Rename 'counterfactual' terminology to 'input ablation' in docstrings

Aligns with extension terminology update — these are ablation studies, not
true counterfactuals.

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

backend/model_service.py

@@ -3606,8 +3606,8 @@ class InterventionResponse(BaseModel):
 @app.post("/analyze/intervention")
 async def run_intervention(request: InterventionRequest, authenticated: bool = Depends(verify_api_key)):
     """
-    Run a counterfactual intervention on a cached generation run.
-    Re-evaluates a token position under modified conditions (masking, ablation, temperature sweep).
+    Run an input ablation on a cached generation run.
+    Re-evaluates a token position under modified conditions (input masking, component ablation, temperature sweep).
     """
     if not manager.model:
         raise HTTPException(status_code=503, detail="Model not loaded")

docs/rq1-mapping.md

@@ -89,7 +89,7 @@ This document maps each of the 4 visualizations (Attention, Token Size & Confide
 **Your extension:**
 - **Interactive head selection:** Developer chooses which head/layer to inspect in real-time
 - **Code-specific annotations:** Highlight syntactic elements (keywords, identifiers, operators) with domain-specific color coding
-- **Counterfactual queries:** "What if I remove this docstring? How does attention redistribute?"
+- **Input ablation queries:** "What changes if I remove this docstring? How does attention redistribute?"
 - **Task-embedded evaluation:** Developers use the tool during actual code review tasks (bug detection, prompt optimization), not just correlation studies
 
 **Paltenghi et al. (2022): "Follow-up Attention: An Empirical Study of Developer and Neural Model Code Exploration"**