S0-R1-C5.5-R1 Study-Causal Semantic Frame Closure

core/quantitative_engine.py

@@ -630,6 +630,56 @@ Keywords: {keywords}"""
             r"\b(?:has|have|had|exerts?|exerted)\s+(?:an?\s+|the\s+)?causal\s+effect\b",
         )
 
+        def direct_match_is_statistical_reporting(match):
+            """Return True only for a statistical-operation -> statistical-artifact frame.
+
+            C5.5-R1: causal-looking verbs such as ``produced``/``resulted in``
+            are ambiguous in quantitative prose.  A bootstrap procedure producing
+            a confidence interval is statistical reporting, not a claim that one
+            study construct causes another.  The exemption therefore requires
+            *both* a statistical-operation source immediately before the verb and
+            a statistical-artifact target immediately after it.  Merely seeing
+            ``analysis`` or ``model`` is not enough; e.g. ``the analysis produced
+            higher engagement`` remains fail-closed.
+            """
+            prefix = value[max(0, match.start() - 180):match.start()]
+            suffix = value[match.end():min(len(value), match.end() + 180)]
+
+            # Keep the source local to the current grammatical frame rather than
+            # letting an earlier sentence mention of an analysis whitelist a later
+            # construct-level causal assertion.
+            source = re.split(r"[.;!?]", prefix)[-1]
+            source_pattern = (
+                r"\b(?:bootstrap(?:\s+resampling|\s+procedure|\s+computation)?|"
+                r"resampling(?:\s+procedure|\s+algorithm)?|"
+                r"regression\s+(?:analysis|model)|"
+                r"association\s+(?:analysis|model)|"
+                r"mediation\s+(?:analysis|model)|"
+                r"estimation(?:\s+procedure|\s+method)?|estimator|"
+                r"statistical\s+(?:analysis|model|procedure|method)|"
+                r"analysis|analyses|model|procedure|method|algorithm|"
+                r"computation|calculation)\b"
+            )
+            if not re.search(source_pattern, source, flags=re.I):
+                return False
+
+            # The target must be a recognisable statistical artefact.  Do not use
+            # generic outcomes such as ``result`` or ``change`` because those can
+            # also be substantive dependent variables.
+            target_pattern = (
+                r"^\s*(?:(?:an?|the|this|that|these|those|following|reported|"
+                r"standardized|standardised|numerical|estimated|95\s*percent|95%)\s+){0,6}"
+                r"(?:confidence\s+intervals?|interval\s+estimates?|"
+                r"coefficient(?:\s+estimates?)?s?|standard\s+errors?|"
+                r"parameter\s+estimates?|point\s+estimates?|indirect\s+estimates?|"
+                r"direct\s+estimates?|effect-size\s+estimates?|"
+                r"correlations?|covariances?|variances?|loadings?|"
+                r"test\s+statistics?|p[- ]?values?|vif(?:s|\s+values)?|"
+                r"cronbach(?:'s)?\s+alpha|composite\s+reliability|"
+                r"average\s+variance\s+extracted|ave|numerical\s+estimates?|estimates?)\b"
+            )
+            return bool(re.search(target_pattern, suffix, flags=re.I))
+
         def direct_match_is_disclaimed(match):
             prefix = value[max(0, match.start() - 220):match.start()]
             if re.search(
@@ -670,6 +720,8 @@ Keywords: {keywords}"""
                 if direct_match_is_disclaimed(match):
                     saw_disclaimed_direct = True
                     continue
+                if direct_match_is_statistical_reporting(match):
+                    continue
                 return "AFFIRMATIVE_CAUSAL"
 
         affirmative_noun_patterns = (