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/progress_tracker.py

@@ -1,477 +0,0 @@
-#!/usr/bin/env python3
-"""
-Progress Tracker — Pipeline 10.8
-Track improvement metrics over time. Are we getting better?
-
-Metrics tracked:
- 1. Test coverage  — % of Python functions with associated tests (test:source file ratio + line coverage if available)
- 2. Doc coverage   — % of Python callables with docstrings (AST-based)
- 3. Issue close rate — closed / (opened + closed) per week (Gitea API)
- 4. Dep freshness  — % of requirements pinned vs outdated (pip list --outdated)
-
-Output:
- - metrics/snapshots/YYYY-MM-DD.json  — one snapshot per run
- - metrics/TRENDS.md                  — cumulative markdown table
- - stdout summary
-
-Usage:
-    python3 scripts/progress_tracker.py
-    python3 scripts/progress_tracker.py --json
-    python3 scripts/progress_tracker.py --output metrics/TRENDS.md
-
-Weekly cron:
-    0 9 * * 1 cd /path/to/compounding-intelligence && python3 scripts/progress_tracker.py
-"""
-
-import argparse
-import json
-import os
-import re
-import subprocess
-import sys
-from collections import defaultdict
-from datetime import datetime, timezone, timedelta
-from pathlib import Path
-from typing import Any, Dict, List, Optional, Tuple
-
-# ── Configuration ──────────────────────────────────────────────────────────
-
-SCRIPT_DIR = Path(__file__).resolve().parent
-REPO_ROOT = SCRIPT_DIR.parent
-METRICS_DIR = REPO_ROOT / "metrics"
-SNAPSHOTS_DIR = METRICS_DIR / "snapshots"
-TOKEN_PATH = Path.home() / ".config" / "gitea" / "token"
-GITEA_API_BASE = "https://forge.alexanderwhitestone.com/api/v1"
-ORG = "Timmy_Foundation"
-
-# Ensure paths exist
-SNAPSHOTS_DIR.mkdir(parents=True, exist_ok=True)
-
-
-# ── Helpers ─────────────────────────────────────────────────────────────────
-
-def run_cmd(cmd: List[str], cwd: Path = REPO_ROOT) -> str:
-    """Run a shell command and return stdout (stderr merged)."""
-    result = subprocess.run(
-        cmd, capture_output=True, text=True, cwd=cwd, timeout=30
-    )
-    if result.returncode != 0:
-        return ""
-    return result.stdout.strip()
-
-
-def slugify_date(dt: datetime) -> str:
-    return dt.strftime("%Y-%m-%d")
-
-
-def snapshot_path(dt: datetime) -> Path:
-    return SNAPSHOTS_DIR / f"{slugify_date(dt)}.json"
-
-
-def load_snapshots() -> List[Dict[str, Any]]:
-    """Load all existing snapshots sorted by date."""
-    snapshots = []
-    for f in sorted(SNAPSHOTS_DIR.glob("*.json")):
-        try:
-            with open(f) as fp:
-                snapshots.append(json.load(fp))
-        except Exception:
-            continue
-    return snapshots
-
-
-# ── Metric 1: Test Coverage ─────────────────────────────────────────────────
-
-def collect_test_coverage() -> Dict[str, Any]:
-    """
-    Compute test coverage metrics.
-    Counts test_*.py and *_test.py files vs non-test .py source files.
-    Also attempts to read .coverage if present.
-    """
-    all_py = list(REPO_ROOT.rglob("*.py"))
-    
-    source_files = []
-    test_files = []
-    
-    for p in all_py:
-        try:
-            rel_parts = p.relative_to(REPO_ROOT).parts
-        except ValueError:
-            continue
-        
-        # Skip hidden/cache/temp dirs (check only relative parts)
-        if any(part.startswith('.') or part.startswith('__') for part in rel_parts):
-            continue
-        if any(part in ('node_modules', 'venv', '.venv', 'env', '.pytest_cache') for part in rel_parts):
-            continue
-        
-        if p.name.startswith("test_") or p.name.endswith("_test.py"):
-            test_files.append(p)
-        else:
-            source_files.append(p)
-    
-    # Try to get line coverage from .coverage
-    coverage_percent = None
-    coverage_tool = None
-    coverage_file = REPO_ROOT / ".coverage"
-    if coverage_file.exists():
-        try:
-            import coverage  # type: ignore
-            # Use coverage API if available
-            cov = coverage.Coverage(data_file=str(coverage_file))
-            cov.load()
-            total = cov.report()
-            coverage_percent = total if isinstance(total, float) else None
-            coverage_tool = "coverage"
-        except Exception:
-            # Fallback: parse `coverage report` output
-            out = run_cmd(["coverage", "report", "--skip-empty"])
-            if out:
-                for line in out.splitlines():
-                    if "TOTAL" in line:
-                        parts = line.split()
-                        if len(parts) >= 2:
-                            try:
-                                coverage_percent = float(parts[-1].rstrip('%'))
-                                coverage_tool = "coverage"
-                                break
-                            except ValueError:
-                                pass
-    
-    return {
-        "test_files": len(test_files),
-        "source_files": len(source_files),
-        "test_to_source_ratio": round(len(test_files) / len(source_files), 4) if source_files else 0.0,
-        "coverage_tool": coverage_tool,
-        "coverage_percent": coverage_percent,
-    }
-
-
-# ── Metric 2: Doc Coverage ──────────────────────────────────────────────────
-
-def collect_doc_coverage() -> Dict[str, Any]:
-    """
-    Check AST of Python files for docstrings.
-    Returns: callables_total, callables_with_doc, doc_coverage_percent
-    """
-    import ast
-    
-    all_py = list(REPO_ROOT.rglob("*.py"))
-    
-    source_files = []
-    test_files = []
-    
-    for p in all_py:
-        try:
-            rel_parts = p.relative_to(REPO_ROOT).parts
-        except ValueError:
-            continue
-        
-        if any(part.startswith('.') or part.startswith('__') for part in rel_parts):
-            continue
-        if any(part in ('node_modules', 'venv', '.venv', 'env', '.pytest_cache') for part in rel_parts):
-            continue
-        
-        if p.name.startswith("test_") or p.name.endswith("_test.py"):
-            test_files.append(p)
-        else:
-            source_files.append(p)
-    
-    total_callables = 0
-    with_doc = 0
-    
-    for p in source_files + test_files:
-        try:
-            with open(p) as f:
-                tree = ast.parse(f.read(), filename=str(p))
-            for node in ast.walk(tree):
-                if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
-                    total_callables += 1
-                    doc = ast.get_docstring(node)
-                    if doc and doc.strip():
-                        with_doc += 1
-        except Exception:
-            continue
-    
-    return {
-        "callables_total": total_callables,
-        "callables_with_doc": with_doc,
-        "doc_coverage_percent": round((with_doc / total_callables * 100) if total_callables else 0.0, 2),
-    }
-
-
-# ── Metric 3: Issue Close Rate ──────────────────────────────────────────────
-
-def collect_issue_metrics() -> Dict[str, Any]:
-    """
-    Use Gitea API to get issue open/close stats for the last 7 days.
-    Returns counts and close rate.
-    """
-    token = ""
-    if TOKEN_PATH.exists():
-        token = TOKEN_PATH.read_text().strip()
-    
-    if not token:
-        return {
-            "opened_last_7d": None,
-            "closed_last_7d": None,
-            "close_rate": None,
-            "total_open": None,
-            "note": "Gitea token not available"
-        }
-    
-    try:
-        from urllib.request import Request, urlopen
-        from urllib.error import HTTPError, URLError
-    except ImportError:
-        return {"error": "urllib not available"}
-    
-    now = datetime.now(timezone.utc)
-    week_ago = now - timedelta(days=7)
-    since = week_ago.strftime("%Y-%m-%d")
-    
-    headers = {"Authorization": f"token {token}"}
-    base_url = f"{GITEA_API_BASE}/repos/{ORG}/compounding-intelligence/issues"
-    
-    try:
-        # Get issues from last 7 days
-        url = f"{base_url}?state=all&since={since}&per_page=100"
-        req = Request(url, headers=headers)
-        with urlopen(req, timeout=15) as resp:
-            issues = json.loads(resp.read())
-        
-        opened = 0
-        closed = 0
-        for issue in issues:
-            created = datetime.fromisoformat(issue["created_at"].replace("Z", "+00:00"))
-            if created >= week_ago:
-                opened += 1
-            if issue.get("state") == "closed":
-                closed_at_str = issue.get("closed_at")
-                if closed_at_str:
-                    closed_at = datetime.fromisoformat(closed_at_str.replace("Z", "+00:00"))
-                    if closed_at >= week_ago:
-                        closed += 1
-        
-        # Total open issues
-        req2 = Request(f"{base_url}?state=open&per_page=1", headers=headers)
-        with urlopen(req2, timeout=15) as resp:
-            total_open = int(resp.headers.get("X-Total-Count", "0"))
-        
-        total = opened + closed
-        close_rate = closed / total if total > 0 else 0.0
-        
-        return {
-            "opened_last_7d": opened,
-            "closed_last_7d": closed,
-            "close_rate": round(close_rate, 4),
-            "total_open": total_open,
-        }
-    except Exception as e:
-        return {
-            "opened_last_7d": None,
-            "closed_last_7d": None,
-            "close_rate": None,
-            "total_open": None,
-            "error": str(e)[:100],
-            "note": "Gitea API unavailable"
-        }
-
-
-# ── Metric 4: Dependency Freshness ─────────────────────────────────────────
-
-def collect_dep_freshness() -> Dict[str, Any]:
-    """
-    Check requirements.txt for outdated dependencies using pip list --outdated.
-    Returns freshness percentage and outdated list.
-    """
-    req_file = REPO_ROOT / "requirements.txt"
-    if not req_file.exists():
-        return {
-            "total_deps": 0,
-            "outdated_deps": 0,
-            "freshness_percent": 100.0,
-            "outdated_list": [],
-            "note": "requirements.txt not found"
-        }
-    
-    # Parse requirements (very simple: take name before comparison op)
-    reqs = []
-    with open(req_file) as f:
-        for line in f:
-            line = line.strip()
-            if not line or line.startswith("#"):
-                continue
-            m = re.match(r"^([a-zA-Z0-9_.-]+)", line)
-            if m:
-                reqs.append(m.group(1))
-    
-    if not reqs:
-        return {"total_deps": 0, "outdated_deps": 0, "freshness_percent": 100.0, "outdated_list": []}
-    
-    # Query pip for outdated packages (may fail if pip not available)
-    outdated_names = set()
-    try:
-        out = run_cmd(["pip", "list", "--outdated", "--format=json"])
-        if out:
-            data = json.loads(out)
-            outdated_names = {item["name"].lower() for item in data}
-    except Exception:
-        pass
-    
-    outdated = [p for p in reqs if p.lower() in outdated_names]
-    total = len(reqs)
-    outdated_count = len(outdated)
-    freshness = round(((total - outdated_count) / total * 100) if total else 100.0, 1)
-    
-    return {
-        "total_deps": total,
-        "outdated_deps": outdated_count,
-        "freshness_percent": freshness,
-        "outdated_list": outdated,
-    }
-
-
-# ── Snapshot & Trends ───────────────────────────────────────────────────────
-
-def take_snapshot() -> Dict[str, Any]:
-    """Collect all metrics and return a snapshot dict."""
-    now = datetime.now(timezone.utc)
-    test_cov = collect_test_coverage()
-    doc_cov = collect_doc_coverage()
-    issues = collect_issue_metrics()
-    deps = collect_dep_freshness()
-    
-    return {
-        "timestamp": now.isoformat(),
-        "date": slugify_date(now),
-        "metrics": {
-            "test_coverage": test_cov,
-            "doc_coverage": doc_cov,
-            "issues": issues,
-            "dependencies": deps,
-        }
-    }
-
-
-def save_snapshot(snapshot: Dict[str, Any]) -> Path:
-    path = snapshot_path(datetime.fromisoformat(snapshot["timestamp"]))
-    with open(path, "w") as f:
-        json.dump(snapshot, f, indent=2)
-    return path
-
-
-def generate_trends(snapshots: List[Dict[str, Any]], output_path: Optional[Path] = None) -> str:
-    """Generate markdown trends table; optionally write to file."""
-    if not snapshots:
-        msg = "# Progress Tracker — Trends\n\nNo snapshots yet. Run `progress_tracker.py` to create the first snapshot."
-        if output_path:
-            output_path.parent.mkdir(parents=True, exist_ok=True)
-            output_path.write_text(msg)
-        return msg
-    
-    lines = [
-        "# Progress Tracker — Trends",
-        f"\nLast updated: {datetime.now(timezone.utc).strftime('%Y-%m-%d %H:%M UTC')}",
-        f"\nSnapshots: {len(snapshots)}\n",
-        "| Date | Test Files → Source | Doc Coverage | Issues Closed/Opened (7d) | Dep Freshness |",
-        "|------|---------------------|--------------|---------------------------|---------------|",
-    ]
-    
-    for snap in reversed(snapshots):  # chronological
-        date = snap["date"]
-        m = snap["metrics"]
-        tc = m["test_coverage"]
-        test_str = f"{tc['test_files']}/{tc['source_files']} ({tc['test_to_source_ratio']:.2f})"
-        doc_str = f"{m['doc_coverage']['doc_coverage_percent']:.1f}%"
-        issues_str = f"{m['issues'].get('closed_last_7d','-')}/{m['issues'].get('opened_last_7d','-')}"
-        dep_str = f"{m['dependencies'].get('freshness_percent','?')}%"
-        lines.append(f"| {date} | {test_str} | {doc_str} | {issues_str} | {dep_str} |")
-    
-    # Current snapshot summary
-    cur = snapshots[-1]
-    cm = cur["metrics"]
-    lines.append(f"\n## Current Snapshot ({cur['date']})\n")
-    
-    tc = cm["test_coverage"]
-    cov_line = f"- Test coverage: {tc['coverage_percent']:.1f}% (via {tc['coverage_tool']})\n" if tc["coverage_percent"] else "- Test coverage: (pytest-cov not configured)\n"
-    lines.append(cov_line)
-    lines.append(f"- Doc coverage: {cm['doc_coverage']['doc_coverage_percent']:.1f}%")
-    
-    im = cm["issues"]
-    if im.get("close_rate") is not None:
-        lines.append(f"- Issue close rate (7d): {im['close_rate']*100:.1f}% ({im['closed_last_7d']} closed, {im['opened_last_7d']} opened)")
-    else:
-        lines.append(f"- Issue metrics: {im.get('note','unavailable')}")
-    
-    dd = cm["dependencies"]
-    lines.append(f"- Dep freshness: {dd.get('freshness_percent','?')}% outdated ({dd.get('outdated_deps',0)}/{dd.get('total_deps',0)} deps)")
-    if dd.get('outdated_list'):
-        lines.append(f"  Outdated: {', '.join(dd['outdated_list'][:5])}")
-    
-    content = "\n".join(lines) + "\n"
-    
-    if output_path:
-        output_path.parent.mkdir(parents=True, exist_ok=True)
-        output_path.write_text(content)
-    
-    return content
-
-
-# ── Main ─────────────────────────────────────────────────────────────────────
-
-def main() -> int:
-    parser = argparse.ArgumentParser(description="Progress Tracker — 10.8")
-    parser.add_argument("--json", action="store_true", help="Emit snapshot as JSON only")
-    parser.add_argument("--output", type=Path, default=METRICS_DIR / "TRENDS.md",
-                        help="Write trends markdown to this file")
-    args = parser.parse_args()
-    
-    snapshot = take_snapshot()
-    all_snapshots = load_snapshots()
-    path_written = save_snapshot(snapshot)
-    
-    if args.json:
-        print(json.dumps(snapshot, indent=2))
-        return 0
-    
-    trends = generate_trends(all_snapshots + [snapshot], output_path=args.output)
-    
-    # Print current snapshot summary
-    print(f"Snapshot saved: {path_written}\n")
-    print(f"Progress Tracker — {snapshot['date']}")
-    print("=" * 50)
-    
-    m = snapshot["metrics"]
-    tc = m["test_coverage"]
-    print(f"Test files: {tc['test_files']}  |  Source files: {tc['source_files']}  |  Ratio: {tc['test_to_source_ratio']:.3f}")
-    if tc["coverage_percent"] is not None:
-        print(f"Line coverage: {tc['coverage_percent']:.1f}%  (via {tc['coverage_tool']})")
-    else:
-        print("Line coverage: (not available — run `pytest --cov`)")
-    
-    print()
-    dc = m["doc_coverage"]
-    print(f"Callables with docstrings: {dc['callables_with_doc']}/{dc['callables_total']}  ({dc['doc_coverage_percent']:.1f}%)")
-    
-    print()
-    im = m["issues"]
-    if im.get("close_rate") is not None:
-        print(f"Issues (7d): {im['closed_last_7d']} closed / {im['opened_last_7d']} opened  → close rate: {im['close_rate']*100:.1f}%")
-        print(f"Total open: {im['total_open']}")
-    else:
-        print(f"Issues: {im.get('note','unavailable')}")
-    
-    print()
-    dd = m["dependencies"]
-    print(f"Dependencies: {dd.get('total_deps',0)} total, {dd.get('outdated_deps',0)} outdated")
-    if dd.get('outdated_list'):
-        shown = dd['outdated_list'][:5]
-        print(f"Outdated: {', '.join(shown)}" + ("..." if len(dd['outdated_list']) > 5 else ""))
-    
-    print(f"\nTrends written to: {args.output}")
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())

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.")