agent-alignment-protocol 0.1.4__tar.gz → 0.1.6__tar.gz

{agent_alignment_protocol-0.1.4 → agent_alignment_protocol-0.1.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-alignment-protocol
-Version: 0.1.4
+Version: 0.1.6
 Summary: Agent Alignment Protocol - The missing alignment layer for the agent protocol stack
 Project-URL: Homepage, https://github.com/mnemom/aap
 Project-URL: Documentation, https://github.com/mnemom/aap#readme

{agent_alignment_protocol-0.1.4 → agent_alignment_protocol-0.1.6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "agent-alignment-protocol"
-version = "0.1.4"
+version = "0.1.6"
 description = "Agent Alignment Protocol - The missing alignment layer for the agent protocol stack"
 readme = "README.md"
 license = "Apache-2.0"

{agent_alignment_protocol-0.1.4 → agent_alignment_protocol-0.1.6}/src/aap/verification/__init__.py

@@ -52,6 +52,7 @@ from aap.verification.divergence import (
 )
 from aap.verification.features import (
     FeatureExtractor,
+    compute_centroid,
     compute_similarity_with_tfidf,
     cosine_similarity,
 )
@@ -87,6 +88,7 @@ __all__ = [
     "detect_divergence",
     # Feature extraction
     "FeatureExtractor",
+    "compute_centroid",
     "cosine_similarity",
     "compute_similarity_with_tfidf",
     # Result models

{agent_alignment_protocol-0.1.4 → agent_alignment_protocol-0.1.6}/src/aap/verification/constants.py

@@ -58,4 +58,4 @@ TFIDF_META_WEIGHT: float = 0.10
 
 # Version
 # -------
-ALGORITHM_VERSION: str = "1.1.0"
+ALGORITHM_VERSION: str = "1.2.0"

{agent_alignment_protocol-0.1.4 → agent_alignment_protocol-0.1.6}/src/aap/verification/divergence.py

@@ -23,7 +23,7 @@ from aap.verification.constants import (
     DEFAULT_SIMILARITY_THRESHOLD,
     DEFAULT_SUSTAINED_TURNS_THRESHOLD,
 )
-from aap.verification.features import FeatureExtractor, cosine_similarity
+from aap.verification.features import FeatureExtractor, compute_centroid, cosine_similarity
 from aap.verification.models import (
     DriftAlert,
     DriftAnalysis,
@@ -68,23 +68,39 @@ class DivergenceDetector:
         """Detect divergence alerts for a sequence of traces.
 
         Analyzes traces chronologically, computing similarity between each
-        trace's behavior and the declared alignment. Generates alerts for
-        sustained divergence (consecutive traces below threshold).
+        trace and a baseline centroid computed from the first N traces.
+        Generates alerts for sustained divergence (consecutive traces below
+        threshold).
 
         Args:
             card: Alignment Card dict per SPEC Section 4
-            traces: List of AP-Trace dicts in chronological order
+            traces: List of AP-Trace dicts (sorted chronologically internally)
 
         Returns:
             List of DriftAlert objects for detected divergences
         """
-        if not traces or len(traces) < self.sustained_turns_threshold:
+        if not traces:
+            return []
+
+        # Sort traces chronologically
+        sorted_traces = sorted(traces, key=lambda t: t.get("timestamp", ""))
+
+        # Compute baseline window size
+        baseline_size = max(self.sustained_turns_threshold, min(10, len(sorted_traces) // 4))
+
+        # Need enough traces for baseline + sustained threshold
+        if len(sorted_traces) < baseline_size + self.sustained_turns_threshold:
             return []
 
-        # Extract card features once
-        card_features = self._feature_extractor.extract_card_features(card)
         card_id = card.get("card_id", "")
 
+        # Extract features for baseline traces and compute centroid
+        baseline_features = [
+            self._feature_extractor.extract_trace_features(t)
+            for t in sorted_traces[:baseline_size]
+        ]
+        baseline_centroid = compute_centroid(baseline_features)
+
         alerts: list[DriftAlert] = []
         low_similarity_streak: list[tuple[dict, float]] = []
 
@@ -92,10 +108,18 @@ class DivergenceDetector:
         escalation_rates: list[float] = []
         value_usage: dict[str, int] = defaultdict(int)
 
-        for trace in traces:
-            # Compute similarity to card
+        # Include baseline traces in escalation/value tracking
+        for trace in sorted_traces[:baseline_size]:
+            escalation = trace.get("escalation", {})
+            escalation_rates.append(1.0 if escalation.get("required") else 0.0)
+            for value in trace.get("decision", {}).get("values_applied", []):
+                value_usage[value] += 1
+
+        # Iterate from after baseline to end
+        for trace in sorted_traces[baseline_size:]:
+            # Compute similarity to baseline centroid
             trace_features = self._feature_extractor.extract_trace_features(trace)
-            similarity = cosine_similarity(trace_features, card_features)
+            similarity = cosine_similarity(trace_features, baseline_centroid)
 
             # Track escalation rate
             escalation = trace.get("escalation", {})

{agent_alignment_protocol-0.1.4 → agent_alignment_protocol-0.1.6}/src/aap/verification/features.py

@@ -351,6 +351,18 @@ class FeatureExtractor:
         return features
 
 
+def compute_centroid(vectors: list[dict[str, float]]) -> dict[str, float]:
+    """Compute the centroid (element-wise average) of sparse feature vectors."""
+    if not vectors:
+        return {}
+    centroid: dict[str, float] = {}
+    for vec in vectors:
+        for key, value in vec.items():
+            centroid[key] = centroid.get(key, 0.0) + value
+    n = len(vectors)
+    return {k: v / n for k, v in centroid.items()}
+
+
 def cosine_similarity(a: dict[str, float], b: dict[str, float]) -> float:
     """Compute cosine similarity between two sparse feature vectors.