MCMA-OpenMemory epistemic report

Type: agentic-system-analyses/types/agentic-system-epistemic-report.md

Source-and-claim boundary

The question is how retained statements acquire authority to answer questions or inform generation, and whether that authority follows from a content check. The functional inclusions, external exclusions and frozen source are those of boundary.md; source identity and evidence layers are in its Source register. All direct reads below use SRC-1 at that revision. No target execution was performed. The runtime account is supplied implementation analysis, not a trace of operation.

The system's own qualification matters: EPI-CLM-knowledge-policy distinguishes epistemic metadata from proof. It makes concrete reuse-policy claims rather than promising an automatic truth oracle. Encryption, permission checks, revision identity and similarity are not evaluated here as evidence of factual truth. They matter where they change admissibility or lineage.

Entry point or operation	SRC-1 source path	Coverage and records, or remaining limit
CLI knowledge-put/check/show/validate; semantic-index/check/topk	packages/core/src/Cli/CliApplication.php	Direct dispatch read: EPI-RTE-knowledge-validation, EPI-RTE-knowledge-assessment; semantic implementation supplied by RT-RTE-ask-answer. CLI validation accepts supplied state, confidence, reason and optional provenance; it is not an independent factual investigation.
Ordinary ask, generated capture, direct exact/semantic answers	packages/core/src/Ask/AskService.php	Supplied RT-RTE-ask-answer; directly inspected knowledge assessment and directAnswer consumer supplement it.
Web confirm/discard and knowledge inline edit	packages/core/src/Web/WebApplication.php	Direct reads of validation dispatch and editKnowledgeObject; EPI-RTE-knowledge-validation, EPI-RTE-answer-replacement. Additional routes omitted from runtime's separate declarations, not contradictions of its bounded account.
Explicit capture and organizer	packages/core/src/Memory/ExplicitMemoryService.php	Direct capture/output-check/fallback reads; RT-RTE-explicit-capture, RT-OBJ-organizer-instructions.
Interaction persistence, recovery, approval/discard and mutation	packages/core/src/Interaction/InteractionArchiveService.php	Supplied RT-RTE-web-request, RT-RTE-interaction-approval, RT-RTE-memory-mutation. No independent execution evidence of review or synchronization.
Catalog classification and thematic synchronization	packages/core/src/Interaction/InteractionCatalogService.php; packages/core/src/Interaction/ThematicSummaryService.php	Direct reads; EPI-RTE-catalog-classification, EPI-RTE-thematic-summary. These are not generative factual criticism.
Canonical broad recall	packages/core/src/Context/BroadMemoryRecallBuilder.php	Direct canonicalRecall/canonical-item reads plus supplied ask consumer; RT-RTE-ask-answer. Separate from strict knowledge policy.
Conversation, multi-memory and connector reference delivery	packages/core/src/Context/ConversationContextBuilder.php; packages/core/src/Context/MultiMemoryContextBuilder.php	Supplied RT-RTE-context-supply, RT-BAP-memory-reference. Model compliance and semantic fidelity remain unobserved.
Permission, deployment, billing and raw library paths	packages/core/src/Library.php; install.sh	Supplied RT-RTE-permission-change, RT-RTE-deploy-checkout, RT-CLM-permission-scope, RT-RTE-web-request. Operational admission, not truth evaluation.
Shipped evaluators	tests/integration/ask-orchestration.php	Selected fake embedding/generation fixture directly read; tests establish specified branch expectations, not measured natural-language accuracy. No test result or quality experiment supplied.
Hooks/tools and wider repository paths	packages/core/src/Cli/CliApplication.php; packages/core/src/Ask/GenerationProvider.php	CLI dispatch searched, material knowledge branches read. Connector/tool-schema account is supplied in runtime, not independently exhaustively searched here. Other tests, historical tools, full migration/recovery and all normative specifications were not exhaustively inspected. This prevents a system-complete negative about every possible evaluator or extension.

Missing external model internals prevent an inference about internal checking or parameter learning. Missing concrete source propositions, generated outputs and evidence-consuming owner reviews prevent classifying actual generated claims as entailed or ampliative, or crediting any accepted ampliative result. Missing run/causal evidence prevents activation, reliability and benefit conclusions. Available but uninspected repository material is not an access gap or evidence of absence.

Epistemic-object inventory

Object ID	Candidate truth-apt content or none	Claimed role	Evidence source ID and local anchor	Gap/limit
EPI-OBJ-knowledge-record	Answer assertions over the record's question; symbolic metadata about their status	KnowledgeRecord	SRC-1, packages/core/src/Knowledge/KnowledgeRecord.php	Concrete records unprovided; confidence is not calibrated evidence.
EPI-OBJ-canonical-memory	User-supplied or model-organized assertions	Explicit canonical memory	SRC-1, packages/core/src/Memory/ExplicitMemoryService.php	Source text retained; meaning preservation of organized text not checked.
EPI-OBJ-interaction-answer	Question/answer assertions, when answer is truth-apt	Interaction answer	See RT-RTE-web-request, SRC-1, packages/core/src/Interaction/InteractionArchiveService.php	Archive membership is not acceptance.
EPI-OBJ-interaction-catalog	Attribution of topics, projects and named entities to an interaction	Catalog	SRC-1, packages/core/src/Interaction/InteractionCatalogService.php	Array/string checks do not verify attribution.
EPI-OBJ-thematic-summary	Topic label plus bounded answer excerpt	Derived thematic summary	SRC-1, packages/core/src/Interaction/ThematicSummaryService.php	First-sentence extraction can omit qualifications; not an independent finding.
RT-OBJ-organizer-instructions	None as a new factual finding; behavior guidance	See record	See record	Requests preservation; does not check compliance.
RT-OBJ-permission-policy	None required: executable grants	See record	See record	Symbolic operational authority, not evidence for answer truth.
RT-CMP-ollama-generation-provider, RT-CMP-llama-cpp-generation, RT-CMP-bedrock-converse-generation	Generated text can contain assertions; exact candidates not determinable	See records	See records	Opaque inference does not establish entailment, novelty or verification.
RT-CMP-ollama-embedding-provider, RT-CMP-llama-cpp-embedding, RT-CMP-bedrock-titan-embedding	No individually stated proposition in vectors	See records	See records	Numerical retrieval representations are not answer oracles; model fixity limits remain local to each record.

Authority-route ledger

Each row is one function, even when its implementation record also supports another function. Generic progression/read-back fields belong to the cited declaring record: see retains that ownership. The classifications and warrant limits here are epistemic assessments, not copied route declarations. All statuses below concern inspected source or supplied implementation findings, not observed operation.

Route ID	Route function	Architectural status	Object/candidate ID	Content/update relation	Transition or check target	Evaluator/condition and domain	Activation and timing	Possible or observed result	Implemented force	Epistemic authority and scope	Operational authority: behavior permitted, blocked, or changed	Behavioral-authority path: consumer, channel, force, horizon	Evidence source ID and local anchor	Claim IDs or none	Mismatch marker or none	Gap/limit
RT-RTE-ask-answer	content transformation	implemented	See record; EPI-OBJ-interaction-answer	truth-apt transformation: indeterminate	Question/context to generated text	Model inference; no independently supplied reference answer	See record	Candidate answer	See record	Candidate production only	See record	See record and RT-BAP-memory-reference	See record	none	none	No concrete output/premises: non-ampliative, entailed and ampliative alternatives unresolved.
RT-RTE-ask-answer	retention	implemented	EPI-OBJ-knowledge-record	no content change	Generated candidate to knowledge capture	Software capture rule, not a factual evaluator	See record	Unverified by default; configurable metadata	See record	Retained candidate, not accepted claim	See record	See record	See record	none	none	Default does not prevent later owner/configuration promotion.
RT-RTE-explicit-capture	content transformation	implemented	EPI-OBJ-canonical-memory	truth-apt transformation: acquisition/import	User content enters archive; source warrant unknown	User save intent and optional organizer	See record	Source original plus normalized content	See record	Imported assertions, not newly warranted facts	See record	See record	SRC-1, packages/core/src/Memory/ExplicitMemoryService.php	none	none	Organizer subedge below is assessed separately.
RT-RTE-explicit-capture	content transformation	implemented	EPI-OBJ-canonical-memory	truth-apt transformation: indeterminate	Source text to model-normalized text	Preservation instruction; semantic compliance unmeasured	See record	Organized text, or source-preserving fallback	See record	Fallback preserves source text; valid model output has unknown fidelity	See record	See record	SRC-1, packages/core/src/Memory/ExplicitMemoryService.php	none	none	Calling this always non-ampliative would confuse requested behavior with demonstrated preservation.
RT-RTE-explicit-capture	check/evidence production	implemented	EPI-OBJ-canonical-memory	no content change	Organizer JSON fields, lengths and enums	PHP structural validator; schema domain only	See record	Structurally valid output or fallback	See record	Licenses parsing/classification shape, not equivalence or truth	See record	See record	SRC-1, packages/core/src/Memory/ExplicitMemoryService.php	none	none	sourceText parameter is not compared with content in validateOrganizerOutput.
RT-RTE-explicit-capture	disposition/acceptance	implemented	EPI-OBJ-canonical-memory	no content change	Capture result to confirmed/verified metadata	Software assignment following explicit save, not post-organization human review	See record	Confirmed canonical; verified mirror if write succeeds	See record	User-authority convention, not evidence-consuming acceptance of the model transformation	See record	See record	SRC-1, packages/core/src/Memory/ExplicitMemoryService.php	none	none	Mirror and canonical commit can diverge.
RT-RTE-explicit-capture	retention	implemented	EPI-OBJ-canonical-memory, EPI-OBJ-knowledge-record	no content change	Organized/source text to encrypted revisions	See record	See record	Durable canonical and optional mirror/index	See record	Retention carries assigned labels, not new factual evidence	See record	See record	See record	none	none	Not lifecycle integration after evidenced acceptance.
EPI-RTE-knowledge-assessment	check/evidence production	implemented	EPI-OBJ-knowledge-record	no content change	Record admissibility for direct reuse	PHP metadata policy; state, confidence, age/current and reuse-policy domain	See record	reuse/revalidate/reject	See record	Metadata eligibility only; provenance count not independent corroboration	See record	See record	See record	EPI-CLM-knowledge-policy	none	No evidence-source evaluation occurs in assess.
RT-RTE-ask-answer	operational admission/selection/consumption	implemented	EPI-OBJ-knowledge-record	no content change	Assessed exact/semantic record to terminal answer	PHP consumes assessment and semantic thresholds	See record	Direct remembered answer or fallback	See record	Direct reuse licenses dependence on stored labels, not new truth judgment	See record	See record	See record	EPI-CLM-knowledge-policy	none	Similarity selects relevance; cannot establish that answer fits a paraphrased question.
RT-RTE-ask-answer	operational admission/selection/consumption	implemented	EPI-OBJ-canonical-memory	truth-apt transformation: non-ampliative reshaping	Canonical items to concatenated direct answer	Broad recall: lexical subject and maturity-based mapped metadata	See record	Direct canonical recall before strict assessment	See record	Confirmed becomes verified/0.95; not a fresh truth check	See record	See record	SRC-1, packages/core/src/Context/BroadMemoryRecallBuilder.php	EPI-CLM-knowledge-policy	scope distinction	currentRequired is not supplied here; strict-knowledge claim cannot be generalized to all recall.
EPI-RTE-knowledge-validation	disposition/acceptance	implemented	EPI-OBJ-knowledge-record	no content change	Caller decision to persisted validation metadata	Caller chooses state/confidence/reason; PHP shape validation	See record	Confirm/validate or retract	See record	Explicit reliance decision; evidence basis and review criterion not established	See record	See record	See record	EPI-CLM-knowledge-policy	none	An owner decision can be meaningful without constituting an evidenced factual acceptance in supplied evidence.
EPI-RTE-answer-replacement	content transformation	implemented	EPI-OBJ-knowledge-record	truth-apt transformation: acquisition/import	Owner-supplied replacement answer	Caller text; software bounds/permission checks	See record	Verified/0.95 edited answer	See record	Warrant of imported replacement unknown	See record	See record	See record	none	none	Editing is not a test of the original assertion.
RT-RTE-interaction-approval	disposition/acceptance	implemented	EPI-OBJ-interaction-answer	no content change	Archived answer to approved/discarded status	Owner/caller choice; software transition	See record	verified/0.95 or retracted/0.0	See record	Review decision recorded; no observed content review or named evidence criterion	See record	See record	See record	none	none	Existing knowledge can be validated without replacement by archived text.
EPI-RTE-catalog-classification	content transformation	implemented	EPI-OBJ-interaction-catalog	truth-apt transformation: indeterminate	Interaction to semantic labels	Optional model with data-only/no-invention guidance	See record	Classified catalog or fallback	See record	Candidate attribution, not corroboration of answer	See record	See record	See record	none	none	Labels may be extracted or inferred; no concrete output individuates the edge.
EPI-RTE-catalog-classification	check/evidence production	implemented	EPI-OBJ-interaction-catalog	no content change	Catalog JSON and label lengths/types	PHP decode/clean/list limits	See record	Sanitized metadata or fallback	See record	Structural usability only	See record	See record	See record	none	none	No factual names/relations check despite prompt prohibition on invention.
EPI-RTE-thematic-summary	content transformation	implemented	EPI-OBJ-thematic-summary	truth-apt transformation: non-ampliative reshaping	Archived answer to topic-prefixed first excerpt	PHP deterministic extraction, no provider	See record	Bounded derived text	See record	Carries source assertion and metadata; creates no independent corroboration	See record	See record	See record	none	none	Compression can remove qualifications; see concrete deduction in record.
EPI-RTE-thematic-summary	retention	implemented	EPI-OBJ-thematic-summary	no content change	Derived payload to versioned object	PHP persists and derives trusted flag from source metadata	See record	Active or superseded/untrusted object	See record	Inherited status, not acceptance	See record	See record	See record	none	none	Direct canonical recall explicitly excludes these summaries. Other downstream readers not exhaustively established.
RT-RTE-memory-mutation	disposition/acceptance	implemented	EPI-OBJ-canonical-memory	no content change	User edit/delete request to update/tombstone	Deterministic target resolution and nonempty-edit checks	See record	Mutation or needs-target/content response	See record	User-authorized revision/withdrawal, not truth verification	See record	See record	See record	none	none	Retraction/index synchronization is not atomic; history persists.
RT-RTE-context-supply	operational admission/selection/consumption	implemented	EPI-OBJ-knowledge-record, EPI-OBJ-interaction-answer, EPI-OBJ-canonical-memory	truth-apt transformation: non-ampliative reshaping	Stored material to bounded provider reference	Deterministic selection/clipping; separate consumer policies	See record	Context with uncertainty fields	See record	Reference eligibility can be weaker than direct-answer eligibility	See record	See RT-BAP-memory-reference	See record	none	none	Delivery not endorsement or demonstrated influence; clipping may lose scope.
RT-RTE-permission-change	behavior/policy adaptation	implemented	RT-OBJ-permission-policy	non-truth-apt policy/content update: role grants	Submitted policy to later access rules	PHP shape/current-policy/owner-recovery conditions	See record	Admitted or rejected grants	See record	No answer warrant	See record	See RT-BAP-policy-decision	See record	none	none	Direct operational authority without truth-apt learning; no improvement objective inferred.

System-claim versus route comparison

Claim ID	Claimed operation or warrant	Claim source ID/anchor and evidence layer	Doctrine/design support	Implemented route IDs	Observed-run support	Causal support and design limits	Supported conclusion	Mismatch/unknown
EPI-CLM-knowledge-policy	See record	See record	See record	EPI-RTE-knowledge-assessment, RT-RTE-ask-answer, EPI-RTE-knowledge-validation	None supplied	No quality comparison; metadata can be caller-assigned	Strict knowledge retrieval enforces its policy, not factual correctness. This is consistent with the source's explicit metadata-not-proof qualification.	Canonical broad recall is a different path; current-data blocking cannot be generalized to it. “Exact miss” in the document's semantic diagram is narrower than runtime's semantic fallback after a nonreusable exact hit.
RT-CLM-permission-scope	See record	See record	See record	See record	See record	See record	Actor-aware authorization limits covered reads and writes, not the truth of authorized content.	See record; host and raw-API scope matters.

Bounded conclusion

MCMA retains imported statements and generated candidates, preserves source/lineage metadata, and makes both direct-answer and model-reference use possible. Acquisition is not production of warranted propositions. User source originals remain available in explicit memory; the organizer's output may differ, and its structural validator supplies no meaning comparison. Generated candidates default to unverified, but explicit saving, owner validation and inline editing assign stronger labels through separate routes. These are different admission mechanisms, not a universal human-review gate. See RT-RTE-ask-answer, RT-RTE-explicit-capture, EPI-RTE-knowledge-validation, EPI-RTE-answer-replacement.

The material automatic check targets eligibility of a metadata-bearing record for direct knowledge reuse. Its criterion is supported/verified state, sufficient confidence, acceptable freshness/current requirement and reuse policy, not independent evidence of answer truth. Its result enforces direct reuse/fallback in that scope. Semantic similarity and reranking add selection force, not new factual warrant. Canonical recall instead gives confirmed memory direct-answer authority through a maturity-to-validation mapping and does not consume currentRequired. Retraction in knowledge therefore cannot be assumed to withdraw the canonical source automatically. See EPI-RTE-knowledge-assessment, RT-RTE-ask-answer, EPI-RTE-knowledge-validation.

Owner decisions record changed reliance; their actual evidence, acceptance criterion, intended factual scope and exercise are unprovided. Thematic synchronization connects retained outputs to interaction references and topics, but this is derived organization and retention, not demonstrated lifecycle integration after evidence-consuming acceptance. Deterministic excerpt construction supports non-ampliative reshaping, not a new explanation. No inspected concrete candidate establishes an ampliative proposition accepted against evidence; this is a limit of this source/run evidence, not a verdict that every possible user review lacks warrant. See RT-RTE-interaction-approval, EPI-RTE-thematic-summary.

Later reference consumption is a weaker authority path than direct answering. Selection can deliver material that strict direct reuse would require revalidating; connector instructions preserve its advisory/untrusted role but cannot enforce model obedience. No supplied observation demonstrates context influence or benefit. Parameter updates are not sent by the inspected inference connectors; exact model versions and provider-internal state remain uninspected as specified in the component records. Policy grants can directly change covered access behavior without any truth-apt claim route. Retention, classification and changing labels alone establish neither improved capacity, reflection, autonomous criticism nor exercised self-improvement. These remain independent questions, not one system-wide score. See RT-RTE-context-supply, RT-BAP-memory-reference, RT-RTE-permission-change.

Shared records

Components

none declared in this member

Operative objects

EPI-OBJ-knowledge-record

Label: KnowledgeRecord

Natural-language or JSON answer plus symbolic intent, provenance, confidence, validation history, freshness and relation fields. Storage: encrypted versioned library JSON under intent-derived knowledge references. SRC-1, packages/core/src/Knowledge/KnowledgeRecord.php, packages/core/src/Knowledge/KnowledgeService.php. Distinct from supplied policy/instruction objects: this is retained answer content consumed by RT-RTE-ask-answer, not that route itself. No supplied operative-object counterpart after comparison.

'evidence_count' => count($provenance), --- packages/core/src/Knowledge/KnowledgeRecord.php

if (($epistemic['evidence_count'] ?? -1) !== count($record['provenance'])) throw new RuntimeException('Knowledge evidence_count mismatch');

--- packages/core/src/Knowledge/KnowledgeRecord.php

These checks establish a count of entries, not independent evidence assessment.

EPI-OBJ-canonical-memory

Label: Explicit canonical memory

Natural-language source and organized content with symbolic classification, retrieval and organization metadata, stored as encrypted versioned JSON under memory://user references. SRC-1, packages/core/src/Memory/ExplicitMemoryService.php. Output object of RT-RTE-explicit-capture, distinct from its static organizer instructions and from the knowledge mirror; no supplied object counterpart.

'source' => [ 'type' => 'explicit-user-request', 'original' => $sourceText, ],

--- packages/core/src/Memory/ExplicitMemoryService.php

EPI-OBJ-interaction-answer

Label: Interaction answer

Natural-language/JSON answer and question with symbolic route, provider, validation and relation metadata in the encrypted interaction archive; the storage and later consumers are supplied by RT-RTE-web-request, SRC-1, packages/core/src/Interaction/InteractionArchiveService.php. Distinct from knowledge capture: archiving is independent of rememberGenerated. This declares the retained object, not a duplicate of the web route; no supplied operative-object counterpart. Evidence passages for independent archive writing and validation are in the cited runtime route and RT-RTE-interaction-approval.

EPI-OBJ-interaction-catalog

Label: Interaction catalog

Natural-language semantic labels/title consumed as symbolic lists in interaction catalog metadata, retained inside encrypted archive JSON. SRC-1, packages/core/src/Interaction/InteractionCatalogService.php, with persistence supplied by RT-RTE-interaction-approval. Distinct material part from the answer: classification affects organization, not its truth. No supplied operative-object counterpart. Its load-bearing prompt/cleaning evidence is in EPI-RTE-catalog-classification.

EPI-OBJ-thematic-summary

Label: Thematic summary

Natural-language topic-prefixed answer excerpt plus symbolic validation, source_storage_hash and interaction references. Encrypted library JSON under topic/interaction-specific references; independently revisioned derived objects, not canonical question/answer transcripts. SRC-1, packages/core/src/Interaction/ThematicSummaryService.php. Distinct from the archive and knowledge mirror; supplied RT-RTE-interaction-approval mentions synchronization but declares no counterpart object.

'source_storage_hash'=>$sourceStorageHash, --- packages/core/src/Interaction/ThematicSummaryService.php

Routes

EPI-RTE-knowledge-assessment

Label: Knowledge metadata assessment

KnowledgeRecord::assess validates a record and computes eligibility; KnowledgeService::directAnswer consumes reusable before including answer/provenance/relations. Owner: PHP code, not a model or human evaluator. Context: currentRequired, minConfidence and time. Effects: returned symbolic decision, not changed source content. SRC-1, packages/core/src/Knowledge/KnowledgeRecord.php, packages/core/src/Knowledge/KnowledgeService.php. This isolates a reusable check used within RT-RTE-ask-answer but also exposed by CLI knowledge-check and direct APIs; not coextensive with the supplied encompassing ask route. No supplied standalone assessment record.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: reusable/decision/reasons/stale and status metadata; directAnswer includes stored answer only for reuse.
• Later read-back: inapplicable — assessment itself writes no retained result; it reads the current record on each invocation.
• Delegated visibility: inapplicable — deterministic PHP check delegates no evaluator call.
• Selection predicate: disputed/retracted or never-direct rejects; insufficient state/confidence, unacceptable age or nonimmutable current request yields revalidate; otherwise reuse.
• Invalidation or expiry: reference time is last_validated_at or captured_at; immutable bypasses age/current, always bypasses stale-only rejection, not validation/confidence/current conditions.
• Activation or effect: Direct-answer consumers enforce the reusable result; no independent factual check is licensed.
• Evidence limits: Metadata eligibility is implemented; source-truth evaluation, deployed execution and accuracy are unobserved.

if (!in_array($state, ['supported','verified'], true)) $reasons[] = 'validation-insufficient'; if ($confidence < $minConfidence) $reasons[] = 'confidence-below-threshold';

--- packages/core/src/Knowledge/KnowledgeRecord.php

if ($stale && $policy !== 'always') $reasons[] = 'stale'; if ($currentRequired) $reasons[] = 'current-data-requested';

--- packages/core/src/Knowledge/KnowledgeRecord.php

if ($assessment['reusable']) { $result['answer'] = $record['answer']; $result['provenance'] = $record['provenance']; $result['relations'] = $record['relations'];

--- packages/core/src/Knowledge/KnowledgeService.php

PHP proposes the decision, the caller supplies thresholds/current requirement, and consuming PHP branches enforce it. No human per-item approval or answer oracle is invoked by this check. Mode: open retrieval requests, not an improvement experiment. Revalidate is a disposition request, not evidence of completed external research.

EPI-RTE-knowledge-validation

Label: Knowledge validation state update

CLI knowledge-validate and web memory confirm/discard reach KnowledgeService::validateKnowledge/validateId and KnowledgeRecord::withValidation. Caller proposes state, confidence, reason and optional additional provenance; PHP checks structure/permissions and retains a new revision. SRC-1, packages/core/src/Cli/CliApplication.php, packages/core/src/Web/WebApplication.php, packages/core/src/Knowledge/KnowledgeService.php, packages/core/src/Knowledge/KnowledgeRecord.php. Distinct from RT-RTE-interaction-approval: target is a knowledge record directly, not an interaction and its associated synchronization. Related RT-RTE-ask-answer later consumes the result but does not own this admission route.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: Updated revision/state/confidence; web additionally returns inspection and optional semantic synchronization.
• Later read-back: Knowledge assessment and answer selection use the new state/time/confidence; history and appended provenance persist.
• Delegated visibility: Web resolves session principal then performs owner-role update; optional embedding receives content for refresh, not factual review authority.
• Selection predicate: Existing actor-readable record, valid state/confidence and nonempty reason; web restricts actions to confirm/discard, CLI accepts supported state vocabulary.
• Invalidation or expiry: New status supersedes current eligibility; web discard maps to retracted/0.0. This route does not edit canonical maturity in a related explicit memory.
• Activation or effect: State update can permit or block subsequent direct knowledge reuse; actual use unobserved.
• Evidence limits: Actor string is not evidence of human deliberation. Optional provenance and a reason do not prove evidence was inspected or a factual criterion met.

$targetState=$action==='confirm'?'verified':'retracted'; $targetConfidence=$action==='confirm'?0.95:0.0; --- packages/core/src/Web/WebApplication.php

$record['epistemic']['last_validated_at'] = $at;

--- packages/core/src/Knowledge/KnowledgeRecord.php

$record = KnowledgeRecord::withValidation($record, $state, $confidence, $reason, $additionalProvenance); --- packages/core/src/Knowledge/KnowledgeService.php

Trigger: explicit caller validation. Software can reject malformed/unauthorized updates, not independently contradict an authorized factual judgment. Guidance is the state/schema vocabulary; persisted change is metadata/history/provenance. Later updates can change status again; revision lineage is not an automatic rollback transaction. No answer oracle; open owner/caller review, no improvement curriculum. No source text for an evidence-consuming truth acceptance criterion is supplied.

EPI-RTE-answer-replacement

Label: Owner answer replacement

Web editKnowledgeObject calls KnowledgeService::replaceAnswerId, replacing answer text and setting stable/year-long reuse metadata and verified/0.95 status. SRC-1, packages/core/src/Web/WebApplication.php, packages/core/src/Knowledge/KnowledgeService.php. Distinct from RT-RTE-memory-mutation, which targets canonical memories, and from validation-only updates above. It is not a renamed interaction-approval route.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: Revision hashes, verified status and optional refreshed semantic entry; web reinspection/billing result.
• Later read-back: Replaced answer and refreshed metadata reach knowledge assessment and retrieval; current hash changes.
• Delegated visibility: Caller supplies replacement; server uses owner; optional embedding provider indexes, not judges, it.
• Selection predicate: Existing valid knowledge ID; trimmed nonempty replacement no more than 32768 bytes; actor-authorized update.
• Invalidation or expiry: Old revision hash retained; freshness becomes stable with 31536000 seconds; later state changes can withdraw direct reuse.
• Activation or effect: New answer text becomes current and labelled verified without a separate content evaluator.
• Evidence limits: No observed edit, truth check, actual later use or atomic semantic-refresh guarantee.

$record['answer']=[ 'format'=>'text', 'value'=>$answer, ];

--- packages/core/src/Knowledge/KnowledgeService.php

$record=KnowledgeRecord::withValidation( $record, 'verified', 0.95, 'owner-edited-in-library', $additionalProvenance ); --- packages/core/src/Knowledge/KnowledgeService.php

Caller proposes/decides content; software enforces shape and access but supplies no factual veto or answer oracle. Trigger is open user editing; bounds and permission checks guide admission, not an improvement objective. Rejection is for bad input/access; previous hashes provide history, no automatic rollback. This imports a replacement with unknown source warrant rather than deriving it from the earlier record.

EPI-RTE-catalog-classification

Label: Interaction catalog classification Part of: RT-RTE-interaction-approval

This is the optional classification subroute of interaction approval, not the whole approval decision. InteractionCatalogService sends question, answer and existing catalog to a generator with data-only/no-invention instructions; PHP decodes and cleans the result or uses deterministic fallback. SRC-1, packages/core/src/Interaction/InteractionCatalogService.php. Natural-language labels become symbolic organization fields; answer truth remains unevaluated.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: Catalog with provider/usage metadata or fallback/error status.
• Later read-back: Approval persists catalog in interaction; thematic topic selection consumes topics/projects/entities. General topic correctness is unverified.
• Delegated visibility: Model receives question/answer/existing catalog, not Library access.
• Selection predicate: Configured provider attempts classification; absent provider or exception uses fallback; arrays/labels are cleaned and bounded.
• Invalidation or expiry: Empty returned topics fall back to existing topics; later classification may replace topic organization and supersede derived summaries.
• Activation or effect: Semantic labels influence retained thematic paths; they do not revise the answer or establish its truth.
• Evidence limits: No observed classification, fidelity metric or causal benefit; provider internals uninspected.

• Do not invent names, relationships, projects, people or characters. --- packages/core/src/Interaction/InteractionCatalogService.php

'classification_status'=>'classified-on-owner-approval', --- packages/core/src/Interaction/InteractionCatalogService.php

$item=self::clean($item,80); if($item!==''&&!in_array($item,$out,true)) $out[]=$item; if(count($out)>=8) break;

--- packages/core/src/Interaction/InteractionCatalogService.php

Proposer: optional model, or fallback from existing catalog. Decider/veto: PHP JSON/label processing with exception fallback; owner approves interaction before this returned classification, not each new label afterward. Mode is open review, not an experiment. No answer oracle. Guidance requests source-faithful cataloging; what persists is labels/status, not checked criticisms of the answer.

EPI-RTE-thematic-summary

Label: Deterministic thematic synchronization

InteractionArchiveService synchronization invokes ThematicSummaryService::sync, which derives topics from catalog or canonical classification, extracts an answer excerpt, persists per-topic payloads and supersedes obsolete topic paths. SRC-1, packages/core/src/Interaction/ThematicSummaryService.php; archive invocation is supplied by RT-RTE-web-request and RT-RTE-interaction-approval. Distinct submechanism spanning persistence and validation triggers, not a duplicate of either encompassing route. It performs extraction/retention, not generative criticism.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: Relevant/no-topic result, summary revision metadata and zero model usage.
• Later read-back: Existing derived objects are reloaded for update/supersession. Broad canonical recall excludes thematic summaries; no new direct-answer authority is established for this derived text.
• Delegated visibility: inapplicable — deterministic library operations make no model call.
• Selection predicate: Catalog topics take precedence, otherwise projects/entities and nongeneric canonical category labels, bounded to fallback topic limits.
• Invalidation or expiry: Changed desired paths mark old objects superseded and trusted_for_conclusions=false; current payload records source_storage_hash. This hash alone is not a consumer-side freshness check.
• Activation or effect: Retained topic organization and trust flag derive from existing source metadata; no observed later factual reliance.
• Evidence limits: Does not independently verify the answer; trusted flag is not evidence of a downstream consumer enforcing acceptance. Broader thematic readers were not exhaustively inspected.

$answerText=self::excerpt($answerText,220); --- packages/core/src/Interaction/ThematicSummaryService.php

$parts=['Tema: '.$topic.'.']; if($answerText!=='') $parts[]='Punto derivado: '.$answerText;

--- packages/core/src/Interaction/ThematicSummaryService.php

$sentences=preg_split('/(?<=[.!?])\s+/u',$value,2); $first=trim((string)($sentences[0]??$value)); return $first!==''?$first:$value;

--- packages/core/src/Interaction/ThematicSummaryService.php

'trusted_for_conclusions'=>$state==='verified'&&$confidence>=0.75, --- packages/core/src/Interaction/ThematicSummaryService.php

Concrete source deduction, not execution: with topic Release and answer The release is ready. Only for staging., clean leaves the short ASCII text unchanged, excerpt splits at the first period-plus-space, and summaryText produces Tema: Release. Punto derivado: The release is ready. The second sentence's scope restriction is lost. This establishes deterministic non-ampliative extraction with a fidelity limit, not novel factual derivation. Source premises remain no more warranted than their input.

Supersession evidence is supplied by RT-RTE-interaction-approval. The canonical exclusion is an implemented countercase to treating every confirmed derived summary as a direct answer:

if( (is_array($content)&&isset($content['thematic_summary_version'])) || preg_match('#^memory://user/temas/[^/]+/resumenes/#',$logicalRef)===1 ){ return null; }

--- packages/core/src/Context/BroadMemoryRecallBuilder.php

Claims

EPI-CLM-knowledge-policy

Label: Remembered knowledge reuse policy

SRC-1 doctrine/design, docs/KNOWLEDGE-REUSE.md: exact reuse is metadata-gated; semantic similarity does not override permissions or epistemic policy, and nonimmutable current requests require revalidation. This is not a proof claim. Distinct from supplied RT-CLM-permission-scope: it concerns answer admissibility and qualification of epistemic metadata, not only authorization APIs.

Only reuse returns the remembered answer. --- docs/KNOWLEDGE-REUSE.md

Similarity does not override permissions or epistemic policy. --- docs/KNOWLEDGE-REUSE.md

Confidence and provenance are epistemic metadata, not absolute proof. --- docs/KNOWLEDGE-REUSE.md

A caller can set currentRequired=true. Non-immutable remembered knowledge then requires revalidation even when semantic similarity is high. --- docs/KNOWLEDGE-REUSE.md

Evidenced absences

none declared in this member

Behavioral-authority paths

none declared in this member