scripts/build_viewer.py

#!/usr/bin/env python3
"""Build a static cross-dataset source-document and scaffold viewer."""

from __future__ import annotations

import argparse
import csv
import hashlib
import json
import shutil
from pathlib import Path
from typing import Any, Iterable


ROOT = Path(__file__).resolve().parents[1]
INFO = Path("/home/azureuser/projects/information-scaffolds")
SESSION_INPUTS = Path(
    "/home/azureuser/.copilot/session-state/6a2da107-0228-4316-ad25-f248cf796604/"
    "files/structure-generation-viewer-inputs"
)
DATA_ROOT = Path("/mnt/ramdisk/blobstore/timchen0618/data")

SHAPE_ALIASES = {
    "tabular_records": "tables",
    "tables": "tables",
    "relation_graphs_and_mappings": "knowledge_graphs",
    "knowledge_graphs": "knowledge_graphs",
    "claim_and_theme_summaries": "claims",
    "claims": "claims",
    "chronology_and_timeline_indexes": "timelines",
    "timelines": "timelines",
    "qa_shortcuts_and_templates": "legacy_qa_shortcuts",
}

SHAPE_LABELS = {
    "tables": "Tables",
    "knowledge_graphs": "Knowledge graphs",
    "claims": "Claims",
    "timelines": "Timelines",
    "legacy_qa_shortcuts": "Legacy QA shortcuts",
}

DATASETS = {
    "monaco_dev": {
        "label": "MoNaCo-dev",
        "source": {
            "kind": "raw_dir",
            "path": INFO / "outputs/rawtext_corpus_monacodev/monaco_dev/scaffolds",
        },
        "runs": [
            {
                "id": "legacy_full",
                "label": "Legacy full extraction - 50-doc subset",
                "job": "sweet_screw_frnspf41x5",
                "schema": "legacy five-shape",
                "root": INFO
                / "outputs/e2e_dev_scaffolds/monaco_dev/named-outputs/scaffolds_dir",
            },
            {
                "id": "current",
                "label": "Current schema-v3 extraction",
                "job": "quirky_cabbage_zdpp543s3z",
                "schema": "3",
                "root": SESSION_INPUTS / "monaco_dev/current/scaffolds_dir",
                "quality": SESSION_INPUTS / "monaco_dev/current/schema_quality_report/report",
            },
        ],
    },
    "wiki_opentable_dev": {
        "label": "Open-WikiTable-dev",
        "source": {
            "kind": "raw_dir",
            "path": INFO
            / "outputs/rawtext_corpus_wikiotdev/wiki_opentable_dev/scaffolds",
        },
        "runs": [
            {
                "id": "legacy_full",
                "label": "Legacy full extraction - 50-doc subset",
                "job": "cyan_muscle_ylws4v86bz",
                "schema": "legacy five-shape",
                "root": INFO
                / "outputs/e2e_dev_scaffolds/wiki_opentable_dev/named-outputs/scaffolds_dir",
            },
            {
                "id": "current",
                "label": "Current schema-v3 extraction",
                "job": "purple_eye_8f31ymzz71",
                "schema": "3",
                "root": SESSION_INPUTS / "wiki_opentable_dev/current/scaffolds_dir",
                "quality": SESSION_INPUTS
                / "wiki_opentable_dev/current/schema_quality_report/report",
            },
        ],
    },
    "phantom_wiki": {
        "label": "PhantomWiki",
        "source": {
            "kind": "corpus",
            "path": DATA_ROOT / "eval/phantom_wiki/unified/corpus.unified.jsonl",
        },
        "runs": [
            {
                "id": "legacy_full",
                "label": "Legacy full extraction - 50-doc subset",
                "job": "teal_jicama_qxvpyf3jsd",
                "schema": "legacy five-shape",
                "root": INFO
                / "outputs/e2e_runs/new-datasets-full-20260711/phantom_wiki/"
                "named-outputs/scaffolds_dir",
            },
            {
                "id": "current",
                "label": "Current schema-v3 extraction",
                "job": "mango_avocado_7qnv0vlf46",
                "schema": "3",
                "root": SESSION_INPUTS / "phantom_wiki/current/scaffolds_dir",
                "quality": SESSION_INPUTS
                / "phantom_wiki/current/schema_quality_report/report",
            },
        ],
    },
}


def read_json(path: Path) -> Any:
    return json.loads(path.read_text(encoding="utf-8"))


def read_jsonl(path: Path) -> Iterable[dict[str, Any]]:
    with path.open(encoding="utf-8") as handle:
        for line_number, line in enumerate(handle, 1):
            if not line.strip():
                continue
            value = json.loads(line)
            if not isinstance(value, dict):
                raise ValueError(f"{path}:{line_number}: expected an object")
            yield value


def parse_doc_id(text: str, fallback: str) -> str:
    for raw in text.splitlines():
        line = raw.strip()
        if not line:
            continue
        if line.startswith("#") and line.lstrip("#").strip().lower().startswith("id:"):
            value = line.lstrip("#").strip().split(":", 1)[1].strip()
            return value or fallback
        return fallback
    return fallback


def load_selected_sources(source: dict[str, Any]) -> list[dict[str, str]]:
    path = Path(source["path"])
    if source["kind"] == "raw_dir":
        selected: list[Path] = []
        for candidate in path.rglob("*.txt"):
            if candidate.is_file():
                selected.append(candidate)
            if len(selected) == 50:
                break
        selected.sort()
        rows = []
        for candidate in selected:
            text = candidate.read_text(encoding="utf-8")
            rows.append(
                {
                    "doc_id": parse_doc_id(text, candidate.stem),
                    "contents": text,
                    "source_path": str(candidate.relative_to(path)),
                }
            )
        return rows

    rows = []
    for row in read_jsonl(path):
        rows.append(
            {
                "doc_id": str(row["id"]),
                "contents": str(row.get("contents", "")),
                "source_path": str(row["id"]),
            }
        )
        if len(rows) == 50:
            break
    return rows


def artifact_format(filename: str) -> str:
    if filename.endswith(".edges.jsonl"):
        return "jsonl"
    if filename.endswith(".timeline.json"):
        return "json"
    suffix = Path(filename).suffix.lower().lstrip(".")
    return suffix if suffix in {"csv", "json", "jsonl", "md"} else "text"


def load_quality(path: Path | None) -> dict[str, Any] | None:
    if path is None or not path.exists():
        return None
    return read_json(path)


def load_scaffold_run(config: dict[str, Any]) -> dict[str, Any]:
    if config.get("status") == "pending":
        return {
            "id": config["id"],
            "label": config["label"],
            "job": config["job"],
            "schema": config["schema"],
            "status": "pending",
            "documents": {},
            "quality": None,
        }

    root = Path(config["root"])
    top = read_json(root / "_index.json")
    documents: dict[str, list[dict[str, Any]]] = {}
    shape_summaries = []
    for shape in top.get("shapes", []):
        raw_shape = str(shape["shape_id"])
        canonical_shape = SHAPE_ALIASES.get(raw_shape, raw_shape)
        folder = str(shape.get("folder") or raw_shape).rstrip("/")
        shape_index = read_json(root / folder / "_index.json")
        shape_summaries.append(
            {
                "id": canonical_shape,
                "raw_id": raw_shape,
                "label": SHAPE_LABELS.get(canonical_shape, shape.get("display_name", raw_shape)),
                "description": shape_index.get("description", shape.get("definition", "")),
                "n_files": len(shape_index.get("entries", [])),
            }
        )
        for entry in shape_index.get("entries", []):
            doc_id = str(entry["doc_id"])
            filename = str(entry["file"])
            artifact_path = root / folder / filename
            documents.setdefault(doc_id, []).append(
                {
                    "shape": canonical_shape,
                    "shape_label": SHAPE_LABELS.get(canonical_shape, canonical_shape),
                    "raw_shape": raw_shape,
                    "filename": filename,
                    "format": artifact_format(filename),
                    "unit_name": entry.get("unit_name"),
                    "unit_description": entry.get("unit_description"),
                    "content": artifact_path.read_text(encoding="utf-8"),
                }
            )

    return {
        "id": config["id"],
        "label": config["label"],
        "job": config["job"],
        "schema": config["schema"],
        "status": "ready",
        "documents": documents,
        "quality": load_quality(config.get("quality")),
        "summary": {
            "n_docs_seen": top.get("n_docs_seen"),
            "n_files_written": top.get("n_files_written"),
            "shapes_hash": top.get("shapes_hash"),
            "shape_summaries": shape_summaries,
            "validation_metrics": top.get("validation_metrics"),
        },
    }


def record_name(doc_id: str) -> str:
    return hashlib.sha1(doc_id.encode()).hexdigest() + ".json"


def build_dataset(dataset_id: str, config: dict[str, Any], data_dir: Path) -> dict[str, Any]:
    sources = load_selected_sources(config["source"])
    if len(sources) != 50:
        raise ValueError(f"{dataset_id}: expected 50 source documents, got {len(sources)}")
    source_ids = [row["doc_id"] for row in sources]
    if len(set(source_ids)) != 50:
        raise ValueError(f"{dataset_id}: selected source IDs are not unique")

    runs = [load_scaffold_run(run) for run in config["runs"]]
    records_dir = data_dir / "records" / dataset_id
    records_dir.mkdir(parents=True, exist_ok=True)
    index_rows = []
    for position, source in enumerate(sources, 1):
        doc_id = source["doc_id"]
        run_records = {}
        for run in runs:
            artifacts = run["documents"].get(doc_id, [])
            run_records[run["id"]] = {
                "status": run["status"],
                "artifacts": artifacts,
                "n_artifacts": len(artifacts),
            }
        digest = hashlib.sha256(source["contents"].encode()).hexdigest()
        record = {
            "dataset": dataset_id,
            "dataset_label": config["label"],
            "position": position,
            "doc_id": doc_id,
            "source": {
                "contents": source["contents"],
                "source_path": source["source_path"],
                "sha256": digest,
            },
            "runs": run_records,
        }
        filename = record_name(doc_id)
        (records_dir / filename).write_text(
            json.dumps(record, ensure_ascii=False), encoding="utf-8"
        )
        index_rows.append(
            {
                "position": position,
                "doc_id": doc_id,
                "record": f"data/records/{dataset_id}/{filename}",
                "source_preview": source["contents"].replace("\n", " ")[:180],
                "source_sha256": digest,
                "run_counts": {
                    run_id: value["n_artifacts"] for run_id, value in run_records.items()
                },
            }
        )

    return {
        "id": dataset_id,
        "label": config["label"],
        "n_docs": len(sources),
        "runs": [
            {
                key: run.get(key)
                for key in ("id", "label", "job", "schema", "status", "quality", "summary")
            }
            for run in runs
        ],
        "records": index_rows,
    }


def validate_bundle(index: dict[str, Any], data_dir: Path) -> None:
    if len(index["datasets"]) != 3:
        raise ValueError("expected three datasets")
    for dataset in index["datasets"]:
        if dataset["n_docs"] != 50 or len(dataset["records"]) != 50:
            raise ValueError(f"{dataset['id']}: expected 50 records")
        if len({row["doc_id"] for row in dataset["records"]}) != 50:
            raise ValueError(f"{dataset['id']}: duplicate document IDs")
        for row in dataset["records"]:
            path = ROOT / row["record"]
            if not path.exists():
                raise ValueError(f"missing record shard: {path}")
    phantom = next(row for row in index["datasets"] if row["id"] == "phantom_wiki")
    current = next(run for run in phantom["runs"] if run["id"] == "current")
    if current["status"] != "ready":
        raise ValueError("PhantomWiki current run must be ready")


def main() -> int:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--out", type=Path, default=ROOT / "data")
    args = parser.parse_args()

    if args.out.exists():
        shutil.rmtree(args.out)
    args.out.mkdir(parents=True)
    index = {
        "title": "Structure Generation Prompt Viewer",
        "datasets": [
            build_dataset(dataset_id, config, args.out)
            for dataset_id, config in DATASETS.items()
        ],
    }
    (args.out / "index.json").write_text(
        json.dumps(index, ensure_ascii=False, indent=2) + "\n", encoding="utf-8"
    )
    validate_bundle(index, args.out)
    print(
        "Built 150 document records: "
        + ", ".join(f"{row['label']}={row['n_docs']}" for row in index["datasets"])
    )
    return 0


if __name__ == "__main__":
    raise SystemExit(main())