Run pipeline, write Post 08, commit untracked files

Pipeline:
- Extract ideas for 38 new drafts → 462 ideas total
- Convergence analysis: 132 cross-org convergent ideas (33% rate)
- Fetch authors for 102 drafts → 709 authors (up from 403)
- Refresh gap analysis: 12 gaps across full 474-draft corpus
- Update verified counts with new totals

Post 08:
- Complete rewrite of "Agents Building the Agent Analysis" (2,953 words)
- Covers 3 phases: writing team → review cycle → fix cycle
- Meta-irony table mapping team coordination to IETF gap names
- Specific examples from dev journal (SQL injection, consent conflation, ideas mismatch)

Untracked files committed:
- scripts/: backfill-wg-names, classify-unrated, compare-classifiers, download-relevant-text, run-webui
- src/ietf_analyzer/classifier.py: two-stage Ollama classifier
- src/webui/: analytics (GDPR-compliant), auth, obsidian_export
- tests/test_obsidian_export.py (10 tests)
- data/reports/: wg-analysis, generated draft for gap #37

Housekeeping:
- .gitignore: exclude LaTeX artifacts, stale DBs, analytics.db

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

scripts/compare-classifiers.py

@@ -0,0 +1,86 @@
+#!/usr/bin/env python3
+"""Compare Ollama classifier vs Claude ratings to find disagreements."""
+
+import sqlite3
+import sys
+sys.path.insert(0, "src")
+
+from ietf_analyzer.classifier import Classifier
+from ietf_analyzer.config import Config
+
+cfg = Config.load()
+conn = sqlite3.connect(cfg.db_path)
+conn.row_factory = sqlite3.Row
+
+# Get all rated drafts with their Claude ratings
+rows = conn.execute("""
+    SELECT d.name, d.title, d.abstract, r.relevance, r.false_positive,
+           r.novelty, r.maturity, r.overlap, r.momentum,
+           (r.novelty + r.maturity + (5 - r.overlap) + r.momentum + r.relevance) / 5.0 as composite
+    FROM drafts d JOIN ratings r ON d.name = r.draft_name
+    WHERE d.abstract IS NOT NULL AND d.abstract != ''
+    ORDER BY d.name
+""").fetchall()
+
+print(f"Comparing Ollama classifier vs Claude ratings on {len(rows)} drafts...\n")
+
+with Classifier(cfg) as clf:
+    agree = 0
+    disagree_ollama_yes_claude_no = []  # Ollama says relevant, Claude says FP
+    disagree_ollama_no_claude_yes = []  # Ollama says irrelevant, Claude says relevant
+
+    for i, r in enumerate(rows):
+        is_rel, sim, method = clf.classify(r["title"], r["abstract"])
+
+        # Claude's view: false_positive=1 OR relevance<=2 means "not really relevant"
+        claude_relevant = not r["false_positive"] and r["relevance"] >= 3
+
+        if is_rel == claude_relevant:
+            agree += 1
+        elif is_rel and not claude_relevant:
+            disagree_ollama_yes_claude_no.append({
+                "name": r["name"], "title": r["title"][:60],
+                "sim": sim, "method": method,
+                "relevance": r["relevance"], "fp": r["false_positive"],
+                "composite": r["composite"],
+            })
+        else:
+            disagree_ollama_no_claude_yes.append({
+                "name": r["name"], "title": r["title"][:60],
+                "sim": sim, "method": method,
+                "relevance": r["relevance"], "fp": r["false_positive"],
+                "composite": r["composite"],
+            })
+
+        if (i + 1) % 50 == 0:
+            print(f"  Processed {i+1}/{len(rows)}...")
+
+print(f"\n{'='*70}")
+print(f"AGREEMENT: {agree}/{len(rows)} ({100*agree/len(rows):.1f}%)")
+print(f"{'='*70}")
+
+print(f"\nOllama=RELEVANT but Claude=NOT relevant ({len(disagree_ollama_yes_claude_no)}):")
+print(f"  (These are cases where Ollama wastes Claude tokens on irrelevant drafts)")
+for d in sorted(disagree_ollama_yes_claude_no, key=lambda x: x["sim"], reverse=True)[:15]:
+    fp_label = " [FP]" if d["fp"] else ""
+    print(f"  sim={d['sim']:.3f} ({d['method']:18s}) rel={d['relevance']}{fp_label} | {d['name']}")
+    print(f"    {d['title']}")
+
+print(f"\nOllama=IRRELEVANT but Claude=RELEVANT ({len(disagree_ollama_no_claude_yes)}):")
+print(f"  (These are cases where Ollama would have incorrectly filtered out good drafts)")
+for d in sorted(disagree_ollama_no_claude_yes, key=lambda x: x["relevance"], reverse=True)[:15]:
+    print(f"  sim={d['sim']:.3f} ({d['method']:18s}) rel={d['relevance']} comp={d['composite']:.1f} | {d['name']}")
+    print(f"    {d['title']}")
+
+# Summary stats
+total_fp_by_claude = sum(1 for r in rows if r["false_positive"] or r["relevance"] <= 2)
+total_relevant_by_claude = len(rows) - total_fp_by_claude
+print(f"\n{'='*70}")
+print(f"Claude thinks: {total_relevant_by_claude} relevant, {total_fp_by_claude} not relevant")
+print(f"Ollama would let through: {agree + len(disagree_ollama_yes_claude_no) - len(disagree_ollama_no_claude_yes)} (saves {len(disagree_ollama_no_claude_yes) - len(disagree_ollama_yes_claude_no)} Claude calls)")
+print(f"\nToken savings if Ollama pre-filters:")
+print(f"  Correctly rejected: {agree - total_relevant_by_claude + len(rows) - agree - len(disagree_ollama_yes_claude_no)} drafts")
+print(f"  Incorrectly rejected (missed): {len(disagree_ollama_no_claude_yes)} drafts")
+print(f"  Incorrectly passed (wasted): {len(disagree_ollama_yes_claude_no)} drafts")
+
+conn.close()