Add unified explainability support and librarian storage for (#693)

Add unified explainability support and librarian storage for all retrieval engines

Implements consistent explainability/provenance tracking
across GraphRAG, DocumentRAG, and Agent retrieval
engines. All large content (answers, thoughts, observations)
is now stored in librarian rather than as inline literals in
the knowledge graph.

Explainability API:
- New explainability.py module with entity classes (Question,
  Exploration, Focus, Synthesis, Analysis, Conclusion) and
  ExplainabilityClient
- Quiescence-based eventual consistency handling for trace
  fetching
- Content fetching from librarian with retry logic

CLI updates:
- tg-invoke-graph-rag -x/--explainable flag returns
  explain_id
- tg-invoke-document-rag -x/--explainable flag returns
  explain_id
- tg-invoke-agent -x/--explainable flag returns explain_id
- tg-list-explain-traces uses new explainability API
- tg-show-explain-trace handles all three trace types

Agent provenance:
- Records session, iterations (think/act/observe), and conclusion
- Stores thoughts and observations in librarian with document
  references
- New predicates: tg:thoughtDocument, tg:observationDocument

DocumentRAG provenance:
- Records question, exploration (chunk retrieval), and synthesis
- Stores answers in librarian with document references

Schema changes:
- AgentResponse: added explain_id, explain_graph fields
- RetrievalResponse: added explain_id, explain_graph fields
- agent_iteration_triples: supports thought_document_id,
  observation_document_id

Update tests.
This commit is contained in:
cybermaggedon 2026-03-12 21:40:09 +00:00 committed by GitHub
parent aecf00f040
commit 35128ff019
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
24 changed files with 2736 additions and 846 deletions

View file

@ -82,6 +82,10 @@ from . namespaces import (
TG_GRAPH_RAG_QUESTION, TG_DOC_RAG_QUESTION, TG_AGENT_QUESTION,
# Agent provenance predicates
TG_THOUGHT, TG_ACTION, TG_ARGUMENTS, TG_OBSERVATION, TG_ANSWER,
# Agent document references
TG_THOUGHT_DOCUMENT, TG_OBSERVATION_DOCUMENT,
# Document reference predicate
TG_DOCUMENT,
# Named graphs
GRAPH_DEFAULT, GRAPH_SOURCE, GRAPH_RETRIEVAL,
)
@ -165,6 +169,10 @@ __all__ = [
"TG_GRAPH_RAG_QUESTION", "TG_DOC_RAG_QUESTION", "TG_AGENT_QUESTION",
# Agent provenance predicates
"TG_THOUGHT", "TG_ACTION", "TG_ARGUMENTS", "TG_OBSERVATION", "TG_ANSWER",
# Agent document references
"TG_THOUGHT_DOCUMENT", "TG_OBSERVATION_DOCUMENT",
# Document reference predicate
"TG_DOCUMENT",
# Named graphs
"GRAPH_DEFAULT", "GRAPH_SOURCE", "GRAPH_RETRIEVAL",
# Triple builders

View file

@ -17,7 +17,8 @@ from . namespaces import (
RDF_TYPE, RDFS_LABEL,
PROV_ACTIVITY, PROV_ENTITY, PROV_WAS_DERIVED_FROM, PROV_STARTED_AT_TIME,
TG_QUERY, TG_THOUGHT, TG_ACTION, TG_ARGUMENTS, TG_OBSERVATION, TG_ANSWER,
TG_QUESTION, TG_ANALYSIS, TG_CONCLUSION,
TG_QUESTION, TG_ANALYSIS, TG_CONCLUSION, TG_DOCUMENT,
TG_THOUGHT_DOCUMENT, TG_OBSERVATION_DOCUMENT,
TG_AGENT_QUESTION,
)
@ -73,10 +74,12 @@ def agent_session_triples(
def agent_iteration_triples(
iteration_uri: str,
parent_uri: str,
thought: str,
action: str,
arguments: Dict[str, Any],
observation: str,
thought: str = "",
action: str = "",
arguments: Dict[str, Any] = None,
observation: str = "",
thought_document_id: Optional[str] = None,
observation_document_id: Optional[str] = None,
) -> List[Triple]:
"""
Build triples for one agent iteration (Analysis - think/act/observe cycle).
@ -85,36 +88,53 @@ def agent_iteration_triples(
- Entity declaration with tg:Analysis type
- wasDerivedFrom link to parent (previous iteration or session)
- Thought, action, arguments, and observation data
- Document references for thought/observation when stored in librarian
Args:
iteration_uri: URI of this iteration (from agent_iteration_uri)
parent_uri: URI of the parent (previous iteration or session)
thought: The agent's reasoning/thought
thought: The agent's reasoning/thought (used if thought_document_id not provided)
action: The tool/action name
arguments: Arguments passed to the tool (will be JSON-encoded)
observation: The result/observation from the tool
observation: The result/observation from the tool (used if observation_document_id not provided)
thought_document_id: Optional document URI for thought in librarian (preferred)
observation_document_id: Optional document URI for observation in librarian (preferred)
Returns:
List of Triple objects
"""
if arguments is None:
arguments = {}
triples = [
_triple(iteration_uri, RDF_TYPE, _iri(PROV_ENTITY)),
_triple(iteration_uri, RDF_TYPE, _iri(TG_ANALYSIS)),
_triple(iteration_uri, RDFS_LABEL, _literal(f"Analysis: {action}")),
_triple(iteration_uri, PROV_WAS_DERIVED_FROM, _iri(parent_uri)),
_triple(iteration_uri, TG_THOUGHT, _literal(thought)),
_triple(iteration_uri, TG_ACTION, _literal(action)),
_triple(iteration_uri, TG_ARGUMENTS, _literal(json.dumps(arguments))),
_triple(iteration_uri, TG_OBSERVATION, _literal(observation)),
]
# Thought: use document reference or inline
if thought_document_id:
triples.append(_triple(iteration_uri, TG_THOUGHT_DOCUMENT, _iri(thought_document_id)))
elif thought:
triples.append(_triple(iteration_uri, TG_THOUGHT, _literal(thought)))
# Observation: use document reference or inline
if observation_document_id:
triples.append(_triple(iteration_uri, TG_OBSERVATION_DOCUMENT, _iri(observation_document_id)))
elif observation:
triples.append(_triple(iteration_uri, TG_OBSERVATION, _literal(observation)))
return triples
def agent_final_triples(
final_uri: str,
parent_uri: str,
answer: str,
answer: str = "",
document_id: Optional[str] = None,
) -> List[Triple]:
"""
Build triples for an agent final answer (Conclusion).
@ -122,20 +142,29 @@ def agent_final_triples(
Creates:
- Entity declaration with tg:Conclusion type
- wasDerivedFrom link to parent (last iteration or session)
- The answer text
- Either document reference (if document_id provided) or inline answer
Args:
final_uri: URI of the final answer (from agent_final_uri)
parent_uri: URI of the parent (last iteration or session if no iterations)
answer: The final answer text
answer: The final answer text (used if document_id not provided)
document_id: Optional document URI in librarian (preferred)
Returns:
List of Triple objects
"""
return [
triples = [
_triple(final_uri, RDF_TYPE, _iri(PROV_ENTITY)),
_triple(final_uri, RDF_TYPE, _iri(TG_CONCLUSION)),
_triple(final_uri, RDFS_LABEL, _literal("Conclusion")),
_triple(final_uri, PROV_WAS_DERIVED_FROM, _iri(parent_uri)),
_triple(final_uri, TG_ANSWER, _literal(answer)),
]
if document_id:
# Store reference to document in librarian (as IRI)
triples.append(_triple(final_uri, TG_DOCUMENT, _iri(document_id)))
elif answer:
# Fallback: store inline answer
triples.append(_triple(final_uri, TG_ANSWER, _literal(answer)))
return triples

View file

@ -92,6 +92,10 @@ TG_ARGUMENTS = TG + "arguments"
TG_OBSERVATION = TG + "observation"
TG_ANSWER = TG + "answer"
# Agent document references (for librarian storage)
TG_THOUGHT_DOCUMENT = TG + "thoughtDocument"
TG_OBSERVATION_DOCUMENT = TG + "observationDocument"
# Named graph URIs for RDF datasets
# These separate different types of data while keeping them in the same collection
GRAPH_DEFAULT = "" # Core knowledge facts (triples extracted from documents)