scripts/verify_all_transcripts.py

#!/usr/bin/env python3
"""
Agnostic Deep Transcript & Corpus Verification Script.
Audits:
- Word counts & character lengths
- SHA-256 hash match against transcript content
- Residual HTML tags, CSS blocks, or web scraper artifacts
- Date, Year, Location, Period, and Event Type validation
- Duplicate detection & provenance consistency
"""

from __future__ import annotations
import argparse
import json
import re
import hashlib
from pathlib import Path
from collections import defaultdict

def parse_args():
    parser = argparse.ArgumentParser(description="Verify transcript quality, formatting, and cryptographic hashes.")
    parser.add_argument(
        "--corpus-dir",
        type=Path,
        default=Path(__file__).resolve().parent.parent,
        help="Path to the dataset directory."
    )
    parser.add_argument(
        "--min-words",
        type=int,
        default=50,
        help="Minimum transcript word count threshold for non-stub speeches."
    )
    return parser.parse_args()

def verify_corpus():
    args = parse_args()
    data_dir = args.corpus_dir.resolve() / "data"
    speeches_file = data_dir / "speeches.jsonl"
    sources_file = data_dir / "sources.jsonl"

    if not speeches_file.exists():
        print(f"[ERROR] Speeches file does not exist: {speeches_file}")
        return 1

    print("=== Starting Deep Transcript Verification ===")
    print(f"Data Directory: {data_dir}")
    
    total_speeches = 0
    clean_speeches = 0
    empty_speeches = 0
    short_speeches = 0
    html_residue_issues = 0
    hash_mismatches = 0
    year_issues = 0
    period_issues = 0

    stats = {
        "word_counts": [],
        "char_counts": [],
        "by_period": defaultdict(int),
        "by_year": defaultdict(int),
        "by_rights": defaultdict(int),
        "by_status": defaultdict(int),
    }

    html_tag_pattern = re.compile(r"<(div|span|p|a\s|script|style|table|tr|td|body|html)[^>]*>", re.IGNORECASE)

    with speeches_file.open("r", encoding="utf-8") as f:
        for idx, line in enumerate(f, 1):
            total_speeches += 1
            rec = json.loads(line.strip())
            
            transcript = rec.get("transcript", "")
            title = rec.get("title", "")
            date_str = rec.get("date", "")
            year = rec.get("year")
            period = rec.get("period", "")
            rights = rec.get("rights_status", "")
            status = rec.get("transcript_status", "")
            sha = rec.get("sha256", "")

            stats["by_period"][period] += 1
            stats["by_year"][year or 0] += 1
            stats["by_rights"][rights] += 1
            stats["by_status"][status] += 1

            # Length & Word count
            words = len(transcript.split())
            chars = len(transcript)
            stats["word_counts"].append(words)
            stats["char_counts"].append(chars)

            if not transcript.strip():
                empty_speeches += 1
                continue
            elif words < args.min_words and status not in ("stub_placeholder", "video_only"):
                short_speeches += 1

            # SHA-256 Check
            computed_sha = hashlib.sha256(transcript.encode("utf-8")).hexdigest()
            if sha and sha != computed_sha:
                hash_mismatches += 1

            # HTML Residue Check
            if html_tag_pattern.search(transcript):
                html_residue_issues += 1

            # Year Check
            if not year or year < 1900 or year > 2100:
                year_issues += 1

            # Period check
            if not period:
                period_issues += 1

            clean_speeches += 1

    print(f"\nVerification Results:")
    print(f"  Total Canonical Speeches Audited: {total_speeches}")
    print(f"  Verified High-Quality / Clean:    {clean_speeches}")
    print(f"  Empty Transcripts (Stubs/Video):  {empty_speeches}")
    print(f"  Suspiciously Short (<{args.min_words} words):   {short_speeches}")
    print(f"  HTML Residue / Tag Issues:        {html_residue_issues}")
    print(f"  Hash Mismatches:                  {hash_mismatches}")
    print(f"  Year Anomaly Count:               {year_issues}")
    print(f"  Period Anomaly Count:             {period_issues}")
    
    total_words = sum(stats["word_counts"])
    avg_words = total_words / total_speeches if total_speeches else 0
    print(f"\nCorpus Text Statistics:")
    print(f"  Total Words in Transcripts:     {total_words:,}")
    print(f"  Average Words per Speech:       {avg_words:,.1f}")
    print(f"  Max Words in Single Speech:     {max(stats['word_counts']) if stats['word_counts'] else 0:,}")

    print("\nPeriod Distribution:")
    for p, cnt in sorted(stats["by_period"].items()):
        print(f"  - {p}: {cnt} speeches")

    print("\nRights Status Breakdown:")
    for r, cnt in sorted(stats["by_rights"].items()):
        print(f"  - {r}: {cnt} records")

    return 0

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