research/workspace.py

"""Research workspace: turn case notes into analytic artifacts (the sandbox).

The tiny head reasons; the workspace materializes. Given a case ledger (NOTE
lines, search hits, verdicts) it renders, saves, and returns markdown documents:
  - timeline      events with dates, sorted, source-tagged
  - evidence      claim/evidence/value rows (the discrepancy table)
  - series        two value series as an ASCII chart (pattern display)
  - crossref      a theme/symbol mapped to every source that mentions it

Deterministic only - no model inference here. Artifacts are saved under
data/artifacts/ so a case produces durable documents, not just chat.
"""
import json
import re
from pathlib import Path

ROOT = Path(__file__).resolve().parents[1]
ART_DIR = ROOT / "data" / "artifacts"

DATE_RE = re.compile(r"\b((?:19|20)\d{2}(?:-\d{1,2}(?:-\d{1,2})?)?)\b")
VALUE_RE = re.compile(r"\b(\d{1,2}:\d{2}|\d+(?:,\d{3})*\.?\d*%?)\b")
SOURCE_RE = re.compile(r"\[([a-z0-9_./-]+)\]|(https?://\S+)")

ART_KINDS = ("timeline", "evidence", "series", "crossref")


def _slug(s):
    return re.sub(r"[^a-z0-9]+", "_", s.lower()).strip("_") or "artifact"


def save_md(title, md):
    ART_DIR.mkdir(parents=True, exist_ok=True)
    path = ART_DIR / f"{_slug(title)}.md"
    path.write_text(md + ("\n" if not md.endswith("\n") else ""), encoding="utf-8")
    return path


def _src(line):
    m = SOURCE_RE.search(line)
    return m.group(1) or m.group(2) if m else "-"


def split_rows(ledger):
    """Split a ledger into (date_rows, value_rows, bare_rows)."""
    dates, values, bare = [], [], []
    for ln in ledger:
        ln = ln.strip()
        if not ln:
            continue
        dm = DATE_RE.search(ln)
        if dm:
            dates.append((dm.group(1), ln, _src(ln)))
            continue
        vm = VALUE_RE.search(ln)
        if vm:
            values.append((ln, _src(ln)))
        else:
            bare.append(ln)
    dates.sort(key=lambda r: r[0])
    return dates, values, bare


def render_timeline(ledger, title="Timeline"):
    rows, _, _ = split_rows(ledger)
    if not rows:
        return None
    out = [f"# {title}", "", "| Date | Event | Source |", "|---|---|---|"]
    for d, ln, s in rows:
        out.append(f"| {d} | {ln[:140]} | {s} |")
    out.append("")
    out.append("_Ordering is verifiable only where the source records the date; "
               "gaps are as informative as entries._")
    return "\n".join(out)


def render_evidence(ledger, title="Evidence & Discrepancies"):
    rows, _, _ = split_rows(ledger)
    if not rows:
        return None
    out = [f"# {title}", "", "| Date | Statement | Source |", "|---|---|---|"]
    for d, ln, s in rows:
        out.append(f"| {d} | {ln[:140]} | {s} |")
    return "\n".join(out)


def render_series(series, title="Series Comparison"):
    """series: list of (label, [numbers]). ASCII bars side by side."""
    if not series or len(series) < 2:
        return None
    labels = [s[0] for s in series]
    seqs = [list(s[1]) for s in series]
    n = min(len(x) for x in seqs)
    if n == 0:
        return None
    out = [f"# {title}", "", f"| {' | '.join(labels)} |", f"|{'---|' * len(labels)}"]
    for i in range(n):
        vals = [x[i] for x in seqs]
        out.append("| " + " | ".join(f"{v:.4g}" for v in vals) + " |")
    out.append("")
    out.append("Points (index)  " + "   ".join(f"[{i}]" for i in range(n)))
    for j, (lab, seq) in enumerate(series):
        mx = max(seq) or 1
        bars = ["#" * max(1, round(v / mx * 20)) for v in seq]
        out.append(f"{lab}: " + " ".join(bars))
    out.append("")
    out.append("_The chart only compares values; it asserts nothing about cause._")
    return "\n".join(out)


def render_crossref(theme, lines, title="Cross-Reference"):
    """theme: a term/symbol; lines: source-tagged ledger/notes."""
    out = [f"# {title}", "", f"Theme/symbol: **{theme}**", "", "| Source | Context |", "|---|---|"]
    hit = 0
    low = theme.lower()
    for ln in lines:
        if low in ln.lower():
            out.append(f"| {_src(ln)} | {ln[:150]} |")
            hit += 1
    if not hit:
        out.append("| - | (no mention in this case's documents) |")
    out.append("")
    out.append("_Absence of a mention is a finding, not an error: note it explicitly._")
    return "\n".join(out)


def synthesize(ledger, title, series=None, theme=None, lines=None):
    """Compose every artifact available for a case into one saved document."""
    arts = []
    tl = render_timeline(ledger, title=f"{title} - Timeline")
    if tl:
        arts.append(tl)
    ev = render_evidence(ledger, title=f"{title} - Evidence & Discrepancies")
    if ev:
        arts.append(ev)
    ch = render_series(series, title=f"{title} - Series Comparison") if series else None
    if ch:
        arts.append(ch)
    cr = render_crossref(theme, lines or ledger, title=f"{title} - Cross-Reference") if theme else None
    if cr:
        arts.append(cr)
    if not arts:
        return None
    doc = "\n\n---\n\n".join(arts)
    ART_DIR.mkdir(parents=True, exist_ok=True)
    path = ART_DIR / f"{_slug(title)}.md"
    path.write_text(doc + "\n", encoding="utf-8")
    return str(path), doc


def parse_series_arg(arg):
    """'title | label1:1,2,3 | label2:4,5,6' -> (title, [(label, [nums])])."""
    parts = [p.strip() for p in arg.split("|")]
    title = parts[0] or "Series Comparison"
    series = []
    for p in parts[1:]:
        if ":" not in p:
            continue
        lab, vs = p.split(":", 1)
        nums = []
        for v in vs.replace(" ", "").split(","):
            try:
                nums.append(float(v))
            except ValueError:
                pass
        if nums:
            series.append((lab.strip(), nums))
    return title, series