render.py

"""HTML report for System One vs Laya vs DeepSeek comparison."""
from __future__ import annotations

import html


def _esc(s) -> str:
    return html.escape(str(s) if s is not None else "")


def _bar(probs, options, chosen_index, fill_class="bar-fill"):
    parts = []
    for i, (p, label) in enumerate(zip(probs or [], options or [])):
        w = max(2, int(float(p) * 100))
        hot = i == chosen_index
        parts.append(
            f'<div class="bar-row{" hot" if hot else ""}">'
            f'<span class="bar-label">{_esc(label)}</span>'
            f'<span class="bar-track"><span class="{fill_class}" style="width:{w}%"></span></span>'
            f'<span class="bar-pct">{float(p) * 100:.1f}%</span></div>'
        )
    return "".join(parts)


def _choice_of(block: dict | None) -> tuple[str, float]:
    if not block:
        return "—", 0.0
    return str(block.get("chosen") or "—"), float(block.get("conf") or 0.0)


def _norm(s: str | None) -> str:
    return (s or "").strip().lower()


def _agree_pill(a: str | None, b: str | None, label: str) -> str:
    if not a or not b or a == "—" or b == "—":
        return f'<span class="pill muted">{_esc(label)}: —</span>'
    if _norm(a) == _norm(b):
        return f'<span class="pill ok">{_esc(label)}: match</span>'
    return f'<span class="pill warn">{_esc(label)}: differ</span>'


def render_report(jev: dict, laya: dict, deepseek: dict, qs: list[dict]) -> str:
    from deepseek_lane import answers_by_id

    ds_answers = answers_by_id(deepseek.get("validation") or {})
    jev_qs = {b["id"]: b for b in (jev.get("questions") or [])}
    laya_qs = {b["id"]: b for b in (laya.get("questions") or [])}

    all3 = 0
    comparable = 0
    rows = []
    for q in qs:
        jid = q["id"]
        jb = jev_qs.get(jid)
        lb = laya_qs.get(jid)
        j_choice, j_conf = _choice_of(jb)
        l_choice, l_conf = _choice_of(lb)
        ds_val = ds_answers.get(jid)

        picks = [p for p in (j_choice, l_choice, ds_val) if p and p != "—"]
        if len(picks) >= 2:
            comparable += 1
            if len({_norm(p) for p in picks}) == 1 and len(picks) == 3:
                all3 += 1

        j_bars = _bar(jb["probs"], jb["options"], jb["chosen_index"]) if jb else ""
        l_bars = (
            _bar(lb["probs"], lb["options"], lb["chosen_index"], fill_class="bar-fill laya")
            if lb
            else ""
        )

        ds_status = ""
        for a in (deepseek.get("validation") or {}).get("answers") or []:
            if a["id"] == jid:
                ds_status = a["status"]
                break

        pills = " ".join(
            [
                _agree_pill(j_choice, l_choice, "S1↔Laya"),
                _agree_pill(j_choice, ds_val, "S1↔DS"),
                _agree_pill(l_choice, ds_val, "Laya↔DS"),
            ]
        )

        jev_err = ""
        laya_body = l_bars
        if laya.get("error") and not lb:
            laya_body = f'<p class="meta err">{_esc(laya.get("error"))}</p>'

        rows.append(
            f'<article class="q-card">'
            f'<h3>{_esc(jid)} · {_esc(q["type"])}</h3>'
            f'<p class="q-text">{_esc(q["question"])}</p>'
            f'<div class="cols3">'
            f'<div class="lane jev"><div class="lane-title">System One</div>'
            f'{j_bars or jev_err}'
            f'<p class="pick">→ <b>{_esc(j_choice)}</b> ({j_conf * 100:.1f}%)</p></div>'
            f'<div class="lane laya"><div class="lane-title">Laya</div>'
            f'{laya_body}'
            f'<p class="pick">→ <b>{_esc(l_choice)}</b> ({l_conf * 100:.1f}%)</p></div>'
            f'<div class="lane ds"><div class="lane-title">DeepSeek (API)</div>'
            f'<p class="pick">→ <b>{_esc(ds_val or "—")}</b>'
            f' <span class="status">{_esc(ds_status)}</span></p>'
            f'<div class="pills">{pills}</div></div>'
            f"</div></article>"
        )

    ratio = f"{all3}/{comparable}" if comparable else "—"
    plan = jev.get("plan") or {}
    warn = ""
    if plan.get("beyond_train_len"):
        warn += (
            '<p class="warn">State exceeds System One training length (384 tokens) — '
            "treat S1 probabilities as exploratory.</p>"
        )
    if plan.get("truncated"):
        warn += '<p class="warn">State was truncated to fit the System One scorer limit.</p>'
    if jev.get("error"):
        warn += f'<p class="warn">System One: {_esc(jev["error"])}</p>'
    if laya.get("error"):
        warn += f'<p class="warn">Laya: {_esc(laya["error"])}</p>'

    ds_err = deepseek.get("error")
    ds_meta = (
        f'<p class="meta err">DeepSeek: {_esc(ds_err)}</p>'
        if ds_err
        else (
            f'<p class="meta">DeepSeek: {_esc(deepseek.get("route", ""))} · '
            f'valid JSON {deepseek.get("validation", {}).get("n_valid", 0)}/{len(qs)} · '
            f'Laya: {_esc(laya.get("checkpoint", "—"))}</p>'
        )
    )

    return f"""
<div class="jev-report">
  <div class="stats">
    <div class="stat y"><div class="n">{jev.get("total_ms", 0):.0f} ms</div><div class="l">System One</div></div>
    <div class="stat g"><div class="n">{laya.get("total_ms", 0):.0f} ms</div><div class="l">Laya</div></div>
    <div class="stat b"><div class="n">{deepseek.get("latency_ms", 0):.0f} ms</div><div class="l">DeepSeek API</div></div>
    <div class="stat p"><div class="n">{ratio}</div><div class="l">All-3 agree</div></div>
  </div>
  {warn}
  {ds_meta}
  {"".join(rows)}
</div>
"""


def render_error(msg: str) -> str:
    return f'<div class="jev-report"><p class="meta err">{_esc(msg)}</p></div>'