src.dackar.RCA.signal_evidence.models
=====================================

.. py:module:: src.dackar.RCA.signal_evidence.models


Classes
-------

.. autoapisummary::

   src.dackar.RCA.signal_evidence.models.AnomalyRecord
   src.dackar.RCA.signal_evidence.models.PropagationEdge
   src.dackar.RCA.signal_evidence.models.NodeTopology
   src.dackar.RCA.signal_evidence.models.ScoredChain


Module Contents
---------------

.. py:class:: AnomalyRecord

   A single flagged anomaly on one sensor over a time interval.

   Fields carry the sensor and (resolved) component ids, the anomaly window,
   a ``pattern`` label, a ``severity`` in ``[0, 1]``, the ``source`` that
   produced it (``"telemetry_summary"`` or ``"historian"``), and optional raw
   values / units.


   .. py:attribute:: sensor_id
      :type:  str


   .. py:attribute:: component_id
      :type:  Optional[str]


   .. py:attribute:: timestamp_start
      :type:  datetime.datetime


   .. py:attribute:: timestamp_end
      :type:  datetime.datetime


   .. py:attribute:: pattern
      :type:  str


   .. py:attribute:: severity
      :type:  float


   .. py:attribute:: source
      :type:  str


   .. py:attribute:: raw_value_start
      :type:  Optional[float]
      :value: None



   .. py:attribute:: raw_value_peak
      :type:  Optional[float]
      :value: None



   .. py:attribute:: units
      :type:  Optional[str]
      :value: None



   .. py:method:: to_interval()

      Return this record's ``[start, end]`` as a temporal ``Interval``.



.. py:class:: PropagationEdge

   A directed propagation edge between two anomalies (by list index).

   ``allen_rel`` captures the temporal relation between the two anomaly
   intervals and ``allen_score`` its raw event-calibrated relevance prior
   (kept for provenance). ``prop_score`` is the *propagation*-calibrated
   weight actually used to rank chains: it rewards a demonstrated causal lead
   at least as much as a co-temporal overlap and discounts co-temporal edges,
   so a clean ``precedes`` lead is never outranked by an ``overlaps`` edge the
   way the raw event score would (MR#49 review). ``edge_type`` is the KG
   relation (``containment`` / ``connectivity`` / ``mixed``), and
   ``onset_lag_h`` the onset lead in hours from source to target.


   .. py:attribute:: from_idx
      :type:  int


   .. py:attribute:: to_idx
      :type:  int


   .. py:attribute:: allen_rel
      :type:  str


   .. py:attribute:: allen_score
      :type:  float


   .. py:attribute:: prop_score
      :type:  float


   .. py:attribute:: edge_type
      :type:  str


   .. py:attribute:: onset_lag_h
      :type:  float


.. py:class:: NodeTopology

   In/out degree and derived pattern (linear/divergence/convergence/hub/isolated) for one anomaly node.


   .. py:attribute:: anomaly_idx
      :type:  int


   .. py:attribute:: in_degree
      :type:  int


   .. py:attribute:: out_degree
      :type:  int


   .. py:attribute:: pattern_type
      :type:  str


.. py:class:: ScoredChain

   A scored propagation path plus the factors that produced its ``path_score``.

   ``path_score`` is driven by ``mean_propagation_score`` (the lead-time-aware
   edge weight); ``mean_allen_score`` is retained alongside as the mean of the
   raw event-calibrated relation priors, for provenance.


   .. py:attribute:: chain_id
      :type:  str


   .. py:attribute:: path
      :type:  list[int]


   .. py:attribute:: path_score
      :type:  float


   .. py:attribute:: topology_alignment_factor
      :type:  float


   .. py:attribute:: lag_consistency_factor
      :type:  float


   .. py:attribute:: mean_allen_score
      :type:  float


   .. py:attribute:: mean_propagation_score
      :type:  float


   .. py:attribute:: hub_boost
      :type:  float


   .. py:attribute:: root_pattern_type
      :type:  str


   .. py:attribute:: nodes
      :type:  list[dict]