Add ML model comparison panel showing BoW, TF-IDF, Naive Bayes, and LDA results

All four classical classifiers now run on every verification request (concurrently
with Layer 2 evidence fetch) and their verdicts, confidence scores, and top
triggered features appear in a side-by-side panel in the result UI between
the Score Breakdown and Layer 1/Layer 2 cards. LDA topic labels (lda_topic_N)
are highlighted in teal to distinguish them from TF-IDF feature chips.

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

api/schemas.py

@@ -94,6 +94,15 @@ class Layer2Result(BaseModel):
     claim_method: Optional[str] = Field(None, description="How the claim was extracted: sentence_scoring | sentence_heuristic | passthrough")
 
 
+# ── Classifier Comparison ─────────────────────────────────────────────────────
+
+class ClassifierComparisonEntry(BaseModel):
+    name: str                           # "BoW", "TF-IDF", "Naive Bayes", "LDA"
+    verdict: Verdict
+    confidence: float = Field(..., ge=0.0, le=100.0)
+    top_features: list[str] = []        # up to 3 top features / lda_topic_N label
+
+
 # ── Main Response ─────────────────────────────────────────────────────────────
 
 class VerificationResponse(BaseModel):
@@ -111,6 +120,10 @@ class VerificationResponse(BaseModel):
     processing_time_ms: Optional[float] = None
     extracted_text: Optional[str] = Field(None, description="Raw text extracted from the URL / image / video for transparency")
     ocr_text: Optional[str] = Field(None, description="Text extracted from an image via OCR (when image_url was provided alongside text)")
+    classifier_comparison: list[ClassifierComparisonEntry] = Field(
+        default_factory=list,
+        description="Per-classifier results from all classical ML models (BoW, TF-IDF, NB, LDA)",
+    )
 
 
 # ── History / Trends ──────────────────────────────────────────────────────────

frontend/src/pages/VerifyPage.jsx

@@ -885,6 +885,69 @@ export default function VerifyPage() {
                         </p>
                     </div>
 
+                    {/* Model Comparison Panel */}
+                    {result.classifier_comparison?.length > 0 && (
+                        <div className="card p-5 fade-up-3">
+                            <SectionHeading>Model Comparison — Classical ML</SectionHeading>
+                            <p className="text-xs mb-4" style={{ color: 'var(--text-muted)', fontFamily: 'var(--font-body)', lineHeight: 1.6 }}>
+                                All four classical classifiers trained on the same dataset — same input, different algorithms.
+                            </p>
+                            <div style={{ display: 'grid', gridTemplateColumns: 'repeat(auto-fit, minmax(160px, 1fr))', gap: 10 }}>
+                                {result.classifier_comparison.map((clf) => (
+                                    <div key={clf.name} style={{
+                                        background: 'rgba(255,255,255,0.03)',
+                                        border: '1px solid rgba(255,255,255,0.08)',
+                                        borderRadius: 6, padding: '10px 12px',
+                                    }}>
+                                        <div style={{
+                                            fontSize: '0.65rem', fontFamily: 'var(--font-mono)',
+                                            letterSpacing: '0.08em', color: 'var(--text-muted)',
+                                            textTransform: 'uppercase', marginBottom: 6,
+                                        }}>
+                                            {clf.name}
+                                        </div>
+                                        <div style={{
+                                            display: 'inline-block', fontSize: '0.7rem', fontWeight: 600,
+                                            padding: '2px 8px', borderRadius: 3, marginBottom: 6,
+                                            background: clf.verdict === 'Credible' ? 'rgba(34,197,94,0.15)' :
+                                                        clf.verdict === 'Likely Fake' ? 'rgba(239,68,68,0.15)' :
+                                                        'rgba(234,179,8,0.15)',
+                                            color: clf.verdict === 'Credible' ? 'var(--accent-green)' :
+                                                   clf.verdict === 'Likely Fake' ? '#f87171' : 'var(--accent-gold)',
+                                        }}>
+                                            {clf.verdict}
+                                        </div>
+                                        <div style={{ fontSize: '0.68rem', color: 'var(--text-muted)', marginBottom: 4 }}>
+                                            {clf.confidence.toFixed(1)}% confidence
+                                        </div>
+                                        <div style={{ height: 3, background: 'rgba(255,255,255,0.08)', borderRadius: 2, marginBottom: 8 }}>
+                                            <div style={{
+                                                height: '100%', borderRadius: 2, width: `${clf.confidence}%`,
+                                                background: clf.verdict === 'Credible' ? 'var(--accent-green)' :
+                                                            clf.verdict === 'Likely Fake' ? '#f87171' : 'var(--accent-gold)',
+                                            }} />
+                                        </div>
+                                        {clf.top_features?.length > 0 && (
+                                            <div style={{ display: 'flex', flexWrap: 'wrap', gap: 3 }}>
+                                                {clf.top_features.map((f, i) => (
+                                                    <span key={i} style={{
+                                                        fontSize: '0.6rem', padding: '1px 5px', borderRadius: 2,
+                                                        background: f.startsWith('lda_topic') ? 'rgba(6,182,212,0.12)' : 'rgba(220,38,38,0.1)',
+                                                        color: f.startsWith('lda_topic') ? 'var(--accent-cyan)' : '#f87171',
+                                                        border: `1px solid ${f.startsWith('lda_topic') ? 'rgba(6,182,212,0.3)' : 'rgba(220,38,38,0.25)'}`,
+                                                        fontFamily: 'var(--font-mono)',
+                                                    }}>
+                                                        {f}
+                                                    </span>
+                                                ))}
+                                            </div>
+                                        )}
+                                    </div>
+                                ))}
+                            </div>
+                        </div>
+                    )}
+
                     {/* Row 3: Layer cards (2 col, collapses to 1 on mobile) */}
                     <div className="grid grid-cols-1 sm:grid-cols-2 gap-4 fade-up-4">
                         {/* Layer 1 */}

scoring/engine.py

@@ -14,6 +14,7 @@ from config import get_settings
 from api.schemas import (
     VerificationResponse, Verdict, Language, DomainTier,
     Layer1Result, Layer2Result, EntitiesResult, EvidenceSource, Stance,
+    ClassifierComparisonEntry,
 )
 
 logger = logging.getLogger(__name__)
@@ -31,6 +32,39 @@ def _get_nlp(key: str, factory):
         _nlp_cache[key] = factory()
     return _nlp_cache[key]
 
+# ── Classical classifier comparison ──────────────────────────────────────────
+# Runs all four classical ML classifiers on every request for the demo panel.
+# Each classifier trains once on first call and is cached via _get_nlp().
+
+async def _run_comparison(text: str) -> list[ClassifierComparisonEntry]:
+    """Run BoW, TF-IDF, Naive Bayes, and LDA classifiers and return comparison entries."""
+    _COMPARISON_CLASSIFIERS = [
+        ("BoW",         "cmp_bow",   lambda: __import__("ml.bow_classifier", fromlist=["BoWClassifier"]).BoWClassifier()),
+        ("TF-IDF",      "cmp_tfidf", lambda: __import__("ml.tfidf_classifier", fromlist=["TFIDFClassifier"]).TFIDFClassifier()),
+        ("Naive Bayes", "cmp_nb",    lambda: __import__("ml.naive_bayes_classifier", fromlist=["NaiveBayesClassifier"]).NaiveBayesClassifier()),
+        ("LDA",         "cmp_lda",   lambda: __import__("ml.lda_analysis", fromlist=["LDAFeatureClassifier"]).LDAFeatureClassifier()),
+    ]
+
+    def _predict_all():
+        results = []
+        for name, key, factory in _COMPARISON_CLASSIFIERS:
+            try:
+                clf = _get_nlp(key, factory)
+                r = clf.predict(text)
+                results.append(ClassifierComparisonEntry(
+                    name=name,
+                    verdict=Verdict(r.verdict),
+                    confidence=r.confidence,
+                    top_features=r.triggered_features[:3],
+                ))
+            except Exception as exc:
+                logger.warning("Comparison classifier %s failed: %s", name, exc)
+        return results
+
+    loop = asyncio.get_event_loop()
+    return await loop.run_in_executor(None, _predict_all)
+
+
 # ── Domain credibility lookup ─────────────────────────────────────────────────
 _DOMAIN_DB_PATH = Path(__file__).parent.parent / "domain_credibility.json"
 _DOMAIN_DB: dict = {}
@@ -173,6 +207,9 @@ async def run_verification(
     evidence_sources: list[EvidenceSource] = []
     l2_verdict = Verdict.UNVERIFIED
 
+    # Run classifier comparison concurrently with evidence fetch
+    comparison_task = asyncio.create_task(_run_comparison(proc.cleaned))
+
     if settings.news_api_key:
         try:
             query_entities = ner_result.persons + ner_result.organizations + ner_result.locations
@@ -278,6 +315,8 @@ async def run_verification(
     verdict = _map_verdict(final_score)
 
     # ── Step 10: Assemble response ────────────────────────────────────────────
+    comparison = await comparison_task
+
     result = VerificationResponse(
         verdict=verdict,
         confidence=round(max(l1.confidence, evidence_score / 100 * 100), 1),
@@ -295,6 +334,7 @@ async def run_verification(
         language=language,
         domain_credibility=get_domain_tier(source_domain) if source_domain else None,
         input_type=input_type,
+        classifier_comparison=comparison,
     )
 
     # ── Record to Firestore (falls back to in-memory if Firebase not configured) ─