@iowarp/clio-coder 0.4.6 → 0.4.8

package/library/plugins/materio/assets/scripts/verify_citations.py

@@ -0,0 +1,616 @@
+#!/usr/bin/env python3
+"""
+verify_citations; resolve the citations in Materio research output against a
+bibliographic registry (Crossref) to surface unresolved identities and
+metadata conflicts for researcher review.
+
+It reads LITERATURE.md and/or .bib files, extracts citation candidates
+(BibTeX entries, quoted paper titles from the "Key Papers" table the
+literature agent writes, and quoted titles on other table rows that carry a
+4-digit year), then queries Crossref by DOI or by title and classifies each
+candidate:
+
+  VERIFIED      matched a real published work (title similar, year consistent)
+  MISMATCH      a work was found but the metadata disagrees (wrong year, or
+                only a weak title match); requires source inspection
+  NOT_FOUND     no plausible match in Crossref; inspect the source manually
+  UNVERIFIABLE  no extractable title (e.g. a placeholder like "[ML disc. 2024]")
+
+Design notes:
+  * No DOIs are assumed; the literature agent usually emits none, so title+
+    year resolution is the primary path.
+  * Network is OPTIONAL. `--offline` (or a failed request) falls back to
+    structural checks + placeholder detection so the tool still runs in CI or
+    air-gapped sessions; unresolved entries become UNVERIFIABLE (offline).
+  * Crossref is free and needs no key; we identify via a descriptive
+    User-Agent. HTTP 429 / 5xx responses are retried once after a short
+    backoff before an entry is reported UNVERIFIABLE.
+
+Exit status: 1 for NOT_FOUND/MISMATCH findings; 2 for incomplete input coverage.
+Zero deps (Python stdlib only).
+
+Usage (invoked from a project root, where .claude/ and .research/ live):
+  python3 assets/scripts/verify_citations.py .research/LITERATURE.md
+  python3 assets/scripts/verify_citations.py --offline path.bib
+  python3 assets/scripts/verify_citations.py --json task-01.bib
+"""
+import argparse
+import difflib
+import json
+import re
+import sys
+import time
+import urllib.error
+import urllib.parse
+import urllib.request
+
+CROSSREF = "https://api.crossref.org/works"
+UA = "materio-verify-citations/0.1 (Materio research assistant; citation verifier)"
+TITLE_MATCH = 0.75      # >= this title similarity => same paper
+TITLE_WEAK = 0.50       # [WEAK, MATCH) => MISMATCH (found but doubtful)
+REQ_TIMEOUT = 15
+REQ_PAUSE = 0.2         # polite spacing between Crossref calls
+RETRY_BACKOFF = 2.0     # seconds before the single retry on HTTP 429 / 5xx
+
+
+# ---- extraction -----------------------------------------------------------
+
+class Cite:
+    __slots__ = ("title", "year", "authors", "doi", "source", "raw")
+
+    def __init__(self, title=None, year=None, authors=None, doi=None,
+                 source="", raw=""):
+        self.title = title
+        self.year = year
+        self.authors = authors
+        self.doi = doi
+        self.source = source      # where in the file it came from
+        self.raw = raw            # original text snippet
+
+
+def _norm(s):
+    """Lowercase, strip non-alphanumerics to single spaces; for comparison."""
+    return re.sub(r"[^a-z0-9]+", " ", (s or "").lower()).strip()
+
+
+def similarity(a, b):
+    return difflib.SequenceMatcher(None, _norm(a), _norm(b)).ratio()
+
+
+# Parse BibTeX groups without importing or executing anything. Nested title
+# braces and one-line entries are common in real researcher bibliographies.
+def _group(text, start):
+    opening = text[start]
+    closing = '}' if opening == '{' else ')' if opening == '(' else '"'
+    depth, index, quoted = 1, start + 1, False
+    while index < len(text):
+        char = text[index]
+        if char == "\\":
+            index += 2
+            continue
+        if opening == '"':
+            if char == '"':
+                return text[start + 1:index], index + 1
+        else:
+            if char == '"':
+                quoted = not quoted
+            if not quoted:
+                if char == opening:
+                    depth += 1
+                elif char == closing:
+                    depth -= 1
+                    if depth == 0:
+                        return text[start + 1:index], index + 1
+        index += 1
+    raise ValueError("unterminated BibTeX value")
+
+
+def _bibtex_entries(text):
+    index = 0
+    while index < len(text):
+        match = re.search(r"@(\w+)\s*([{(])", text[index:])
+        if not match:
+            return
+        start = index + match.start()
+        group_start = index + match.end() - 1
+        try:
+            body, end = _group(text, group_start)
+        except ValueError:
+            return
+        index = end
+        if match.group(1).lower() not in {"comment", "preamble", "string"}:
+            yield body, text[start:end]
+
+
+def _bibtex_fields(body):
+    index = body.find(',') + 1
+    fields = {}
+    while index > 0 and index < len(body):
+        match = re.match(r"\s*,?\s*([\w-]+)\s*=\s*", body[index:])
+        if not match:
+            break
+        name = match.group(1).lower()
+        index += match.end()
+        if index >= len(body):
+            break
+        if body[index] in '{"':
+            try:
+                value, index = _group(body, index)
+            except ValueError:
+                break
+        else:
+            end = body.find(',', index)
+            if end == -1:
+                end = len(body)
+            value, index = body[index:end].strip(), end
+        fields[name] = re.sub(r"[{}]", "", value).strip()
+    return fields
+
+
+def extract_bibtex(text, source):
+    """Extract balanced BibTeX entries, preserving complete title contents."""
+    cites = []
+    for body, raw in _bibtex_entries(text):
+        fields = _bibtex_fields(body)
+        title = fields.get("title")
+        if title:
+            cites.append(Cite(title=title, year=fields.get("year"),
+                              authors=fields.get("author"), doi=fields.get("doi"),
+                              source=source, raw=raw[:160]))
+    return cites
+
+
+# A quoted title: straight or curly quotes wrapping 20+ non-quote characters
+# (the length floor rules out short quoted phrases and acronyms).
+_QUOTED = re.compile(r"[\"“]([^\"”“]{20,})[\"”]")
+
+# A 4-digit publication year.
+_YEAR = re.compile(r"\b(?:19|20)\d{2}\b")
+
+# An inline citation key like [Gong 2014] or [du Plessis 2019].
+_INLINE = re.compile(r"\[([A-Za-z][\w.\- ]*?\s(?:19|20)\d{2}[a-z]?)\]")
+
+# Placeholder-looking inline keys: contain a hyphenated/lowercase filler token
+# rather than a proper surname (e.g. "ML domain-discovery 2024").
+_PLACEHOLDER_HINT = re.compile(r"[a-z]+-[a-z]+|\b(?:ml|opt|disc|tbd|xxx)\b",
+                               re.IGNORECASE)
+
+# Markdown table separator row: |---|:---:|
+_TABLE_SEP = re.compile(r"^\s*\|?\s*:?-{3,}")
+
+# An author list as the literature agent writes it at the head of a Key
+# Papers row: "King W.E. et al.", "W.E. King and R. Smith",
+# "Tang M., Pistorius P.C.", "Scipioni Bertoli U. et al.".
+_INITIALS = r"(?:[A-Z]\.\s?){1,3}"
+_SURNAME = r"[A-Z][A-Za-z'’\-]+(?:\s+[A-Z][A-Za-z'’\-]+)?"
+_AUTHOR = rf"(?:{_INITIALS}\s*)?{_SURNAME}(?:\s+{_INITIALS})?"
+_AUTHOR_LIST = re.compile(
+    rf"^{_AUTHOR}(?:\s*(?:,|;|&|\band\b)\s*{_AUTHOR})*"
+    rf"(?:,?\s*et\s+al\.?)?\.?$")
+# ...and at least one multi-author marker or initial, so a bare capitalized
+# phrase ("Additive Manufacturing") is not mistaken for a surname pair.
+_AUTHOR_MARK = re.compile(r"\bet\s+al\b|\band\b|&|,|;|\b[A-Z]\.")
+
+
+def _cells(line):
+    return [c.strip() for c in line.strip().strip("|").split("|")]
+
+
+def _table_kind(header, heading):
+    """
+    Classify a markdown table by its header row (falling back to the section
+    heading above it). The literature agent writes two tables:
+
+      ## Key Papers               | Paper | Year | Why Important |
+      ## Methodological Landscape | Approach | Strengths | Limitations | Key Papers |
+
+    Only the first is bibliographic. The Landscape table's "Key Papers"
+    column repeats titles in shorthand, usually ellipsis-truncated, which
+    produced false NOT_FOUND/MISMATCH flags, so it is skipped outright.
+    """
+    cells = [c.lower() for c in _cells(header)]
+    first = cells[0] if cells else ""
+    if "approach" in first:
+        return "landscape"
+    if "paper" in first:
+        return "papers"
+    h = (heading or "").lower()
+    if "landscape" in h:
+        return "landscape"
+    if "key papers" in h:
+        return "papers"
+    return "other"
+
+
+def _looks_like_authors(seg):
+    """True if a Key Papers first-cell prefix reads as an author list."""
+    if not seg or len(seg) > 80 or re.search(r"\d", seg):
+        return False
+    return bool(_AUTHOR_LIST.match(seg)) and bool(_AUTHOR_MARK.search(seg))
+
+
+def _row_authors(first_cell):
+    """
+    Author string from the text that precedes the quoted title in a Key
+    Papers row, or None when that text is not an author list. Anything
+    else (a journal in *italics*, a volume number, prose) yields None so the
+    Crossref query is by title alone rather than polluted with junk.
+    """
+    seg = re.split(r"[\"“]", first_cell, 1)[0]
+    seg = re.split(r"[*_(\[]|\s[—–-]\s", seg, 1)[0]
+    seg = seg.strip().strip(",;:—–- ").strip()
+    return seg if _looks_like_authors(seg) else None
+
+
+def _titles_in(line):
+    """Yield (title, truncated) for each quoted title on a line."""
+    for qm in _QUOTED.finditer(line):
+        raw = qm.group(1).strip()
+        truncated = raw.endswith("...") or raw.endswith("…")
+        title = raw.rstrip(".…").rstrip(",;:").strip()
+        yield title, truncated
+
+
+def _row_year(line):
+    """First 4-digit year on a row, looking outside the quoted titles first."""
+    outside = _QUOTED.sub(" ", line)
+    ym = _YEAR.search(outside) or _YEAR.search(line)
+    return ym.group(0) if ym else None
+
+
+def dedup_truncated(cites):
+    """Deduplicate exact identities only; a complete prefix can be another paper.
+
+    Preserve conflicting DOI metadata as separate findings. Explicit ellipsis
+    fragments are omitted by the Markdown extractor, not guessed by title length.
+    """
+    seen, out = set(), []
+    for cite in cites:
+        key = (_norm(cite.title), cite.year, (cite.doi or "").lower())
+        if key not in seen:
+            seen.add(key)
+            out.append(cite)
+    return out
+
+
+def extract_markdown(text, source):
+    """
+    Extract citation candidates from a LITERATURE.md-style document. Only
+    three places are treated as bibliographic:
+      * BibTeX entries embedded in the document
+      * rows of the Key Papers table (title from the quoted string; authors
+        only when the row's first cell starts with an author list)
+      * rows of any other table that carry a 4-digit year
+    Prose lines are never scanned (quoted emphasis phrases are not titles),
+    and the Methodological Landscape table is skipped. Titles that end in an
+    ellipsis are omitted as incomplete fragments. Complete prefix titles remain
+    distinct; exact identity duplicates are removed. Also returns inline [Author Year] keys that look
+    like placeholders.
+    """
+    cites = list(extract_bibtex(text, source))
+    without_bibtex = text
+    for _, raw in _bibtex_entries(text):
+        without_bibtex = without_bibtex.replace(raw, "")
+    lines = without_bibtex.splitlines()
+    seen_titles = {(_norm(c.title), c.year) for c in cites}
+    heading, in_table, kind = None, False, "other"
+    for i, line in enumerate(lines):
+        if "|" not in line:
+            in_table = False
+            if line.lstrip().startswith("#"):
+                heading = line.lstrip("# ").strip()
+            continue
+        if not in_table:
+            in_table = True
+            # A header row is one followed by a |---| separator; a table
+            # without one is classified by its section heading alone.
+            nxt = lines[i + 1] if i + 1 < len(lines) else ""
+            if _TABLE_SEP.match(nxt):
+                kind = _table_kind(line, heading)
+                continue
+            kind = _table_kind("", heading)
+        if _TABLE_SEP.match(line) or kind == "landscape":
+            continue
+        year = _row_year(line)
+        if kind != "papers" and not year:
+            continue
+        authors = _row_authors(_cells(line)[0]) if kind == "papers" else None
+        for title, truncated in _titles_in(line):
+            key = _norm(title)
+            if truncated or (key, year) in seen_titles or len(key) < 12:
+                continue
+            seen_titles.add((key, year))
+            cites.append(Cite(title=title, year=year, authors=authors,
+                              source=source, raw=line.strip()[:160]))
+    return dedup_truncated(cites), _extract_inline(text, source)
+
+
+def _surnames(authors):
+    """
+    Plausible primary-author surname tokens from a free-form author string.
+    Author order is ambiguous here; tables use both "King W.E." (surname
+    first) and "W.E. King" (surname last); so we return every token that
+    looks like a surname (capitalized, mostly letters) and drop initials
+    like "W.E.". The caller matches this set against the work's authors.
+    """
+    if not authors:
+        return set()
+    first = re.split(r"\band\b|;|&", authors)[0]
+    first = re.sub(r"\bet al\.?\b", "", first, flags=re.IGNORECASE)
+    out = set()
+    for tok in re.split(r"[\s,]+", first):
+        letters = re.sub(r"[^A-Za-z]", "", tok)
+        if len(letters) < 2:
+            continue
+        # An initials cluster like "W.E." -> "WE" is all-caps and short; skip.
+        if tok.replace(".", "").isupper() and len(letters) <= 3:
+            continue
+        if tok[:1].isupper():
+            out.add(_norm(letters))
+    return out
+
+
+def _item_surnames(item):
+    """Normalized family names AND their word tokens (handles compound
+    surnames like 'Scipioni Bertoli' whichever way the source split them)."""
+    out = set()
+    for a in (item.get("author") or []):
+        fam = a.get("family")
+        if not fam:
+            continue
+        nf = _norm(fam)
+        out.add(nf)
+        out.update(nf.split())
+    return out
+
+
+def _extract_inline(text, source):
+    """Return inline [Author Year]-style keys that look like placeholders
+    (a non-name filler token rather than a real surname, e.g.
+    "[ML domain-discovery 2024]"); these are unverifiable by construction."""
+    placeholders = []
+    seen = set()
+    for m in _INLINE.finditer(text):
+        key = m.group(1).strip()
+        if key in seen:
+            continue
+        seen.add(key)
+        if _PLACEHOLDER_HINT.search(key):
+            placeholders.append(Cite(title=None, source=source,
+                                     raw="[" + key + "]"))
+    return placeholders
+
+
+# ---- verification ---------------------------------------------------------
+
+def _retryable(code):
+    return code == 429 or 500 <= code < 600
+
+
+def _get(url):
+    """GET JSON. One retry after RETRY_BACKOFF on HTTP 429 / 5xx (Crossref
+    rate limiting or a transient server fault); any other error propagates."""
+    req = urllib.request.Request(url, headers={"User-Agent": UA})
+    for attempt in (0, 1):
+        try:
+            with urllib.request.urlopen(req, timeout=REQ_TIMEOUT) as r:
+                return json.load(r)
+        except urllib.error.HTTPError as e:
+            if attempt == 0 and _retryable(e.code):
+                time.sleep(RETRY_BACKOFF)
+                continue
+            raise
+
+
+DOI_ABSENT = object()   # sentinel: DOI definitively not in Crossref (HTTP 404)
+
+
+def crossref_by_doi(doi):
+    """Return the work dict, DOI_ABSENT if Crossref lacks the DOI (other agencies may own it),
+    or None on a real network error."""
+    try:
+        data = _get(CROSSREF + "/" + urllib.parse.quote(doi))
+        return data.get("message")
+    except urllib.error.HTTPError as e:
+        return DOI_ABSENT if e.code == 404 else None
+    except Exception:
+        return None
+
+
+def crossref_by_title(title, authors=None, year=None):
+    q = title
+    if authors:
+        q = authors + " " + q
+    params = {"rows": "5", "query.bibliographic": q}
+    url = CROSSREF + "?" + urllib.parse.urlencode(params)
+    try:
+        data = _get(url)
+        return data.get("message", {}).get("items", [])
+    except Exception:
+        return None      # None = network failure (vs [] = no results)
+
+
+def _item_year(item):
+    for k in ("published-print", "published-online", "issued"):
+        dp = item.get(k, {}).get("date-parts") if item else None
+        if dp and dp[0] and dp[0][0]:
+            return str(dp[0][0])
+    return None
+
+
+def verify_one(c, offline):
+    """Return (status, detail_dict)."""
+    if not c.title:
+        return "UNVERIFIABLE", {"reason": "no extractable title (placeholder)"}
+    if offline:
+        return "UNVERIFIABLE", {"reason": "offline mode; not resolved"}
+
+    # DOI path
+    if c.doi:
+        item = crossref_by_doi(c.doi)
+        if item is DOI_ABSENT:
+            return "UNVERIFIABLE", {"doi": c.doi,
+                                    "reason": "DOI absent from Crossref; another registration agency may own it"}
+        if item is None:
+            return "UNVERIFIABLE", {"reason": "network error resolving DOI"}
+        rt = (item.get("title") or [""])[0]
+        sim = similarity(c.title, rt)
+        ry = _item_year(item)
+        year_ok = not (c.year and ry) or (c.year.isdigit() and ry.isdigit() and abs(int(c.year) - int(ry)) <= 1)
+        if sim >= TITLE_MATCH and year_ok:
+            return "VERIFIED", {"doi": c.doi, "matched_title": rt,
+                                "similarity": round(sim, 2)}
+        return "MISMATCH", {"doi": c.doi, "matched_title": rt,
+                            "similarity": round(sim, 2),
+                            "reason": "DOI metadata needs review: weak title match or inconsistent year"}
+
+    # Title path
+    items = crossref_by_title(c.title, c.authors, c.year)
+    if items is None:
+        return "UNVERIFIABLE", {"reason": "network error querying Crossref"}
+    best, best_sim = None, 0.0
+    for it in items:
+        rt = (it.get("title") or [""])[0]
+        s = similarity(c.title, rt)
+        if s > best_sim:
+            best, best_sim = it, s
+    if best is None:
+        return "NOT_FOUND", {"reason": "no Crossref results"}
+    rt = (best.get("title") or [""])[0]
+    ry = _item_year(best)
+    detail = {"matched_title": rt, "similarity": round(best_sim, 2),
+              "doi": best.get("DOI"), "matched_year": ry}
+
+    year_ok = not (c.year and ry) or (c.year.isdigit() and ry.isdigit() and abs(int(c.year) - int(ry)) <= 1)
+    # Author+year corroboration rescues abbreviated titles (e.g. a source
+    # that shortened "...laser powder-bed fusion" to "...LPBF"): the primary
+    # author's surname appearing in the matched work, with a consistent year,
+    # is strong independent evidence it is the same paper.
+    author_ok = bool(_surnames(c.authors) & _item_surnames(best))
+
+    if best_sim >= TITLE_MATCH:
+        if not year_ok:
+            detail["reason"] = f"year {c.year} vs Crossref {ry}"
+            return "MISMATCH", detail
+        return "VERIFIED", detail
+    if best_sim >= TITLE_WEAK:
+        if author_ok and year_ok:
+            detail["reason"] = "abbreviated title; confirmed by author + year"
+            return "VERIFIED", detail
+        detail["reason"] = "only a weak title match; verify manually"
+        return "MISMATCH", detail
+    detail["reason"] = "closest Crossref result is unrelated"
+    return "NOT_FOUND", detail
+
+
+# ---- driver ---------------------------------------------------------------
+
+def collect(paths):
+    cites, placeholders = [], []
+    for p in paths:
+        try:
+            with open(p, encoding="utf-8") as f:
+                text = f.read()
+        except OSError as e:
+            print(f"WARN: cannot read {p}: {e}", file=sys.stderr)
+            continue
+        if p.endswith(".bib"):
+            cites += extract_bibtex(text, p)
+        else:
+            md_cites, md_ph = extract_markdown(text, p)
+            cites += md_cites
+            placeholders += md_ph
+    return dedup_truncated(cites), placeholders
+
+
+def main():
+    ap = argparse.ArgumentParser(description="Verify citations against Crossref.")
+    ap.add_argument("paths", nargs="+", help="LITERATURE.md and/or .bib files")
+    ap.add_argument("--offline", action="store_true",
+                    help="skip network; structural + placeholder checks only")
+    ap.add_argument("--json", action="store_true", help="machine-readable output")
+    args = ap.parse_args()
+
+    unreadable = []
+    for path in args.paths:
+        try:
+            with open(path, encoding="utf-8") as source:
+                source.read()
+        except (OSError, UnicodeError) as error:
+            unreadable.append({"file": path, "reason": str(error)})
+    readable = [path for path in args.paths if path not in {item["file"] for item in unreadable}]
+    cites, placeholders = collect(readable)
+
+    results = []
+    for c in cites:
+        status, detail = verify_one(c, args.offline)
+        results.append((c, status, detail))
+        if not args.offline and not c.doi:
+            time.sleep(REQ_PAUSE)
+    for ph in placeholders:
+        results.append((ph, "UNVERIFIABLE",
+                        {"reason": "self-flagged placeholder citation key"}))
+
+    counts = {}
+    for _, status, _ in results:
+        counts[status] = counts.get(status, 0) + 1
+
+    if args.json:
+        out = {
+            "coverage": {"files_requested": len(args.paths), "files_read": len(readable), "unreadable": unreadable, "candidates": len(cites)},
+            "summary": counts,
+            "results": [
+                {"title": c.title, "year": c.year, "doi": c.doi,
+                 "source": c.source, "raw": c.raw,
+                 "status": status, "detail": detail}
+                for c, status, detail in results
+            ],
+        }
+        print(json.dumps(out, indent=2))
+    else:
+        _print_report(results, counts, args.offline)
+
+    bad = counts.get("NOT_FOUND", 0) + counts.get("MISMATCH", 0)
+    if unreadable or not results:
+        if not args.json:
+            print("Citation coverage incomplete: unreadable inputs or no extractable citations", file=sys.stderr)
+        return 2
+    return 1 if bad else 0
+
+
+_ICON = {"VERIFIED": "OK  ", "MISMATCH": "WARN", "NOT_FOUND": "FAIL",
+         "UNVERIFIABLE": "??  "}
+
+
+def _print_report(results, counts, offline):
+    order = ["NOT_FOUND", "MISMATCH", "UNVERIFIABLE", "VERIFIED"]
+    for status in order:
+        group = [(c, d) for c, s, d in results if s == status]
+        if not group:
+            continue
+        print(f"\n{status} ({len(group)}):")
+        for c, d in group:
+            label = c.title or c.raw or "?"
+            line = f"  [{_ICON[status]}] {label[:88]}"
+            print(line)
+            bits = []
+            if d.get("reason"):
+                bits.append(d["reason"])
+            if d.get("matched_title"):
+                tag = "closest:" if status == "NOT_FOUND" else "~"
+                bits.append(f"{tag} {d['matched_title'][:70]}")
+            if d.get("doi") and status != "UNVERIFIABLE":
+                bits.append(f"doi:{d['doi']}")
+            if bits:
+                print("           " + " | ".join(bits))
+    mode = " (offline)" if offline else ""
+    counts_str = ", ".join(f"{k}={v}" for k, v in sorted(counts.items())) \
+        or "no citations found"
+    print(f"\nverify-citations{mode}: {counts_str}")
+    bad = counts.get("NOT_FOUND", 0) + counts.get("MISMATCH", 0)
+    if bad:
+        print(f"  -> {bad} citation(s) need attention.")
+
+
+if __name__ == "__main__":
+    sys.exit(main())

package/library/plugins/materio/assets/templates/RESEARCH.md

@@ -0,0 +1,39 @@
+# Research Identity
+
+## Research Area
+- **Domain**: [Structural Materials | Functional Materials | Energy Materials | Biomaterials | Nanomaterials | Polymers | Computational MatSci | Characterization]
+- **Sub-field**: [e.g., High-entropy alloys, Solid-state batteries, Bone scaffolds, 2D materials]
+- **Specific Interest**: [precise topic; phenomena, material system, or application]
+
+## Research Prompt
+[The refined research question; one sentence. What does this research answer?]
+
+## Scope
+### In Scope
+- [specific materials, phenomena, length scales, conditions]
+- [specific characterization methods or computational approaches]
+
+### Out of Scope
+- [explicit exclusions; what this research will NOT address]
+
+## Researcher Profile
+- **Career Stage**: [undergrad | grad student | postdoc | PI | industry researcher]
+- **Available Resources**: [experimental equipment | computational cluster | literature only | combination]
+- **Timeline**: [months/years]
+- **Institutional Context**: [university lab | national lab | industry R&D | independent]
+
+## Initial Keywords
+- **Primary**: []
+- **Secondary**: []
+- **Exclude from search**: []
+
+## Suggested Research Prompts (generated during interview)
+1. [prompt option A; fundamental understanding focus]
+2. [prompt option B; application focus]
+3. [prompt option C; methodological focus]
+
+## Selected Prompt
+[User-confirmed research prompt after interview]
+
+## Key Decisions
+[Decisions locked in during identify-research; scope choices, methodological constraints, etc.]