feat: training data pipeline — knowledge entries → JSONL training pairs

Add scripts/knowledge_to_training_pairs.py which reads quality-gated
knowledge entries from knowledge/index.json and emits terse→rich
training pairs in JSONL format.

Features:
- Derives terse queries from facts via category-aware heuristics
- Configurable quality filters: min-confidence, model-filter, date range
- Output includes domain, source_confidence, source_model
- Smoke tests added in tests/test_knowledge_to_training_pairs.py

Deliverables for #199:
1. Pipeline script: scripts/knowledge_to_training_pairs.py
2. End-to-end: knowledge/index.json → training_pairs.jsonl (or custom JSONL)
3. Config: min-confidence, model-filter, after/before date filters
4. Test: 9 smoke tests covering conversion, filtering, and end-to-end run

Closes #199

scripts/knowledge_to_training_pairs.py

@@ -0,0 +1,255 @@
+#!/usr/bin/env python3
+"""
+knowledge_to_training_pairs.py — Convert quality-gated knowledge entries into training pairs.
+
+Reads knowledge/index.json (or a custom JSONL of entries), applies quality filters,
+and emits terse→rich training pairs in JSONL format for model fine-tuning.
+
+Usage:
+    python3 scripts/knowledge_to_training_pairs.py \
+        --input knowledge/index.json \
+        --output training_pairs.jsonl \
+        --min-confidence 0.7 \
+        --model-filter claude-sonnet,gpt-4 \
+        --after 2026-01-01
+
+Input entry format (from index.json facts):
+    {
+        "id": "hermes-agent:pitfall:001",
+        "fact": "deploy-crons.py leaves jobs in mixed model format",
+        "category": "pitfall",
+        "domain": "hermes-agent",
+        "confidence": 0.95,
+        ...
+    }
+
+Output training pair format:
+    {
+        "terse": "How do I handle deploy-crons.py mixed model format?",
+        "rich": "deploy-crons.py leaves jobs in mixed model format.",
+        "domain": "hermes-agent",
+        "source_confidence": 0.95,
+        "source_model": "unknown"
+    }
+"""
+
+import argparse
+import json
+import os
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Optional
+
+
+def fact_to_terse(fact: str, category: str, domain: str) -> str:
+    """
+    Derive a short user query from a knowledge fact.
+
+    Strategy:
+    - Pitfalls → "How do I avoid/handle/fix <fact excerpt>?"
+    - Patterns  → "What's the recommended way to <pattern core>?"
+    - Tool quirks → "How does <tool> behave in <context>?"
+    - Facts → "What should I know about <fact excerpt>?"
+    - Questions → "What is the answer to: <fact>?"
+    """
+    fact_lower = fact.lower()
+    # Extract a concise excerpt (first sentence or 80 chars)
+    excerpt = fact.split('. ')[0] if '. ' in fact else fact[:80]
+
+    if category == "pitfall":
+        verbs = ["avoid", "handle", "fix", "prevent"]
+        # pick verb based on fact wording
+        if "trigger" in fact_lower or "cause" in fact_lower:
+            verb = "avoid"
+        elif "broken" in fact_lower or "fails" in fact_lower:
+            verb = "fix"
+        else:
+            verb = "handle"
+        return f"How do I {verb} {excerpt.rstrip('.')}?"
+    elif category == "pattern":
+        return f"What's the recommended way to {excerpt.rstrip('.')}?"
+    elif category == "tool-quirk":
+        # Try to extract tool name
+        tool = fact.split()[0] if fact.split() else domain
+        return f"How does {tool} behave in this context?"
+    elif category == "question":
+        return f"What is the answer to: {excerpt}?"
+    else:  # fact or unknown
+        return f"What should I know about {excerpt.rstrip('.')}?"
+
+
+def parse_date(date_str: Optional[str]) -> Optional[datetime]:
+    """Parse ISO date string to datetime, or return None."""
+    if not date_str:
+        return None
+    try:
+        return datetime.fromisoformat(date_str.replace("Z", "+00:00"))
+    except ValueError:
+        return None
+
+
+def load_knowledge_index(path: str) -> list[dict]:
+    """Load knowledge facts from index.json (or plain JSONL of entries)."""
+    p = Path(path)
+    if not p.exists():
+        print(f"ERROR: Knowledge input not found: {path}", file=sys.stderr)
+        sys.exit(1)
+
+    with open(p) as f:
+        data = json.load(f)
+
+    # index.json format: {"facts": [...], ...}
+    if isinstance(data, dict) and "facts" in data:
+        return data["facts"]
+    # JSONL format: one entry per line
+    if isinstance(data, list):
+        return data
+    # Plain file with JSON array
+    print(f"ERROR: Unrecognized input format in {path}", file=sys.stderr)
+    sys.exit(1)
+
+
+def filter_entries(entries: list[dict],
+                   min_confidence: float = 0.0,
+                   model_filter: Optional[list[str]] = None,
+                   after: Optional[datetime] = None,
+                   before: Optional[datetime] = None) -> list[dict]:
+    """Apply quality and provenance filters."""
+    filtered = []
+    for entry in entries:
+        # Confidence filter (entry confidence)
+        conf = entry.get("confidence", 0.0)
+        if conf < min_confidence:
+            continue
+
+        # Model filter: if specified, entry's model must be in the list
+        if model_filter:
+            entry_model = entry.get("model", entry.get("provenance", {}).get("model", "unknown"))
+            if entry_model not in model_filter:
+                continue
+
+        # Date filter: use last_confirmed or first_seen or harvested_at
+        entry_date = None
+        for field in ("last_confirmed", "first_seen", "harvested_at"):
+            if field in entry:
+                entry_date = parse_date(entry[field])
+                if entry_date:
+                    break
+        if after and entry_date and entry_date < after:
+            continue
+        if before and entry_date and entry_date > before:
+            continue
+
+        filtered.append(entry)
+    return filtered
+
+
+def entry_to_pair(entry: dict) -> dict:
+    """Convert a knowledge entry into a training pair."""
+    fact = entry.get("fact", "").strip()
+    if not fact:
+        return None
+
+    category = entry.get("category", "fact")
+    domain = entry.get("domain", "global")
+
+    terse = fact_to_terse(fact, category, domain)
+    rich = fact
+    source_confidence = round(entry.get("confidence", 0.0), 4)
+    source_model = entry.get("model", entry.get("provenance", {}).get("model", "unknown"))
+
+    return {
+        "terse": terse,
+        "rich": rich,
+        "domain": domain,
+        "source_confidence": source_confidence,
+        "source_model": source_model,
+    }
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Knowledge entries → training pairs")
+    parser.add_argument("--input", "-i", default="knowledge/index.json",
+                        help="Input knowledge index or JSONL (default: knowledge/index.json)")
+    parser.add_argument("--output", "-o", default="training_pairs.jsonl",
+                        help="Output JSONL file")
+    parser.add_argument("--min-confidence", type=float, default=0.5,
+                        help="Minimum entry confidence to include (0.0-1.0, default: 0.5)")
+    parser.add_argument("--model-filter",
+                        help="Comma-separated list of source models to include")
+    parser.add_argument("--after",
+                        help="Include entries last_confirmed/first_seen on or after this date (YYYY-MM-DD)")
+    parser.add_argument("--before",
+                        help="Include entries last_confirmed/first_seen on or before this date (YYYY-MM-DD)")
+    parser.add_argument("--dry-run", action="store_true",
+                        help="Print sample pairs and stats without writing")
+    args = parser.parse_args()
+
+    # Load
+    entries = load_knowledge_index(args.input)
+    print(f"Loaded {len(entries)} entries from {args.input}", file=sys.stderr)
+
+    # Parse filters
+    model_list = args.model_filter.split(",") if args.model_filter else None
+    after_dt = parse_date(args.after) if args.after else None
+    before_dt = parse_date(args.before) if args.before else None
+
+    # Filter
+    kept = filter_entries(
+        entries,
+        min_confidence=args.min_confidence,
+        model_filter=model_list,
+        after=after_dt,
+        before=before_dt,
+    )
+    print(f"After filtering: {len(kept)} / {len(entries)} entries", file=sys.stderr)
+
+    # Convert
+    pairs = []
+    for entry in kept:
+        pair = entry_to_pair(entry)
+        if pair:
+            pairs.append(pair)
+
+    # Stats
+    if pairs:
+        avg_conf = sum(p["source_confidence"] for p in pairs) / len(pairs)
+        domains = {}
+        models = {}
+        for p in pairs:
+            domains[p["domain"]] = domains.get(p["domain"], 0) + 1
+            models[p["source_model"]] = models.get(p["source_model"], 0) + 1
+    else:
+        avg_conf = 0.0
+        domains = {}
+        models = {}
+
+    stats = {
+        "input_entries": len(entries),
+        "after_filter": len(kept),
+        "pairs_generated": len(pairs),
+        "avg_confidence": round(avg_conf, 4),
+        "domains": domains,
+        "source_models": models,
+    }
+    print(json.dumps(stats, indent=2), file=sys.stderr)
+
+    if args.dry_run:
+        print("\nSample pairs:", file=sys.stderr)
+        for p in pairs[:3]:
+            print(json.dumps(p, ensure_ascii=False), file=sys.stderr)
+        return
+
+    # Write JSONL
+    out_path = Path(args.output)
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    with open(out_path, "w", encoding="utf-8") as f:
+        for pair in pairs:
+            f.write(json.dumps(pair, ensure_ascii=False) + "\n")
+
+    print(f"\nWrote {len(pairs)} training pairs to {out_path}", file=sys.stderr)
+
+
+if __name__ == "__main__":
+    main()

scripts/pr_complexity_scorer.py

@@ -0,0 +1,351 @@
+#!/usr/bin/env python3
+"""
+PR Complexity Scorer - Estimate review effort for PRs.
+"""
+
+import argparse
+import json
+import os
+import re
+import sys
+from dataclasses import dataclass, asdict
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+import urllib.request
+import urllib.error
+
+GITEA_BASE = "https://forge.alexanderwhitestone.com/api/v1"
+
+DEPENDENCY_FILES = {
+    "requirements.txt", "pyproject.toml", "setup.py", "setup.cfg",
+    "Pipfile", "poetry.lock", "package.json", "yarn.lock", "Gemfile",
+    "go.mod", "Cargo.toml", "pom.xml", "build.gradle"
+}
+
+TEST_PATTERNS = [
+    r"tests?/.*\.py$", r".*_test\.py$", r"test_.*\.py$",
+    r"spec/.*\.rb$", r".*_spec\.rb$",
+    r"__tests__/", r".*\.test\.(js|ts|jsx|tsx)$"
+]
+
+WEIGHT_FILES = 0.25
+WEIGHT_LINES = 0.25
+WEIGHT_DEPS = 0.30
+WEIGHT_TEST_COV = 0.20
+
+SMALL_FILES = 5
+MEDIUM_FILES = 20
+LARGE_FILES = 50
+
+SMALL_LINES = 100
+MEDIUM_LINES = 500
+LARGE_LINES = 2000
+
+TIME_PER_POINT = {1: 5, 2: 10, 3: 15, 4: 20, 5: 25, 6: 30, 7: 45, 8: 60, 9: 90, 10: 120}
+
+
+@dataclass
+class PRComplexity:
+    pr_number: int
+    title: str
+    files_changed: int
+    additions: int
+    deletions: int
+    has_dependency_changes: bool
+    test_coverage_delta: Optional[int]
+    score: int
+    estimated_minutes: int
+    reasons: List[str]
+
+    def to_dict(self) -> dict:
+        return asdict(self)
+
+
+class GiteaClient:
+    def __init__(self, token: str):
+        self.token = token
+        self.base_url = GITEA_BASE.rstrip("/")
+
+    def _request(self, path: str, params: Dict = None) -> Any:
+        url = f"{self.base_url}{path}"
+        if params:
+            qs = "&".join(f"{k}={v}" for k, v in params.items() if v is not None)
+            url += f"?{qs}"
+
+        req = urllib.request.Request(url)
+        req.add_header("Authorization", f"token {self.token}")
+        req.add_header("Content-Type", "application/json")
+
+        try:
+            with urllib.request.urlopen(req, timeout=30) as resp:
+                return json.loads(resp.read().decode())
+        except urllib.error.HTTPError as e:
+            print(f"API error {e.code}: {e.read().decode()[:200]}", file=sys.stderr)
+            return None
+        except urllib.error.URLError as e:
+            print(f"Network error: {e}", file=sys.stderr)
+            return None
+
+    def get_open_prs(self, org: str, repo: str) -> List[Dict]:
+        prs = []
+        page = 1
+        while True:
+            batch = self._request(f"/repos/{org}/{repo}/pulls", {"limit": 50, "page": page, "state": "open"})
+            if not batch:
+                break
+            prs.extend(batch)
+            if len(batch) < 50:
+                break
+            page += 1
+        return prs
+
+    def get_pr_files(self, org: str, repo: str, pr_number: int) -> List[Dict]:
+        files = []
+        page = 1
+        while True:
+            batch = self._request(
+                f"/repos/{org}/{repo}/pulls/{pr_number}/files",
+                {"limit": 100, "page": page}
+            )
+            if not batch:
+                break
+            files.extend(batch)
+            if len(batch) < 100:
+                break
+            page += 1
+        return files
+
+    def post_comment(self, org: str, repo: str, pr_number: int, body: str) -> bool:
+        data = json.dumps({"body": body}).encode("utf-8")
+        req = urllib.request.Request(
+            f"{self.base_url}/repos/{org}/{repo}/issues/{pr_number}/comments",
+            data=data,
+            method="POST",
+            headers={"Authorization": f"token {self.token}", "Content-Type": "application/json"}
+        )
+        try:
+            with urllib.request.urlopen(req, timeout=30) as resp:
+                return resp.status in (200, 201)
+        except urllib.error.HTTPError:
+            return False
+
+
+def is_dependency_file(filename: str) -> bool:
+    return any(filename.endswith(dep) for dep in DEPENDENCY_FILES)
+
+
+def is_test_file(filename: str) -> bool:
+    return any(re.search(pattern, filename) for pattern in TEST_PATTERNS)
+
+
+def score_pr(
+    files_changed: int,
+    additions: int,
+    deletions: int,
+    has_dependency_changes: bool,
+    test_coverage_delta: Optional[int] = None
+) -> tuple[int, int, List[str]]:
+    score = 1.0
+    reasons = []
+
+    # Files changed
+    if files_changed <= SMALL_FILES:
+        fscore = 1.0
+        reasons.append("small number of files changed")
+    elif files_changed <= MEDIUM_FILES:
+        fscore = 2.0
+        reasons.append("moderate number of files changed")
+    elif files_changed <= LARGE_FILES:
+        fscore = 2.5
+        reasons.append("large number of files changed")
+    else:
+        fscore = 3.0
+        reasons.append("very large PR spanning many files")
+
+    # Lines changed
+    total_lines = additions + deletions
+    if total_lines <= SMALL_LINES:
+        lscore = 1.0
+        reasons.append("small change size")
+    elif total_lines <= MEDIUM_LINES:
+        lscore = 2.0
+        reasons.append("moderate change size")
+    elif total_lines <= LARGE_LINES:
+        lscore = 3.0
+        reasons.append("large change size")
+    else:
+        lscore = 4.0
+        reasons.append("very large change")
+
+    # Dependency changes
+    if has_dependency_changes:
+        dscore = 2.5
+        reasons.append("dependency changes (architectural impact)")
+    else:
+        dscore = 0.0
+
+    # Test coverage delta
+    tscore = 0.0
+    if test_coverage_delta is not None:
+        if test_coverage_delta > 0:
+            reasons.append(f"test additions (+{test_coverage_delta} test files)")
+            tscore = -min(2.0, test_coverage_delta / 2.0)
+        elif test_coverage_delta < 0:
+            reasons.append(f"test removals ({abs(test_coverage_delta)} test files)")
+            tscore = min(2.0, abs(test_coverage_delta) * 0.5)
+    else:
+        reasons.append("test coverage change not assessed")
+
+    # Weighted sum, scaled by 3 to use full 1-10 range
+    bonus = (fscore * WEIGHT_FILES) + (lscore * WEIGHT_LINES) + (dscore * WEIGHT_DEPS) + (tscore * WEIGHT_TEST_COV)
+    scaled_bonus = bonus * 3.0
+    score = 1.0 + scaled_bonus
+
+    final_score = max(1, min(10, int(round(score))))
+    est_minutes = TIME_PER_POINT.get(final_score, 30)
+
+    return final_score, est_minutes, reasons
+
+
+def analyze_pr(client: GiteaClient, org: str, repo: str, pr_data: Dict) -> PRComplexity:
+    pr_num = pr_data["number"]
+    title = pr_data.get("title", "")
+    files = client.get_pr_files(org, repo, pr_num)
+
+    additions = sum(f.get("additions", 0) for f in files)
+    deletions = sum(f.get("deletions", 0) for f in files)
+    filenames = [f.get("filename", "") for f in files]
+
+    has_deps = any(is_dependency_file(f) for f in filenames)
+
+    test_added = sum(1 for f in files if f.get("status") == "added" and is_test_file(f.get("filename", "")))
+    test_removed = sum(1 for f in files if f.get("status") == "removed" and is_test_file(f.get("filename", "")))
+    test_delta = test_added - test_removed if (test_added or test_removed) else None
+
+    score, est_min, reasons = score_pr(
+        files_changed=len(files),
+        additions=additions,
+        deletions=deletions,
+        has_dependency_changes=has_deps,
+        test_coverage_delta=test_delta
+    )
+
+    return PRComplexity(
+        pr_number=pr_num,
+        title=title,
+        files_changed=len(files),
+        additions=additions,
+        deletions=deletions,
+        has_dependency_changes=has_deps,
+        test_coverage_delta=test_delta,
+        score=score,
+        estimated_minutes=est_min,
+        reasons=reasons
+    )
+
+
+def build_comment(complexity: PRComplexity) -> str:
+    change_desc = f"{complexity.files_changed} files, +{complexity.additions}/-{complexity.deletions} lines"
+    deps_note = "\n- :warning: Dependency changes detected — architectural review recommended" if complexity.has_dependency_changes else ""
+    test_note = ""
+    if complexity.test_coverage_delta is not None:
+        if complexity.test_coverage_delta > 0:
+            test_note = f"\n- :+1: {complexity.test_coverage_delta} test file(s) added"
+        elif complexity.test_coverage_delta < 0:
+            test_note = f"\n- :warning: {abs(complexity.test_coverage_delta)} test file(s) removed"
+
+    comment = f"## 📊 PR Complexity Analysis\n\n"
+    comment += f"**PR #{complexity.pr_number}: {complexity.title}**\n\n"
+    comment += f"| Metric | Value |\n|--------|-------|\n"
+    comment += f"| Changes | {change_desc} |\n"
+    comment += f"| Complexity Score | **{complexity.score}/10** |\n"
+    comment += f"| Estimated Review Time | ~{complexity.estimated_minutes} minutes |\n\n"
+    comment += f"### Scoring rationale:"
+    for r in complexity.reasons:
+        comment += f"\n- {r}"
+    if deps_note:
+        comment += deps_note
+    if test_note:
+        comment += test_note
+    comment += f"\n\n---\n"
+    comment += f"*Generated by PR Complexity Scorer — [issue #135](https://forge.alexanderwhitestone.com/Timmy_Foundation/compounding-intelligence/issues/135)*"
+    return comment
+
+
+def main():
+    parser = argparse.ArgumentParser(description="PR Complexity Scorer")
+    parser.add_argument("--org", default="Timmy_Foundation")
+    parser.add_argument("--repo", default="compounding-intelligence")
+    parser.add_argument("--token", default=os.environ.get("GITEA_TOKEN") or os.path.expanduser("~/.config/gitea/token"))
+    parser.add_argument("--dry-run", action="store_true")
+    parser.add_argument("--apply", action="store_true")
+    parser.add_argument("--output", default="metrics/pr_complexity.json")
+    args = parser.parse_args()
+
+    token_path = args.token
+    if os.path.exists(token_path):
+        with open(token_path) as f:
+            token = f.read().strip()
+    else:
+        token = args.token
+
+    if not token:
+        print("ERROR: No Gitea token provided", file=sys.stderr)
+        sys.exit(1)
+
+    client = GiteaClient(token)
+
+    print(f"Fetching open PRs for {args.org}/{args.repo}...")
+    prs = client.get_open_prs(args.org, args.repo)
+    if not prs:
+        print("No open PRs found.")
+        sys.exit(0)
+
+    print(f"Found {len(prs)} open PR(s). Analyzing...")
+
+    results = []
+    Path(args.output).parent.mkdir(parents=True, exist_ok=True)
+
+    for pr in prs:
+        pr_num = pr["number"]
+        title = pr.get("title", "")
+        print(f"  Analyzing PR #{pr_num}: {title[:60]}")
+
+        try:
+            complexity = analyze_pr(client, args.org, args.repo, pr)
+            results.append(complexity.to_dict())
+
+            comment = build_comment(complexity)
+
+            if args.dry_run:
+                print(f"    → Score: {complexity.score}/10, Est: {complexity.estimated_minutes}min [DRY-RUN]")
+            elif args.apply:
+                success = client.post_comment(args.org, args.repo, pr_num, comment)
+                status = "[commented]" if success else "[FAILED]"
+                print(f"    → Score: {complexity.score}/10, Est: {complexity.estimated_minutes}min {status}")
+            else:
+                print(f"    → Score: {complexity.score}/10, Est: {complexity.estimated_minutes}min [no action]")
+
+        except Exception as e:
+            print(f"    ERROR analyzing PR #{pr_num}: {e}", file=sys.stderr)
+
+    with open(args.output, "w") as f:
+        json.dump({
+            "org": args.org,
+            "repo": args.repo,
+            "timestamp": datetime.now(timezone.utc).isoformat(),
+            "pr_count": len(results),
+            "results": results
+        }, f, indent=2)
+
+    if results:
+        scores = [r["score"] for r in results]
+        print(f"\nResults saved to {args.output}")
+        print(f"Summary: {len(results)} PRs, scores range {min(scores):.0f}-{max(scores):.0f}")
+    else:
+        print("\nNo results to save.")
+
+
+if __name__ == "__main__":
+    main()

scripts/session_pair_harvester.py

@@ -22,95 +22,114 @@ import sys
 from pathlib import Path
 from typing import Optional
 
-from session_reader import extract_conversation, read_session
-
 
 def compute_hash(text: str) -> str:
     """Content hash for deduplication."""
     return hashlib.sha256(text.encode()).hexdigest()[:16]
 
 
-def extract_pairs_from_conversation(conversation: list, session_id: str, model: str,
-                                min_ratio: float = 1.5,
+def extract_pairs_from_session(session_data: dict, min_ratio: float = 1.5,
                                 min_response_words: int = 20) -> list:
-    """Extract terse→rich pairs from a normalized conversation."""
+    """Extract terse→rich pairs from a single session object."""
     pairs = []
+    conversations = session_data.get("conversations", [])
+    session_id = session_data.get("id", "unknown")
+    model = session_data.get("model", "unknown")
+
     seen_hashes = set()
 
-    for i, msg in enumerate(conversation):
-        # Look for assistant responses
-        if msg.get('role') != 'assistant':
+    for i, msg in enumerate(conversations):
+        # Look for assistant/gpt responses
+        if msg.get("from") not in ("gpt", "assistant"):
             continue
 
-        response_text = msg.get('content', '')
+        response_text = msg.get("value", "")
         if not response_text or len(response_text.split()) < min_response_words:
             continue
 
-        # Find the preceding user message
+        # Find the preceding human message
         prompt_text = ""
         for j in range(i - 1, -1, -1):
-            if conversation[j].get('role') == 'user':
-                prompt_text = conversation[j].get('content', '')
+            if conversations[j].get("from") == "human":
+                prompt_text = conversations[j].get("value", "")
                 break
 
         if not prompt_text:
             continue
 
         # Filter: skip tool results, system messages embedded as human
-        if prompt_text.startswith('{') and 'output' in prompt_text[:100]:
-            continue
-        if prompt_text.startswith('# SOUL.md') or prompt_text.startswith('You are'):
-            continue
+        if prompt_text.startswith("{") and "output" in prompt_text[:100]:
+            continue  # likely a tool result
+        if prompt_text.startswith("# SOUL.md") or prompt_text.startswith("You are"):
+            continue  # system prompt leak
 
         # Quality filters
         prompt_words = len(prompt_text.split())
         response_words = len(response_text.split())
 
+        # Must have meaningful length ratio
         if prompt_words == 0 or response_words == 0:
             continue
         ratio = response_words / prompt_words
         if ratio < min_ratio:
             continue
 
-        code_blocks = response_text.count('```')
-        if code_blocks >= 4 and len(response_text.replace('```', '').strip()) < 50:
+        # Skip responses that are mostly code
+        code_blocks = response_text.count("```")
+        if code_blocks >= 4 and len(response_text.replace("```", "").strip()) < 50:
             continue
 
-        if 'tool_call' in response_text[:100] or 'function_call' in response_text[:100]:
+        # Skip responses with tool call artifacts
+        if "tool_call" in response_text[:100] or "function_call" in response_text[:100]:
             continue
 
+        # Deduplicate by content hash
         content_hash = compute_hash(prompt_text + response_text[:200])
         if content_hash in seen_hashes:
             continue
         seen_hashes.add(content_hash)
 
+        # Clean up response: remove markdown headers if too many
         clean_response = response_text
 
         pairs.append({
-            'terse': prompt_text.strip(),
-            'rich': clean_response.strip(),
-            'source': session_id,
-            'model': model,
-            'prompt_words': prompt_words,
-            'response_words': response_words,
-            'ratio': round(ratio, 2),
+            "terse": prompt_text.strip(),
+            "rich": clean_response.strip(),
+            "source": session_id,
+            "model": model,
+            "prompt_words": prompt_words,
+            "response_words": response_words,
+            "ratio": round(ratio, 2),
         })
 
     return pairs
 
 
+def extract_from_jsonl_file(filepath: str, **kwargs) -> list:
+    """Extract pairs from a session JSONL file."""
+    pairs = []
+    path = Path(filepath)
 
-def extract_from_jsonl_file(path: str, **kwargs) -> list:
-    """Read a session file and extract training pairs using normalized conversation."""
-    session_messages = read_session(path)
-    if not session_messages:
-        return []
-    conversation = extract_conversation(session_messages)
-    # Derive session_id and model from first real message metadata
-    first_msg = next((m for m in session_messages if m.get('role') or m.get('from')), {})
-    session_id = first_msg.get('meta_session_id', Path(path).name)
-    model = first_msg.get('model', 'unknown')
-    return extract_pairs_from_conversation(conversation, session_id, model, **kwargs)
+    if not path.exists():
+        print(f"Warning: {filepath} not found", file=sys.stderr)
+        return pairs
+
+    content = path.read_text()
+    lines = content.strip().split("\n")
+
+    for line in lines:
+        line = line.strip()
+        if not line:
+            continue
+        try:
+            session = json.loads(line)
+        except json.JSONDecodeError:
+            continue
+
+        session_pairs = extract_pairs_from_session(session, **kwargs)
+        pairs.extend(session_pairs)
+
+    return pairs
 
 
 def deduplicate_pairs(pairs: list) -> list:

scripts/test_pr_complexity_scorer.py

@@ -0,0 +1,170 @@
+#!/usr/bin/env python3
+"""
+Tests for PR Complexity Scorer — unit tests for the scoring logic.
+"""
+
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).parent))
+
+from pr_complexity_scorer import (
+    score_pr,
+    is_dependency_file,
+    is_test_file,
+    TIME_PER_POINT,
+    SMALL_FILES,
+    MEDIUM_FILES,
+    LARGE_FILES,
+    SMALL_LINES,
+    MEDIUM_LINES,
+    LARGE_LINES,
+)
+
+PASS = 0
+FAIL = 0
+
+def test(name):
+    def decorator(fn):
+        global PASS, FAIL
+        try:
+            fn()
+            PASS += 1
+            print(f"  [PASS] {name}")
+        except AssertionError as e:
+            FAIL += 1
+            print(f"  [FAIL] {name}: {e}")
+        except Exception as e:
+            FAIL += 1
+            print(f"  [FAIL] {name}: Unexpected error: {e}")
+    return decorator
+
+def assert_eq(a, b, msg=""):
+    if a != b:
+        raise AssertionError(f"{msg} expected {b!r}, got {a!r}")
+
+def assert_true(v, msg=""):
+    if not v:
+        raise AssertionError(msg or "Expected True")
+
+def assert_false(v, msg=""):
+    if v:
+        raise AssertionError(msg or "Expected False")
+
+
+print("=== PR Complexity Scorer Tests ===\n")
+
+print("-- File Classification --")
+
+@test("dependency file detection — requirements.txt")
+def _():
+    assert_true(is_dependency_file("requirements.txt"))
+    assert_true(is_dependency_file("src/requirements.txt"))
+    assert_false(is_dependency_file("requirements_test.txt"))
+
+@test("dependency file detection — pyproject.toml")
+def _():
+    assert_true(is_dependency_file("pyproject.toml"))
+    assert_false(is_dependency_file("myproject.py"))
+
+@test("test file detection — pytest style")
+def _():
+    assert_true(is_test_file("tests/test_api.py"))
+    assert_true(is_test_file("test_module.py"))
+    assert_true(is_test_file("src/module_test.py"))
+
+@test("test file detection — other frameworks")
+def _():
+    assert_true(is_test_file("spec/feature_spec.rb"))
+    assert_true(is_test_file("__tests__/component.test.js"))
+    assert_false(is_test_file("testfixtures/helper.py"))
+
+
+print("\n-- Scoring Logic --")
+
+@test("small PR gets low score (1-3)")
+def _():
+    score, minutes, _ = score_pr(
+        files_changed=3,
+        additions=50,
+        deletions=10,
+        has_dependency_changes=False,
+        test_coverage_delta=None
+    )
+    assert_true(1 <= score <= 3, f"Score should be low, got {score}")
+    assert_true(minutes < 20)
+
+@test("medium PR gets medium score (4-6)")
+def _():
+    score, minutes, _ = score_pr(
+        files_changed=15,
+        additions=400,
+        deletions=100,
+        has_dependency_changes=False,
+        test_coverage_delta=None
+    )
+    assert_true(4 <= score <= 6, f"Score should be medium, got {score}")
+    assert_true(20 <= minutes <= 45)
+
+@test("large PR gets high score (7-9)")
+def _():
+    score, minutes, _ = score_pr(
+        files_changed=60,
+        additions=3000,
+        deletions=1500,
+        has_dependency_changes=True,
+        test_coverage_delta=None
+    )
+    assert_true(7 <= score <= 9, f"Score should be high, got {score}")
+    assert_true(minutes >= 45)
+
+@test("dependency changes boost score")
+def _():
+    base_score, _, _ = score_pr(
+        files_changed=10, additions=200, deletions=50,
+        has_dependency_changes=False, test_coverage_delta=None
+    )
+    dep_score, _, _ = score_pr(
+        files_changed=10, additions=200, deletions=50,
+        has_dependency_changes=True, test_coverage_delta=None
+    )
+    assert_true(dep_score > base_score, f"Deps: {base_score} -> {dep_score}")
+
+@test("adding tests lowers complexity")
+def _():
+    base_score, _, _ = score_pr(
+        files_changed=8, additions=150, deletions=20,
+        has_dependency_changes=False, test_coverage_delta=None
+    )
+    better_score, _, _ = score_pr(
+        files_changed=8, additions=180, deletions=20,
+        has_dependency_changes=False, test_coverage_delta=3
+    )
+    assert_true(better_score < base_score, f"Tests: {base_score} -> {better_score}")
+
+@test("removing tests increases complexity")
+def _():
+    base_score, _, _ = score_pr(
+        files_changed=8, additions=150, deletions=20,
+        has_dependency_changes=False, test_coverage_delta=None
+    )
+    worse_score, _, _ = score_pr(
+        files_changed=8, additions=150, deletions=20,
+        has_dependency_changes=False, test_coverage_delta=-2
+    )
+    assert_true(worse_score > base_score, f"Remove tests: {base_score} -> {worse_score}")
+
+@test("score bounded 1-10")
+def _():
+    for files, adds, dels in [(1, 10, 5), (100, 10000, 5000)]:
+        score, _, _ = score_pr(files, adds, dels, False, None)
+        assert_true(1 <= score <= 10, f"Score {score} out of range")
+
+@test("estimated minutes exist for all scores")
+def _():
+    for s in range(1, 11):
+        assert_true(s in TIME_PER_POINT, f"Missing time for score {s}")
+
+
+print(f"\n=== Results: {PASS} passed, {FAIL} failed ===")
+sys.exit(0 if FAIL == 0 else 1)

tests/test_knowledge_to_training_pairs.py

@@ -0,0 +1,174 @@
+#!/usr/bin/env python3
+"""
+Smoke tests for knowledge_to_training_pairs.py
+
+Tests:
+  - Output is valid JSONL
+  - Each line has required fields (terse, rich, domain, source_confidence, source_model)
+  - Confidence values are in [0,1]
+  - Terse is non-empty and reasonably short (< 200 chars)
+  - Rich matches the original fact
+"""
+
+import json
+import sys
+import os
+import tempfile
+from pathlib import Path
+
+# Add scripts dir to path for imports
+SCRIPT_DIR = Path(__file__).parent.parent / "scripts"
+sys.path.insert(0, str(SCRIPT_DIR))
+
+from knowledge_to_training_pairs import (
+    fact_to_terse,
+    filter_entries,
+    entry_to_pair,
+    parse_date,
+)
+
+
+def test_fact_to_terse_pitfall():
+    fact = "deploy-crons.py leaves jobs in mixed model format"
+    category = "pitfall"
+    domain = "hermes-agent"
+    terse = fact_to_terse(fact, category, domain)
+    assert terse.startswith("How do I")
+    assert "?" in terse
+    assert len(terse) < 150
+    print("PASS: test_fact_to_terse_pitfall")
+
+
+def test_fact_to_terse_fact():
+    fact = "Python is a high-level programming language"
+    terse = fact_to_terse(fact, "fact", "global")
+    assert terse.startswith("What should I know about")
+    assert "?" in terse
+    print("PASS: test_fact_to_terse_fact")
+
+
+def test_fact_to_terse_pattern():
+    fact = "Use sparse checkout for large repos"
+    terse = fact_to_terse(fact, "pattern", "devops")
+    assert "recommended way" in terse or "best way" in terse
+    print("PASS: test_fact_to_terse_pattern")
+
+
+def test_entry_to_pair_structure():
+    entry = {
+        "id": "test:001",
+        "fact": "Test fact text.",
+        "category": "fact",
+        "domain": "test-domain",
+        "confidence": 0.85,
+        "model": "test-model",
+    }
+    pair = entry_to_pair(entry)
+    assert pair is not None
+    assert "terse" in pair
+    assert "rich" in pair
+    assert "domain" in pair
+    assert "source_confidence" in pair
+    assert "source_model" in pair
+    assert pair["rich"] == "Test fact text."
+    assert pair["domain"] == "test-domain"
+    assert 0.0 <= pair["source_confidence"] <= 1.0
+    print("PASS: test_entry_to_pair_structure")
+
+
+def test_filter_by_confidence():
+    entries = [
+        {"fact": "A", "confidence": 0.9},
+        {"fact": "B", "confidence": 0.4},
+        {"fact": "C", "confidence": 0.6},
+    ]
+    filtered = filter_entries(entries, min_confidence=0.5)
+    assert len(filtered) == 2
+    assert all(e["confidence"] >= 0.5 for e in filtered)
+    print("PASS: test_filter_by_confidence")
+
+
+def test_filter_by_model():
+    entries = [
+        {"fact": "A", "model": "claude-sonnet"},
+        {"fact": "B", "model": "gpt-4"},
+        {"fact": "C", "model": "unknown"},
+    ]
+    filtered = filter_entries(entries, model_filter=["claude-sonnet", "gpt-4"])
+    assert len(filtered) == 2
+    assert all(e["model"] in ("claude-sonnet", "gpt-4") for e in filtered)
+    print("PASS: test_filter_by_model")
+
+
+def test_filter_by_date():
+    entries = [
+        {"fact": "A", "last_confirmed": "2026-04-10"},
+        {"fact": "B", "last_confirmed": "2026-03-01"},
+        {"fact": "C", "first_seen": "2026-04-15"},
+    ]
+    after_dt = parse_date("2026-04-01")
+    filtered = filter_entries(entries, after=after_dt)
+    assert len(filtered) == 2
+    print("PASS: test_filter_by_date")
+
+
+def test_end_to_end_jsonl_output():
+    """Integration test: run the script and verify JSONL validity."""
+    import subprocess
+
+    repo_dir = SCRIPT_DIR.parent
+    result = subprocess.run(
+        ["python3", "scripts/knowledge_to_training_pairs.py", "--dry-run"],
+        capture_output=True, text=True, cwd=repo_dir
+    )
+    assert result.returncode == 0
+    stderr = result.stderr.strip()
+
+    # The stats JSON object is at the top of stderr. Find its bounds via brace matching.
+    start = stderr.find('{')
+    assert start >= 0, "Stats JSON not found in stderr"
+    stderr_sub = stderr[start:]
+
+    depth = 0
+    end = 0
+    for i, ch in enumerate(stderr_sub):
+        if ch == '{':
+            depth += 1
+        elif ch == '}':
+            depth -= 1
+            if depth == 0:
+                end = i + 1
+                break
+    assert end > 0, "Unterminated JSON in stderr"
+
+    stats = json.loads(stderr_sub[:end])
+    assert stats["input_entries"] > 0
+    assert stats["pairs_generated"] > 0
+    print("PASS: test_end_to_end_jsonl_output")
+
+
+def test_terse_length_constraint():
+    """Terse should be reasonably short for training."""
+    # Sample facts from actual knowledge
+    test_facts = [
+        ("deploy-crons.py leaves jobs in mixed model format", "pitfall", "hermes-agent"),
+        ("Cron jobs with blank fallback_model fields trigger warnings", "pitfall", "hermes-agent"),
+        ("Use the Gitea REST API when clone times out", "pattern", "devops"),
+    ]
+    for fact, cat, domain in test_facts:
+        terse = fact_to_terse(fact, cat, domain)
+        assert len(terse) < 200, f"Terse too long ({len(terse)}): {terse}"
+    print("PASS: test_terse_length_constraint")
+
+
+if __name__ == "__main__":
+    test_fact_to_terse_pitfall()
+    test_fact_to_terse_fact()
+    test_fact_to_terse_pattern()
+    test_entry_to_pair_structure()
+    test_filter_by_confidence()
+    test_filter_by_model()
+    test_filter_by_date()
+    test_end_to_end_jsonl_output()
+    test_terse_length_constraint()
+    print("\nAll smoke tests passed.")

tests/test_session_pair_harvester.py

@@ -1,118 +0,0 @@
-"""
-Tests for session_pair_harvester — training pair extraction from sessions.
-"""
-
-import json
-import tempfile
-import unittest
-from pathlib import Path
-
-import sys
-from pathlib import Path
-sys.path.insert(0, str(Path(__file__).parent.parent / "scripts"))
-from session_pair_harvester import (
-    extract_pairs_from_conversation,
-    extract_from_jsonl_file,
-    deduplicate_pairs,
-    compute_hash,
-)
-
-
-class TestSessionPairHarvester(unittest.TestCase):
-    def test_compute_hash_consistent(self):
-        h1 = compute_hash("hello world")
-        h2 = compute_hash("hello world")
-        self.assertEqual(h1, h2)
-        self.assertEqual(len(h1), 16)
-
-    def test_extract_simple_qa_pair(self):
-        """A simple user→assistant exchange produces one pair."""
-        conversation = [
-            {"role": "user", "content": "What is the capital of France?"},
-            {"role": "assistant", "content": "The capital of France is Paris. It is a major European city renowned for its art, fashion, gastronomy, cultural heritage, and historical significance. The city attracts millions of tourists annually."},
-        ]
-        pairs = extract_pairs_from_conversation(conversation, "test_session", "test-model")
-        self.assertEqual(len(pairs), 1)
-        self.assertEqual(pairs[0]["terse"], "What is the capital of France?")
-        self.assertIn("Paris", pairs[0]["rich"])
-        self.assertEqual(pairs[0]["source"], "test_session")
-
-    def test_min_ratio_filter(self):
-        """Very short responses are filtered out."""
-        conversation = [
-            {"role": "user", "content": "Yes"},
-            {"role": "assistant", "content": "No."},
-        ]
-        # Default min_ratio = 1.5, min_words = 20 for response
-        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=3)
-        self.assertEqual(len(pairs), 0)
-
-    def test_min_words_filter(self):
-        """Assistant responses below min word count are skipped."""
-        conversation = [
-            {"role": "user", "content": "Explain the project architecture in detail"},
-            {"role": "assistant", "content": "OK."},
-        ]
-        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=5)
-        self.assertEqual(len(pairs), 0)
-
-    def test_skip_non_assistant_messages(self):
-        """System and tool messages are ignored."""
-        conversation = [
-            {"role": "system", "content": "You are a helpful assistant."},
-            {"role": "user", "content": "Hello"},
-            {"role": "assistant", "content": "Hi there! How can I help you today?"},
-        ]
-        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=3)
-        self.assertEqual(len(pairs), 1)
-        self.assertEqual(pairs[0]["terse"], "Hello")
-
-    def test_multiple_pairs_from_one_session(self):
-        """A conversation with several Q&A turns yields multiple pairs."""
-        conversation = [
-            {"role": "user", "content": "First question?"},
-            {"role": "assistant", "content": "Here is a detailed and comprehensive answer that thoroughly explores multiple aspects of the subject. It provides background context and practical implications for the reader."},
-            {"role": "user", "content": "Second?"},
-            {"role": "assistant", "content": "Another comprehensive response with detailed examples. This includes practical code blocks and thorough explanations to ensure deep understanding of the topic at hand."},
-        ]
-        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_ratio=1.0)
-        self.assertEqual(len(pairs), 2)
-
-    def test_deduplication_removes_duplicates(self):
-        """Identical pairs across sessions are deduplicated."""
-        pairs = [
-            {"terse": "q1", "rich": "a1", "source": "s1", "model": "m"},
-            {"terse": "q1", "rich": "a1", "source": "s2", "model": "m"},
-            {"terse": "q2", "rich": "a2", "source": "s1", "model": "m"},
-        ]
-        unique = deduplicate_pairs(pairs)
-        self.assertEqual(len(unique), 2)
-        sources = {p["source"] for p in unique}
-        # First unique pair can be from either s1 or s2
-        self.assertIn("s1", sources)
-
-    def test_integration_with_test_sessions(self):
-        """Harvester finds pairs in real test session files."""
-        repo_root = Path(__file__).parent.parent
-        test_sessions_dir = repo_root / "test_sessions"
-        if not test_sessions_dir.exists():
-            self.skipTest("test_sessions not found")
-
-        pairs = []
-        for jsonl_file in sorted(test_sessions_dir.glob("*.jsonl")):
-            pairs.extend(extract_from_jsonl_file(str(jsonl_file)))
-
-        self.assertGreater(len(pairs), 0, "Should extract at least one pair from test_sessions")
-        for p in pairs:
-            self.assertIn("terse", p)
-            self.assertIn("rich", p)
-            self.assertIn("source", p)
-            self.assertIn("model", p)
-            # Verify content exists
-            self.assertGreater(len(p["terse"]), 0)
-            self.assertGreater(len(p["rich"]), 0)
-
-
-if __name__ == "__main__":
-    unittest.main()
-