Fix Bug 3: Filter ghost agents from reasoning tree, add MACRO_DSLR to modality label map, add debug logging for agent scores

app.py

@@ -50,6 +50,9 @@ def run_all_agents(img: Image.Image) -> Tuple[List[AgentEvidence], ForensicVerdi
     from agents.modality_detector import detect_modality
     modality = detect_modality(img)
     adj = modality.score_adjustments
+    
+    print(f"[FORENSIQ] Modality: {modality.modality} (conf={modality.confidence})", file=sys.stderr)
+    print(f"[FORENSIQ] Adjustments: {len(adj)} tests recalibrated", file=sys.stderr)
 
     # Signal processing agents (fast, run in parallel with modality adjustments)
     signal_agents = [
@@ -78,6 +81,7 @@ def run_all_agents(img: Image.Image) -> Tuple[List[AgentEvidence], ForensicVerdi
             try:
                 results[name] = future.result()
             except Exception as e:
+                print(f"[FORENSIQ] Agent '{name}' FAILED: {e}", file=sys.stderr)
                 results[name] = AgentEvidence(
                     agent_name=f"{name.title()} Agent (Error)",
                     violation_score=0.0,
@@ -97,9 +101,25 @@ def run_all_agents(img: Image.Image) -> Tuple[List[AgentEvidence], ForensicVerdi
         results.get("text"),
     ]
     ordered = [r for r in ordered if r is not None]
-
-    # Bayesian synthesis
-    verdict = bayesian_synthesis(ordered)
+    
+    # Log agent scores for debugging
+    for a in ordered:
+        status = "ACTIVE" if a.failure_prob < 0.8 else "FAILED"
+        print(f"[FORENSIQ] {a.agent_name}: score={a.violation_score:+.3f} conf={a.confidence:.3f} fail={a.failure_prob:.2f} [{status}]", file=sys.stderr)
+
+    # FIX Bug 3: Filter out failed agents BEFORE Bayesian synthesis
+    # Failed agents (failure_prob >= 0.8) contribute no evidence — they're
+    # ghost entries that shouldn't appear in the reasoning tree or affect 
+    # the active agent count display.
+    active_agents = [r for r in ordered if r.failure_prob < 0.8]
+    failed_agents = [r for r in ordered if r.failure_prob >= 0.8]
+    
+    n_active = len(active_agents)
+    n_total = len(ordered)
+    print(f"[FORENSIQ] Active agents: {n_active}/{n_total}", file=sys.stderr)
+
+    # Bayesian synthesis — only pass active agents
+    verdict = bayesian_synthesis(active_agents)
 
     # Attach modality info to verdict for reporting — include ALL indicators for diagnostics
     verdict.reasoning_tree["modality"] = {
@@ -109,14 +129,22 @@ def run_all_agents(img: Image.Image) -> Tuple[List[AgentEvidence], ForensicVerdi
                        if not isinstance(v, np.ndarray) and k != "modality_scores"},
         "modality_scores": modality.indicators.get("modality_scores", {}),
         "adjustments_applied": len(modality.score_adjustments),
-        "adjustments_list": list(modality.score_adjustments.keys())[:10],  # First 10 for display
+        "adjustments_list": list(modality.score_adjustments.keys())[:10],
     }
+    
+    # Track failed agents in the tree for transparency
+    if failed_agents:
+        verdict.reasoning_tree["failed_agents"] = [
+            {"name": a.agent_name, "reason": a.rationale[:200]} 
+            for a in failed_agents
+        ]
 
     # Generate explanations
     verdict.forensic_report = generate_forensic_report(verdict)
     reasoning = generate_reasoning_tree(verdict)
     verdict.court_brief = generate_court_brief(verdict)
 
+    # Return ALL agents (including failed) for display, but verdict uses only active
     return ordered, verdict, reasoning
 
 
@@ -174,7 +202,6 @@ def create_radar_chart(agent_results: List[AgentEvidence]) -> go.Figure:
     for agent in agent_results:
         short_name = agent.agent_name.replace(" Agent", "").replace(" Characteristics", "")
         names.append(short_name)
-        # Map -1..+1 to 0..100 for display
         display_score = (agent.violation_score + 1) * 50
         scores.append(display_score)
         if agent.violation_score > 0.2:
@@ -184,7 +211,6 @@ def create_radar_chart(agent_results: List[AgentEvidence]) -> go.Figure:
         else:
             colors.append("gold")
 
-    # Close the polygon
     names_closed = names + [names[0]]
     scores_closed = scores + [scores[0]]
 
@@ -199,7 +225,6 @@ def create_radar_chart(agent_results: List[AgentEvidence]) -> go.Figure:
         name="Violation Score",
     ))
 
-    # Add neutral line at 50 (score = 0)
     fig.add_trace(go.Scatterpolar(
         r=[50] * (len(names) + 1),
         theta=names_closed,
@@ -297,7 +322,6 @@ def create_fft_display(agent_results: List[AgentEvidence]) -> go.Figure:
                         yaxis=dict(showticklabels=False, scaleanchor="x"),
                     )
                     return fig
-    # Return empty figure
     fig = go.Figure()
     fig.update_layout(height=400, title="FFT Spectrum (not available)")
     return fig
@@ -431,9 +455,18 @@ def analyze_image(img):
         mod_name = mod_info.get("detected", "UNKNOWN")
         mod_conf = mod_info.get("confidence", 0)
         mod_adj = mod_info.get("adjustments_applied", 0)
-        mod_label = {"PORTRAIT_MODE": "📱 Portrait Mode", "MESSAGING": "💬 Messaging App",
-                     "SMARTPHONE": "📱 Smartphone", "SCREENSHOT": "🖥�� Screenshot",
-                     "DSLR": "📷 DSLR", "UNKNOWN": "❓ Unknown"}.get(mod_name, mod_name)
+        mod_label = {
+            "PORTRAIT_MODE": "📱 Portrait Mode", 
+            "MESSAGING": "💬 Messaging App",
+            "SMARTPHONE": "📱 Smartphone", 
+            "SCREENSHOT": "🖥️ Screenshot",
+            "DSLR": "📷 DSLR", 
+            "MACRO_DSLR": "🔬 Macro DSLR",
+            "UNKNOWN": "❓ Unknown",
+        }.get(mod_name, f"📷 {mod_name}")
+
+        # Count active agents from verdict (which only has active agents now)
+        n_active = len(verdict.agent_results)
 
         verdict_html = f"""
         <div style="background:{bg}; color:white; padding:24px; border-radius:16px;
@@ -442,7 +475,7 @@ def analyze_image(img):
             <div style="font-size:28px; font-weight:700; margin-bottom:4px;">{verdict.verdict}</div>
             <div style="font-size:42px; font-weight:800;">{prob:.1%}</div>
             <div style="font-size:14px; opacity:0.9; margin-top:4px;">
-                Confidence: {verdict.confidence} | Agents: {len([a for a in agent_results if a.failure_prob < 0.8])}/7 active
+                Confidence: {verdict.confidence} | Agents: {n_active}/7 active
             </div>
             <div style="font-size:12px; opacity:0.7; margin-top:6px; border-top:1px solid rgba(255,255,255,0.2); padding-top:6px;">
                 Modality: {mod_label} ({mod_conf:.0%}) | {mod_adj} test(s) recalibrated
@@ -476,6 +509,8 @@ def analyze_image(img):
             _build_agent_details_md(agent_results),
         )
     except Exception as e:
+        import traceback
+        traceback.print_exc(file=sys.stderr)
         error_html = f"""
         <div style="background:linear-gradient(135deg, #6c757d, #495057); color:white;
                      padding:24px; border-radius:16px; text-align:center;">
@@ -518,7 +553,12 @@ def _build_agent_details_md(agent_results: List[AgentEvidence]) -> str:
             note = sf.get("note", "")
             s = sf.get("score", 0)
             ic = "🔴" if s > 0.2 else "🟢" if s < -0.1 else "🟡"
-            md += f"- {ic} **{test}** ({s:+.2f}): {note}\n"
+            # Show modality adjustment info
+            if sf.get("modality_adjusted"):
+                adj_info = f" [×{sf.get('adjustment_multiplier', '?')} modality]"
+            else:
+                adj_info = ""
+            md += f"- {ic} **{test}** ({s:+.2f}{adj_info}): {note}\n"
         md += "\n---\n\n"
     return md