src.dackar.RCA.signal_evidence.models

Classes

AnomalyRecord	A single flagged anomaly on one sensor over a time interval.
PropagationEdge	A directed propagation edge between two anomalies (by list index).
NodeTopology	In/out degree and derived pattern (linear/divergence/convergence/hub/isolated) for one anomaly node.
ScoredChain	A scored propagation path plus the factors that produced its path_score.

Module Contents

class src.dackar.RCA.signal_evidence.models.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.

sensor_id: str

component_id: str | None

timestamp_start: datetime.datetime

timestamp_end: datetime.datetime

pattern: str

severity: float

source: str

raw_value_start: float | None = None

raw_value_peak: float | None = None

units: str | None = None

to_interval()

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

Return type:

orchestrators.temporal_relations.Interval

class src.dackar.RCA.signal_evidence.models.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.

from_idx: int

to_idx: int

allen_rel: str

allen_score: float

prop_score: float

edge_type: str

onset_lag_h: float

class src.dackar.RCA.signal_evidence.models.NodeTopology

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

anomaly_idx: int

in_degree: int

out_degree: int

pattern_type: str

class src.dackar.RCA.signal_evidence.models.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.

chain_id: str

path: list[int]

path_score: float

topology_alignment_factor: float

lag_consistency_factor: float

mean_allen_score: float

mean_propagation_score: float

hub_boost: float

root_pattern_type: str

nodes: list[dict]