards."""
    chain = result.predicted_chain
    if not chain.cascade_events:
        return

    for node in chain.cascade_events:
        confidence = result.confidence_scores.get(node.id, 0)
        sev_col = severity_color(node.severity)
        dcol = domain_color(node.domain)

        # Plain-text label (Streamlit doesn't render HTML here)
        label = (
            f"{node.id}  ·  {node.description[:72]}"
            f"{'…' if len(node.description) > 72 else ''}"
            f"   [{node.severity.upper()}]"
        )

        with st.expander(label, expanded=False):
            meta = (
                f'<div class="node-meta">'
                f'<span style="color:{dcol}">● {node.domain}</span>'
                f' &nbsp;·&nbsp; <span style="color:{sev_col}">'
                f'SEVERITY {node.severity.upper()}</span>'
                f' &nbsp;·&nbsp; +{node.time_offset_hours}h'
            )
            if confidence:
                meta += f' &nbsp;·&nbsp; CONF {confidence:.0%}'
            if node.parent_ids:
                joint_label = (
                    f' <span style="color:{TOKENS["amber"]}">⋈ JOINT</span>'
                    if len(node.parent_ids) >= 2 else ""
                )
                meta += (
                    f' &nbsp;·&nbsp; CAUSED BY {", ".join(node.parent_ids)}'
                    f'{joint_label}'
                )
            meta += "</div>"
            st.markdown(meta, unsafe_allow_html=True)

            st.markdown(
                f'<div style="font-family:Figtree;font-size:14.5px;'
                f'color:{TOKENS["text"]};line-height:1.6;margin-top:10px">'
                f'<span style="color:{TOKENS["text_muted"]};font-family:JetBrains Mono;'
                f'font-size:10px;letter-spacing:0.18em;text-transform:uppercase;'
                f'display:block;margin-bottom:4px">Mechanism</span>'
                f'{node.mechanism}'
                f'</div>',
                unsafe_allow_html=True,
            )