Coherence Ops Framework

Coherence Ops Framework

Path: src/coherence_ops/

The Coherence Ops framework implements the four canonical governance artifacts (DLR / RS / DS / MG) and the coherence audit loop that connects RAL / Σ OVERWATCH runtime exhaust to structured governance, learning, and memory.

The Four Canonical Artifacts

Artifact	Full Name	Class	Question It Answers
DLR	Decision Log Record	DLRBuilder	What policy governed this decision, and was it followed?
RS	Reflection Session	ReflectionSession	What happened, what degraded, what should we learn?
DS	Drift Signal	DriftSignalCollector	What is breaking, how often, and how badly?
MG	Memory Graph	MemoryGraph	What happened before, why, and what changed as a result?

Modules

Module	Class	Purpose
manifest.py	CoherenceManifest	System-level declaration of artifact coverage — the "bill of materials"
dlr.py	DLRBuilder	Build Decision Log Records from sealed episodes
rs.py	ReflectionSession	Aggregate episodes into learning summaries with divergence detection
ds.py	DriftSignalCollector	Collect and bucket drift signals by fingerprint, type, and severity
mg.py	MemoryGraph	Provenance graph for "why did we do this?" queries
audit.py	CoherenceAuditor	Cross-artifact consistency checks (manifest coverage, DLR completeness, drift resolution, MG orphans)
scoring.py	CoherenceScorer	Unified 0–100 coherence score with A/B/C/D/F grading
reconciler.py	Reconciler	Detect and propose repairs for cross-artifact inconsistencies

Scoring Dimensions

The coherence score is computed from four weighted dimensions:

Dimension	Weight	Source	What It Measures
Policy Adherence	25%	DLR	Percentage of episodes with valid policy stamps
Outcome Health	30%	RS	Success rate × verification pass rate
Drift Control	25%	DS	Recurring drift volume and red-severity signals
Memory Completeness	20%	MG	Graph coverage — episodes present vs expected

Data Flow

Sealed episodes and drift events from RAL flow through the framework:

1. DLR Builder extracts policy stamps, DTE references, and outcomes from episodes
2. Reflection Session aggregates batches into outcome distributions, degrade frequency, and takeaways
3. Drift Signal Collector buckets drift events by fingerprint and tracks recurrence
4. Memory Graph builds a provenance graph linking episodes, actions, drift, and patches
5. Coherence Auditor runs cross-artifact consistency checks
6. Coherence Scorer produces a single 0–100 score
7. Reconciler proposes repair actions for detected inconsistencies

Manifest

The CoherenceManifest declares which artifacts a system produces:

• system_id — unique system identifier
• version — manifest version
• artifacts[] — one entry per artifact kind, with schema_version, compliance level (full/partial/scaffold/missing), source, and refresh cadence

Schemas

• schemas/coherence_manifest.schema.json — manifest validation
• schemas/coherence_report.schema.json — audit report validation

See Also

• Coherence Ops Mapping — how RAL maps to DLR/RS/DS/MG
• LLM Data Model — data requirements for RAL
• Architecture — where Coherence Ops sits in the stack
• Mermaid Diagrams — visual pipeline diagram (06-coherence-ops-pipeline.md)