[semantic-memory] Cross-memory entity/concept reconciliation, aliasing, ontology alignment, and reversible identity links

nsaspy commented

2026-09-08 01:57:03 +00:00

Owner

Parent: #4
Depends on: #6
Integrates with: #7
Upstream IR: prolog-rlm#392

Goal

Add persistent, provenance-preserving cross-memory semantic identity reconciliation so repeated mentions of the same real concept/entity/event can be reasoned over together without destructively rewriting source-local identities.

Large corpora will contain:

Prolog-RLM
prolog rlm
the RLM library
lost-rob0t/prolog-rlm
this project
it

and they may or may not refer to the same entity depending on source/context/time.

The memory system needs explicit identity hypotheses and canonical views, not blind string equality and not irreversible merges.

Required relation classes

Support persistent links equivalent to:

alias / synonym
same-as
likely-same-as
different-from
subclass/superclass
concept-equivalent
concept-overlaps
renamed-from / renamed-to
version-of
instance-of
part-of
source-local-refers-to
ambiguous-candidate

Exact vocabulary should reuse upstream IR where possible.

Reversible reconciliation

Never destroy original semantic record/entity IDs.

A reconciliation decision is itself append-only knowledge with provenance:

identity_link(LinkId, A, Relation, B, Evidence, Status).

It may later be superseded/retracted if new evidence shows the merge was wrong.

Canonical/current query views may follow active high-confidence identity links, but history and original source identities remain recoverable.

Evidence for reconciliation

Use symbolic evidence where possible:

exact stable IDs/URLs/repository names;
aliases explicitly stated by sources;
co-occurring properties;
temporal/version relations;
namespace/project scope;
type constraints;
source-local coreference links from compiler;
mutually exclusive facts proving difference;
optional embedding similarity as a sensor;
optional model fallback for fuzzy candidate comparison.

A model suggestion is not an irreversible merge.

Ontology alignment

Domain sources may use different vocabularies for equivalent or related concepts.

Support versioned mappings such as:

source concept `maintainer`
source concept `owner`
project ontology concept `repository_maintainer`

Alignment may be exact, narrower, broader, overlapping or unresolved. Do not flatten all similar words into synonymy.

Events and propositions

Reconciliation should eventually apply not only to entities but also where useful to:

same event described by multiple sources;
same proposition/claim independently stated;
same procedure with variant wording;
same rule/constraint semantically duplicated.

Keep corroboration/dedup separate from destructive identity merging.

Scope / time

Identity may be contextual:

same display name, different people;
renamed project across time;
same role name in different organizations;
source-local pronouns;
versioned component identities.

Reconciliation must consider namespace/time/type constraints.

Query integration

#7 general semantic queries should be able to use active identity/alignment links to join knowledge across memories while reporting when a result depends on uncertain reconciliation.

Result explanations include identity links used in the proof.

Acceptance

Explicit aliases across two memories reconcile to one queryable entity while retaining original IDs.
Same surface name for two incompatible entities remains distinct.
A mistaken identity link can be retracted without rewriting source records.
Renamed/versioned entity relation preserves temporal lineage.
Ontology mapping distinguishes exact/narrower/broader/overlap.
Query can chain rule/fact across memories via an active identity link.
Uncertain identity link makes dependent query result explicitly uncertain/conditional.
Model fallback cannot silently create irreversible merge.
Reconciliation is namespace/project aware.
Explanation identifies identity/alignment evidence used.

Refs #4 #6 #7 and prolog-rlm#392.

Parent: #4 Depends on: #6 Integrates with: #7 Upstream IR: prolog-rlm#392 ## Goal Add persistent, provenance-preserving **cross-memory semantic identity reconciliation** so repeated mentions of the same real concept/entity/event can be reasoned over together without destructively rewriting source-local identities. Large corpora will contain: ```text Prolog-RLM prolog rlm the RLM library lost-rob0t/prolog-rlm this project it ``` and they may or may not refer to the same entity depending on source/context/time. The memory system needs explicit identity hypotheses and canonical views, not blind string equality and not irreversible merges. ## Required relation classes Support persistent links equivalent to: ```text alias / synonym same-as likely-same-as different-from subclass/superclass concept-equivalent concept-overlaps renamed-from / renamed-to version-of instance-of part-of source-local-refers-to ambiguous-candidate ``` Exact vocabulary should reuse upstream IR where possible. ## Reversible reconciliation Never destroy original semantic record/entity IDs. A reconciliation decision is itself append-only knowledge with provenance: ```prolog identity_link(LinkId, A, Relation, B, Evidence, Status). ``` It may later be superseded/retracted if new evidence shows the merge was wrong. Canonical/current query views may follow active high-confidence identity links, but history and original source identities remain recoverable. ## Evidence for reconciliation Use symbolic evidence where possible: - exact stable IDs/URLs/repository names; - aliases explicitly stated by sources; - co-occurring properties; - temporal/version relations; - namespace/project scope; - type constraints; - source-local coreference links from compiler; - mutually exclusive facts proving difference; - optional embedding similarity as a sensor; - optional model fallback for fuzzy candidate comparison. A model suggestion is not an irreversible merge. ## Ontology alignment Domain sources may use different vocabularies for equivalent or related concepts. Support versioned mappings such as: ```text source concept `maintainer` source concept `owner` project ontology concept `repository_maintainer` ``` Alignment may be exact, narrower, broader, overlapping or unresolved. Do not flatten all similar words into synonymy. ## Events and propositions Reconciliation should eventually apply not only to entities but also where useful to: - same event described by multiple sources; - same proposition/claim independently stated; - same procedure with variant wording; - same rule/constraint semantically duplicated. Keep corroboration/dedup separate from destructive identity merging. ## Scope / time Identity may be contextual: - same display name, different people; - renamed project across time; - same role name in different organizations; - source-local pronouns; - versioned component identities. Reconciliation must consider namespace/time/type constraints. ## Query integration #7 general semantic queries should be able to use active identity/alignment links to join knowledge across memories while reporting when a result depends on uncertain reconciliation. Result explanations include identity links used in the proof. ## Acceptance - [ ] Explicit aliases across two memories reconcile to one queryable entity while retaining original IDs. - [ ] Same surface name for two incompatible entities remains distinct. - [ ] A mistaken identity link can be retracted without rewriting source records. - [ ] Renamed/versioned entity relation preserves temporal lineage. - [ ] Ontology mapping distinguishes exact/narrower/broader/overlap. - [ ] Query can chain rule/fact across memories via an active identity link. - [ ] Uncertain identity link makes dependent query result explicitly uncertain/conditional. - [ ] Model fallback cannot silently create irreversible merge. - [ ] Reconciliation is namespace/project aware. - [ ] Explanation identifies identity/alignment evidence used. Refs #4 #6 #7 and prolog-rlm#392.

nsaspy commented

2026-09-08 01:57:35 +00:00

Author

Owner

Machine Spirit #401C handoff — reversible procedure/action/goal alignment

Depth 4C PIFF extends the existing reversible reconciliation principle beyond entities/concepts to procedures, semantic actions and goal/task identities where useful.

Candidate relation classes should distinguish, with provenance and versioned mapping evidence:

exact_procedure_equivalent
variant_of
refines
abstracts
implements_goal
projects_to
overlaps
approximate_correspondence
incompatible
unknown

Do not destructively deduplicate procedures because names/steps look similar. Original source-local procedure/action/goal IDs remain recoverable, and a mistaken mapping must be retractable/supersedable without rewriting history.

Hard boundary:

procedure/action identity mapping
    != host action binding
    != principal mapping
    != capability
    != authority

In particular, two memories both mentioning restart_service, administrator, or an equivalent-looking workflow task cannot create a host binding/principal relation by lexical equality. Query explanations should expose any procedure/action alignment used in a composed result. See prolog-rlm#401 Depth-4C PIFF/PSC design; #401D owns computational realization.

## Machine Spirit #401C handoff — reversible procedure/action/goal alignment Depth 4C PIFF extends the existing reversible reconciliation principle beyond entities/concepts to **procedures, semantic actions and goal/task identities** where useful. Candidate relation classes should distinguish, with provenance and versioned mapping evidence: ```text exact_procedure_equivalent variant_of refines abstracts implements_goal projects_to overlaps approximate_correspondence incompatible unknown ``` Do not destructively deduplicate procedures because names/steps look similar. Original source-local procedure/action/goal IDs remain recoverable, and a mistaken mapping must be retractable/supersedable without rewriting history. Hard boundary: ```text procedure/action identity mapping != host action binding != principal mapping != capability != authority ``` In particular, two memories both mentioning `restart_service`, `administrator`, or an equivalent-looking workflow task cannot create a host binding/principal relation by lexical equality. Query explanations should expose any procedure/action alignment used in a composed result. See prolog-rlm#401 Depth-4C PIFF/PSC design; #401D owns computational realization.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

Machine Spirit #402A identity/ontology durability handoff

Depth 5A (prolog-rlm#402A) confirms this issue's reversible-reconciliation direction and makes it part of the authoritative ledger contract.

Required persistence shape:

original entity/concept/procedure/event IDs
        remain immutable/addressable
              +
append-only identity/ontology mapping events
              +
versioned current canonicalization/alignment projection

A mapping may be exact, likely-same, different, narrower/broader, rename/version, merge, split, overlap, incompatible or unresolved. Activating, superseding or withdrawing a mapping changes the current mapping view, never the historical source-local objects.

Ontology/schema version changes likewise use explicit versioned mapping/adaptation receipts. Historical objects stay interpretable under their original vocabulary/schema. Hartung/Groß/Rahm's invertible ontology-diff work (2010, https://arxiv.org/abs/1010.0122) is a strong design precedent for preserving nontrivial merge/split evolution rather than flattening it to rename-only migration.

Identity links involving principal-like entities remain semantic knowledge only and can never authenticate/bind a runtime principal or transfer authority.

Refs: lost-rob0t/prolog-rlm#402, symbolic-memory#6/#7.

## Machine Spirit #402A identity/ontology durability handoff Depth 5A (`prolog-rlm#402A`) confirms this issue's reversible-reconciliation direction and makes it part of the authoritative ledger contract. Required persistence shape: ```text original entity/concept/procedure/event IDs remain immutable/addressable + append-only identity/ontology mapping events + versioned current canonicalization/alignment projection ``` A mapping may be exact, likely-same, different, narrower/broader, rename/version, merge, split, overlap, incompatible or unresolved. Activating, superseding or withdrawing a mapping changes the **current mapping view**, never the historical source-local objects. Ontology/schema version changes likewise use explicit versioned mapping/adaptation receipts. Historical objects stay interpretable under their original vocabulary/schema. Hartung/Groß/Rahm's invertible ontology-diff work (2010, https://arxiv.org/abs/1010.0122) is a strong design precedent for preserving nontrivial merge/split evolution rather than flattening it to rename-only migration. Identity links involving principal-like entities remain semantic knowledge only and can never authenticate/bind a runtime principal or transfer authority. Refs: lost-rob0t/prolog-rlm#402, symbolic-memory#6/#7.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

MACHINE-SPIRIT #402B handoff — identity/dedup privacy boundary

Reconciliation must not treat globally equal payload hashes as implicit same-entity/same-evidence proof across isolated authority domains. Structural semantic equality, source-episode identity, independent corroboration, and correlated/copy dependence remain separate.

Where equality disclosure is sensitive, content commitments/dedup should be namespace/trust-domain scoped (or otherwise policy-protected), and an unauthorized caller must not learn private equality through object IDs/digests. Redaction of one source cannot erase or collapse independent evidence from another source.

See prolog-rlm#402B.

## MACHINE-SPIRIT #402B handoff — identity/dedup privacy boundary Reconciliation must not treat globally equal payload hashes as implicit same-entity/same-evidence proof across isolated authority domains. Structural semantic equality, source-episode identity, independent corroboration, and correlated/copy dependence remain separate. Where equality disclosure is sensitive, content commitments/dedup should be namespace/trust-domain scoped (or otherwise policy-protected), and an unauthorized caller must not learn private equality through object IDs/digests. Redaction of one source cannot erase or collapse independent evidence from another source. See prolog-rlm#402B.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

MACHINE-SPIRIT #402C handoff — cross-ledger identity/dedup/corroboration

Federation makes #10's reversible identity model mandatory across authority domains, not just memories inside one store.

Hard additions:

same semantic/content fingerprint across ledgers != same assertion episode/source/event;
replicas/backups/mirrors of one logical ledger are duplicate observation paths, never independent corroboration;
one externally signed statement registered in multiple transparency services is one statement origin + multiple registration receipts;
independent issuers asserting the same proposition remain separate evidence episodes even when proposition structure deduplicates;
identity/principal name equality across ledgers never transfers authority;
cross-domain content commitments/hashes must be scoped or deliberately hidden where equality testing would leak private membership;
active mappings remain versioned/reversible and federation cuts retain exact mapping generation so historical replay does not silently adopt today's identity decisions.

Query explanations must identify which cross-ledger identity/alignment/dependence links were used. Full design + C fixtures are on prolog-rlm#402.

### MACHINE-SPIRIT #402C handoff — cross-ledger identity/dedup/corroboration Federation makes #10's reversible identity model mandatory across **authority domains**, not just memories inside one store. Hard additions: - same semantic/content fingerprint across ledgers != same assertion episode/source/event; - replicas/backups/mirrors of one logical ledger are duplicate observation paths, never independent corroboration; - one externally signed statement registered in multiple transparency services is one statement origin + multiple registration receipts; - independent issuers asserting the same proposition remain separate evidence episodes even when proposition structure deduplicates; - identity/principal name equality across ledgers never transfers authority; - cross-domain content commitments/hashes must be scoped or deliberately hidden where equality testing would leak private membership; - active mappings remain versioned/reversible and federation cuts retain exact mapping generation so historical replay does not silently adopt today's identity decisions. Query explanations must identify which cross-ledger identity/alignment/dependence links were used. Full design + C fixtures are on prolog-rlm#402.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

Machine Spirit #402D handoff: identity/ontology/mapping versions are first-class projection invalidation generations. A mapping change invalidates only declared consumers where possible; huge fan-out changes use generation bump + stale partition + bounded rebuild instead of eager rewrite. Sparse federation cuts must fingerprint the mapping generation they used, so historical replay cannot silently adopt today’s identity alignment. See prolog-rlm#402 D18-D19/D32.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

Upstream Machine Spirit #403B handoff — identity attention is candidate-only until reconciled

prolog-rlm#403B adds explicit seek_identity_counterexample / identity-disambiguation retrieval obligations for queries whose reasoning would cross uncertain identity links.

Hard retrieval boundary:

embedding-near / lexical-near / graph-near
    != same entity
    != same event
    != same proposition

Similarity sensors may propose reconciliation candidates, but OATH-TAPS must retrieve active different-from, namespace/time/type conflicts, competing mappings and identity-link provenance before using uncertain joins under profiles that require disambiguation. Cross-namespace similarity is filtered by read authority before scoring/expansion. Source copies/equal hashes do not themselves prove semantic identity or independent corroboration.

Identity/mapping generation belongs in projection lineage so a reconciliation change invalidates only affected attention/query results. Refs lost-rob0t/prolog-rlm#403 #392 #400 and symbolic-memory#6/#7.

## Upstream Machine Spirit #403B handoff — identity attention is candidate-only until reconciled `prolog-rlm#403B` adds explicit `seek_identity_counterexample` / identity-disambiguation retrieval obligations for queries whose reasoning would cross uncertain identity links. Hard retrieval boundary: ```text embedding-near / lexical-near / graph-near != same entity != same event != same proposition ``` Similarity sensors may propose reconciliation candidates, but OATH-TAPS must retrieve active `different-from`, namespace/time/type conflicts, competing mappings and identity-link provenance before using uncertain joins under profiles that require disambiguation. Cross-namespace similarity is filtered by read authority before scoring/expansion. Source copies/equal hashes do not themselves prove semantic identity or independent corroboration. Identity/mapping generation belongs in projection lineage so a reconciliation change invalidates only affected attention/query results. Refs lost-rob0t/prolog-rlm#403 #392 #400 and symbolic-memory#6/#7.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

MACHINE-SPIRIT prolog-rlm#403C handoff — identity/source-dependence for retrieval federation

MOSAIC-TAPS makes #10 a required interoperability input. Cross-backend dedup/join/corroboration must use reversible provenance-carrying identity/alignment and source-dependence evidence rather than string/hash/backend equality.

Hard distinctions:

same retriever hit != same semantic entity
same content hash  != same assertion/event identity
backend independence != source independence
similarity          != same-as

One syndicated origin found by lexical, vector and web retrievers remains one dependence component for corroboration pressure. Uncertain identity links must make dependent fused/joined results conditional and invalidate/rebuild projection receipts when superseded/retracted.

Refs lost-rob0t/prolog-rlm#397 #403 #392 #381 and symbolic-memory#6/#7/#9.

### MACHINE-SPIRIT prolog-rlm#403C handoff — identity/source-dependence for retrieval federation MOSAIC-TAPS makes #10 a required interoperability input. Cross-backend dedup/join/corroboration must use reversible provenance-carrying identity/alignment and source-dependence evidence rather than string/hash/backend equality. Hard distinctions: ```text same retriever hit != same semantic entity same content hash != same assertion/event identity backend independence != source independence similarity != same-as ``` One syndicated origin found by lexical, vector and web retrievers remains one dependence component for corroboration pressure. Uncertain identity links must make dependent fused/joined results conditional and invalidate/rebuild projection receipts when superseded/retracted. Refs lost-rob0t/prolog-rlm#397 #403 #392 #381 and symbolic-memory#6/#7/#9.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

Machine Spirit #404B / CITADEL-EPOCH identity/dependence handoff

Identity/source reconciliation must participate in promotion-evidence independence and contamination checks.

New IDs must not manufacture new validation origins. Aliases, mirrors, same-event descriptions, semantic paraphrases/derivatives, parent/child candidates and repeated views of one underlying episode may remain distinct records while belonging to the same evidence-dependence/origin component where warranted.

Conversely, equality/alias hypotheses are not enough to collapse genuinely independent sources without evidence. Origin-group decisions remain reversible/provenance-bearing and may themselves be uncertain.

Required consequence for #404: record independence != source independence != causal independence; evaluation splitting/grouping must consume active identity/dependence knowledge so near-duplicate or renamed records cannot bypass holdout/contamination policy. Refs prolog-rlm#404B CITADEL fixtures B2, B5-B8, B30.

## Machine Spirit #404B / CITADEL-EPOCH identity/dependence handoff Identity/source reconciliation must participate in promotion-evidence independence and contamination checks. New IDs must not manufacture new validation origins. Aliases, mirrors, same-event descriptions, semantic paraphrases/derivatives, parent/child candidates and repeated views of one underlying episode may remain distinct records while belonging to the same evidence-dependence/origin component where warranted. Conversely, equality/alias hypotheses are not enough to collapse genuinely independent sources without evidence. Origin-group decisions remain reversible/provenance-bearing and may themselves be uncertain. Required consequence for #404: `record independence != source independence != causal independence`; evaluation splitting/grouping must consume active identity/dependence knowledge so near-duplicate or renamed records cannot bypass holdout/contamination policy. Refs prolog-rlm#404B CITADEL fixtures B2, B5-B8, B30.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

Machine Spirit #404C handoff — identity mapping is not evidence independence

PACT-EPOCH transfer assessments may depend on #10 ontology/entity mappings, but keep these concepts separate:

same entity/proposition identity
    != same evidence origin
    != independent validation

Transfer receipts should reference exact mapping/link versions and classify mapping loss/ambiguity. A reversible same-as/ontology link can enable semantic comparison without proving two evaluation datasets, fixtures or authorities are independent. Conversely, origin/dependence claims may use opaque privacy-preserving classes without globally merging semantic identities. Mapping change/retraction must invalidate only dependent transfer assessments. Full design: prolog-rlm#404C.

## Machine Spirit #404C handoff — identity mapping is not evidence independence PACT-EPOCH transfer assessments may depend on #10 ontology/entity mappings, but keep these concepts separate: ```text same entity/proposition identity != same evidence origin != independent validation ``` Transfer receipts should reference exact mapping/link versions and classify mapping loss/ambiguity. A reversible `same-as`/ontology link can enable semantic comparison without proving two evaluation datasets, fixtures or authorities are independent. Conversely, origin/dependence claims may use opaque privacy-preserving classes without globally merging semantic identities. Mapping change/retraction must invalidate only dependent transfer assessments. Full design: prolog-rlm#404C.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

Implementation decomposition

#10 remains the canonical cross-memory identity/ontology requirement. The focused implementation slice is #27 under world-model epic #14.

Federation later consumes the same reversible identity/source-dependence model in #40. Keep identity reconciliation separate from source corroboration and from host principal/authority identity; optional vector/model matches remain candidate evidence, never irreversible merges.

## Implementation decomposition #10 remains the canonical cross-memory identity/ontology requirement. The focused implementation slice is **#27** under world-model epic #14. Federation later consumes the same reversible identity/source-dependence model in #40. Keep identity reconciliation separate from source corroboration and from host principal/authority identity; optional vector/model matches remain candidate evidence, never irreversible merges.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

MACHINE-SPIRIT #404D identity/dependence refinement

Identity reconciliation and learning-evidence dependence must remain separate versioned relations. CLOCKWORK explicitly rejects irreversible canonical union-find/DSU merging for source-origin/dependence truth because mappings can be corrected, retracted or superseded.

Current dependence components may be rebuildable accelerators, but canonical history is direct provenance-bearing links/events. Changing an identity/origin mapping increments the relevant dependency generation and invalidates only learned candidates/transfers whose exposure or validation-independence conclusions depend on it.

A missing/private cross-domain origin mapping yields independence=unknown, never independent. Replicas/mirrors remain dependent origins even when identity reconciliation joins their semantic objects. Refs prolog-rlm#404D and #27/#35.

## MACHINE-SPIRIT #404D identity/dependence refinement Identity reconciliation and learning-evidence dependence must remain separate versioned relations. CLOCKWORK explicitly rejects irreversible canonical union-find/DSU merging for source-origin/dependence truth because mappings can be corrected, retracted or superseded. Current dependence components may be rebuildable accelerators, but canonical history is direct provenance-bearing links/events. Changing an identity/origin mapping increments the relevant dependency generation and invalidates only learned candidates/transfers whose exposure or validation-independence conclusions depend on it. A missing/private cross-domain origin mapping yields `independence=unknown`, never independent. Replicas/mirrors remain dependent origins even when identity reconciliation joins their semantic objects. Refs prolog-rlm#404D and #27/#35.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

prolog-rlm#405B AEGIS-MS identity handoff: semantic reconciliation is never host authentication/authorization. same_as, aliases, roles, ontology links and model-suggested identity hypotheses may support semantic joins, but cannot bind an authenticated host principal or privileged resource. Semantic→host binding must be a separate trusted runtime adapter with explicit generation/provenance and dispatch-time revalidation; changed/retracted identity mappings invalidate dependent semantic results without silently mutating host authority.

prolog-rlm#405B AEGIS-MS identity handoff: **semantic reconciliation is never host authentication/authorization**. `same_as`, aliases, roles, ontology links and model-suggested identity hypotheses may support semantic joins, but cannot bind an authenticated host principal or privileged resource. Semantic→host binding must be a separate trusted runtime adapter with explicit generation/provenance and dispatch-time revalidation; changed/retracted identity mappings invalidate dependent semantic results without silently mutating host authority.

nsaspy commented

2026-09-08 01:57:36 +00:00

Author

Owner

Machine Spirit #405C / CONCORDAT identity handoff

Cross-memory identity reconciliation is now explicitly domain-qualified and directional.

same surface name != same entity
semantic same_as != authenticated principal equality
mapping(A,B) != mapping(B,A)
foreign identity hypothesis != host resource/principal binding

Persist original IDs plus mapping relation/evidence/source+target domains/version/status; never destructively canonicalize away the source-domain identity. exact|narrower|broader|overlap|likely_same|different remain materially distinct and bridge chains accumulate loss/uncertainty rather than silently strengthening to equality.

Conformance should include same-name principals across two trust domains and a mapping-cycle fixture that cannot bootstrap authority/corroboration.

Canonical research: lost-rob0t/prolog-rlm#405C.

## Machine Spirit #405C / CONCORDAT identity handoff Cross-memory identity reconciliation is now explicitly **domain-qualified and directional**. ```text same surface name != same entity semantic same_as != authenticated principal equality mapping(A,B) != mapping(B,A) foreign identity hypothesis != host resource/principal binding ``` Persist original IDs plus mapping relation/evidence/source+target domains/version/status; never destructively canonicalize away the source-domain identity. `exact|narrower|broader|overlap|likely_same|different` remain materially distinct and bridge chains accumulate loss/uncertainty rather than silently strengthening to equality. Conformance should include same-name principals across two trust domains and a mapping-cycle fixture that cannot bootstrap authority/corroboration. Canonical research: lost-rob0t/prolog-rlm#405C.

nsaspy referenced this issue

2026-09-08 01:57:39 +00:00

[EPIC][CORE] Library-first Symbolic Memory v1: durable semantic world model, learning history, federation, and zero-model recall #38

Rows
Columns

[semantic-memory] Cross-memory entity/concept reconciliation, aliasing, ontology alignment, and reversible identity links #39

Goal

Required relation classes

Reversible reconciliation

Evidence for reconciliation

Ontology alignment

Events and propositions

Scope / time

Query integration

Acceptance

Machine Spirit #401C handoff — reversible procedure/action/goal alignment

Machine Spirit #402A identity/ontology durability handoff

MACHINE-SPIRIT #402B handoff — identity/dedup privacy boundary

MACHINE-SPIRIT #402C handoff — cross-ledger identity/dedup/corroboration

Upstream Machine Spirit #403B handoff — identity attention is candidate-only until reconciled

MACHINE-SPIRIT prolog-rlm#403C handoff — identity/source-dependence for retrieval federation

Machine Spirit #404B / CITADEL-EPOCH identity/dependence handoff

Machine Spirit #404C handoff — identity mapping is not evidence independence

Implementation decomposition

MACHINE-SPIRIT #404D identity/dependence refinement

Machine Spirit #405C / CONCORDAT identity handoff