[semantic-memory] General symbolic recall/query over facts, rules, events, time, causality, procedures, constraints, claims, and provenance

nsaspy commented

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

Owner

Parent: #4
Depends on: #6
Integrates with: prolog-rlm#392-#394

Goal

Expose the remembered general semantic world model as a queryable reasoning surface so callers can ask arbitrary knowledge questions without rereading original prose.

The query layer must not be specialized to how do I do X?. It should reason over every semantic class persisted by #6.

Query classes

Support at least:

Entity / definition / taxonomy

What is X?
What type is X?
What is X part of?
What aliases refer to X?
How are X and Y related?

Rule / implication

What follows if conditions A and B hold?
What rules support conclusion C?
What exceptions defeat rule R?

Procedure / planning

How do I do X?
What must happen before Y?
What is the next step after failure Z?
What branch applies under conditions A/B?

Event / state / temporal

What happened before E?
What changed after E?
What was true at time T?
What is currently true?
Which events overlap interval I?

Causal / diagnostic

What may cause symptom S?
What evidence discriminates between hypotheses H1/H2?
What recovery addresses failure F?

Constraint / quantitative

What values satisfy remembered constraints?
How many resources are required?
Which options violate limit L?

Use upstream safe reasoning/constraint semantics rather than arbitrary evaluation.

Normative / policy knowledge

According to remembered source/policy P, what is required/permitted/forbidden?
What exception applies?

This reports remembered policy knowledge only; it does not grant host authority.

Claims / beliefs / conflicting sources

Who claims P?
What sources support or deny P?
Is P asserted, uncertain, conflicted, or unknown?

Attributed claims must remain attributed.

Goals / plans / preferences

What goal did actor A express?
Which option was preferred and under what criteria?

Hypothesis / scenario

Under scenario S, what follows?
What evidence supports hypothesis H?

Scenario facts cannot leak into ordinary world queries.

General API direction

Provide a general query substrate equivalent to:

memory_semantic_query(+QueryIR, +Options, -Outcome).
memory_semantic_ask(+Goal, +Context, +Options, -Outcome).
memory_semantic_projection(+Goal, +Context, +Options, -Outcome).
memory_semantic_explain(+ResultRef, -Outcome).
memory_source_expand(+RecordRef, +Options, -Outcome).

Specialized convenience predicates may sit on top but must reuse one general semantic reasoning layer.

MACHINE-SPIRIT #400A epistemic query contract

Depth 3A requires the query surface to separate evidence support from warrant/acceptance under a chosen epistemic profile.

Do not reduce query status to one mutable truth enum or confidence score.

Conceptual API refinement:

memory_semantic_query(+QueryIR,
                      +Context,
                      +EpistemicProfile,
                      +Options,
                      -Outcome).

memory_epistemic_support(+Proposition,
                         +Context,
                         +Options,
                         -SupportState).

memory_semantic_counterevidence(+ResultRef, -CounterEvidence).

A result should expose at least:

bindings/conclusion
support_state: neither | supported_only | refuted_only | both
acceptance: warranted | unwarranted | undecided | credulous_only | profile_inconsistent | resource_limited
explicit epistemic profile + version
supporting justification refs
counterevidence/refuting justification refs
assumption/environment refs
validity/time/context
unresolved conditions
source/compiler/logical/profile provenance

Candidate profile family from #400A:

support_paraconsistent: report positive/negative support without explosion or automatic winner;
well_founded_default: candidate deterministic default profile for admitted bounded tabled logic-program fragments;
stable_models: optional multi-model profile exposing skeptical vs credulous conclusions;
argumentation: derived argument/attack/defeat view for contested/default queries;
revision_projection: AGM-inspired coherent operational belief projection over explicitly selected evidence/trust policy.

Profile selection is part of the query contract. all memories does not mean one implicit global classical theory.

Cross-memory joins must explicitly state evidence admission/context/trust policy. A source claim, hypothetical proposition or historical fact cannot become an ordinary current-world premise merely because entity IDs join.

Hard query distinctions:

withdraw(P-support) != support(not(P))
defeated_default(P) != explicit_false(P)
temporal_change != contradiction
conflict != explosion
undefined/default-cycle != ordinary false
no stable model != ordinary false
credulous support != skeptical warrant
warranted semantic recommendation != host authority

For model/default profiles with multiple admissible models/extensions, expose plurality rather than choosing the first solver result.

For resource-limited environment/model search, return an explicit bounded undecided/resource-limited outcome with receipt.

Result model

A result should preserve:

answer bindings/conclusion;
independent support/refutation state;
profile-relative acceptance;
supporting fact/rule/event/procedure/constraint refs;
source memory/span refs;
assumptions/preconditions;
defaults/exceptions applied;
temporal/scenario/speaker scope;
unresolved conditions;
conflicts/counterevidence;
trust/provenance summary;
derived-vs-observed classification;
epistemic profile/version and replay receipt.

Reasoning

Use Prolog inference over upstream safe semantic reasoning semantics. Support chaining across multiple memories when namespace/trust policy permits it.

Required properties:

open-world by default;
explicit negative knowledge distinct from missing knowledge;
conflict-preserving/non-explosive base support;
temporal validity-aware;
scenario/hypothesis scope-aware;
attribution-aware;
default/exception-aware;
safe constraint solving where supported;
provenance-carrying explanations;
profile-relative skeptical/credulous/undefined distinctions where relevant.

Bounded projection

Do not return the whole semantic store to downstream models.

Given a goal, select only relevant semantic records, epistemic support/counterevidence, assumptions, rules/defaults and compact provenance refs. A caller can explicitly expand original source when needed.

Acceptance

how-to query derives ordered procedure with correct branch after restart.
definition/taxonomy query works with no model call.
temporal query distinguishes historical from current truth.
causal query returns possible causes and supporting provenance.
safe quantitative fixture is solved through upstream constraint semantics.
default+exception query returns the exception-adjusted conclusion and explanation.
attributed conflicting claims return support on both sides rather than arbitrary winner.
explicit false differs from unknown.
hypothetical/scenario knowledge does not leak to current-world query.
query chains compatible knowledge from multiple memories.
bounded projection excludes unrelated semantic records.
every derived conclusion explains supporting rule/source path.
no LLM/provider is required for recall/reasoning once projections exist.
optional conversational renderer can consume the bounded structured result without owning inference/truth.
withdrawal does not appear as explicit negative evidence.
multiple stable/default models expose skeptical/credulous plurality.
a WFS negative cycle can return undefined with residual explanation.
query results expose counterevidence and profile/version.
resource ceiling never becomes arbitrary top-1 truth.

Refs #4 #6 and prolog-rlm#392 #394 #397 #400.

Parent: #4 Depends on: #6 Integrates with: prolog-rlm#392-#394 ## Goal Expose the remembered **general semantic world model** as a queryable reasoning surface so callers can ask arbitrary knowledge questions without rereading original prose. The query layer must not be specialized to `how do I do X?`. It should reason over every semantic class persisted by #6. ## Query classes Support at least: ### Entity / definition / taxonomy ```text What is X? What type is X? What is X part of? What aliases refer to X? How are X and Y related? ``` ### Rule / implication ```text What follows if conditions A and B hold? What rules support conclusion C? What exceptions defeat rule R? ``` ### Procedure / planning ```text How do I do X? What must happen before Y? What is the next step after failure Z? What branch applies under conditions A/B? ``` ### Event / state / temporal ```text What happened before E? What changed after E? What was true at time T? What is currently true? Which events overlap interval I? ``` ### Causal / diagnostic ```text What may cause symptom S? What evidence discriminates between hypotheses H1/H2? What recovery addresses failure F? ``` ### Constraint / quantitative ```text What values satisfy remembered constraints? How many resources are required? Which options violate limit L? ``` Use upstream safe reasoning/constraint semantics rather than arbitrary evaluation. ### Normative / policy knowledge ```text According to remembered source/policy P, what is required/permitted/forbidden? What exception applies? ``` This reports remembered policy knowledge only; it does not grant host authority. ### Claims / beliefs / conflicting sources ```text Who claims P? What sources support or deny P? Is P asserted, uncertain, conflicted, or unknown? ``` Attributed claims must remain attributed. ### Goals / plans / preferences ```text What goal did actor A express? Which option was preferred and under what criteria? ``` ### Hypothesis / scenario ```text Under scenario S, what follows? What evidence supports hypothesis H? ``` Scenario facts cannot leak into ordinary world queries. ## General API direction Provide a general query substrate equivalent to: ```prolog memory_semantic_query(+QueryIR, +Options, -Outcome). memory_semantic_ask(+Goal, +Context, +Options, -Outcome). memory_semantic_projection(+Goal, +Context, +Options, -Outcome). memory_semantic_explain(+ResultRef, -Outcome). memory_source_expand(+RecordRef, +Options, -Outcome). ``` Specialized convenience predicates may sit on top but must reuse one general semantic reasoning layer. ## MACHINE-SPIRIT #400A epistemic query contract Depth 3A requires the query surface to separate **evidence support** from **warrant/acceptance under a chosen epistemic profile**. Do not reduce query status to one mutable truth enum or confidence score. Conceptual API refinement: ```prolog memory_semantic_query(+QueryIR, +Context, +EpistemicProfile, +Options, -Outcome). memory_epistemic_support(+Proposition, +Context, +Options, -SupportState). memory_semantic_counterevidence(+ResultRef, -CounterEvidence). ``` A result should expose at least: ```text bindings/conclusion support_state: neither | supported_only | refuted_only | both acceptance: warranted | unwarranted | undecided | credulous_only | profile_inconsistent | resource_limited explicit epistemic profile + version supporting justification refs counterevidence/refuting justification refs assumption/environment refs validity/time/context unresolved conditions source/compiler/logical/profile provenance ``` Candidate profile family from #400A: - `support_paraconsistent`: report positive/negative support without explosion or automatic winner; - `well_founded_default`: candidate deterministic default profile for admitted bounded tabled logic-program fragments; - `stable_models`: optional multi-model profile exposing skeptical vs credulous conclusions; - `argumentation`: derived argument/attack/defeat view for contested/default queries; - `revision_projection`: AGM-inspired coherent operational belief projection over explicitly selected evidence/trust policy. Profile selection is part of the query contract. `all memories` does not mean one implicit global classical theory. Cross-memory joins must explicitly state evidence admission/context/trust policy. A source claim, hypothetical proposition or historical fact cannot become an ordinary current-world premise merely because entity IDs join. Hard query distinctions: ```text withdraw(P-support) != support(not(P)) defeated_default(P) != explicit_false(P) temporal_change != contradiction conflict != explosion undefined/default-cycle != ordinary false no stable model != ordinary false credulous support != skeptical warrant warranted semantic recommendation != host authority ``` For model/default profiles with multiple admissible models/extensions, expose plurality rather than choosing the first solver result. For resource-limited environment/model search, return an explicit bounded undecided/resource-limited outcome with receipt. ## Result model A result should preserve: - answer bindings/conclusion; - independent support/refutation state; - profile-relative acceptance; - supporting fact/rule/event/procedure/constraint refs; - source memory/span refs; - assumptions/preconditions; - defaults/exceptions applied; - temporal/scenario/speaker scope; - unresolved conditions; - conflicts/counterevidence; - trust/provenance summary; - derived-vs-observed classification; - epistemic profile/version and replay receipt. ## Reasoning Use Prolog inference over upstream safe semantic reasoning semantics. Support chaining across multiple memories when namespace/trust policy permits it. Required properties: - open-world by default; - explicit negative knowledge distinct from missing knowledge; - conflict-preserving/non-explosive base support; - temporal validity-aware; - scenario/hypothesis scope-aware; - attribution-aware; - default/exception-aware; - safe constraint solving where supported; - provenance-carrying explanations; - profile-relative skeptical/credulous/undefined distinctions where relevant. ## Bounded projection Do not return the whole semantic store to downstream models. Given a goal, select only relevant semantic records, epistemic support/counterevidence, assumptions, rules/defaults and compact provenance refs. A caller can explicitly expand original source when needed. ## Acceptance - [ ] `how-to` query derives ordered procedure with correct branch after restart. - [ ] definition/taxonomy query works with no model call. - [ ] temporal query distinguishes historical from current truth. - [ ] causal query returns possible causes and supporting provenance. - [ ] safe quantitative fixture is solved through upstream constraint semantics. - [ ] default+exception query returns the exception-adjusted conclusion and explanation. - [ ] attributed conflicting claims return support on both sides rather than arbitrary winner. - [ ] explicit false differs from unknown. - [ ] hypothetical/scenario knowledge does not leak to current-world query. - [ ] query chains compatible knowledge from multiple memories. - [ ] bounded projection excludes unrelated semantic records. - [ ] every derived conclusion explains supporting rule/source path. - [ ] no LLM/provider is required for recall/reasoning once projections exist. - [ ] optional conversational renderer can consume the bounded structured result without owning inference/truth. - [ ] withdrawal does not appear as explicit negative evidence. - [ ] multiple stable/default models expose skeptical/credulous plurality. - [ ] a WFS negative cycle can return undefined with residual explanation. - [ ] query results expose counterevidence and profile/version. - [ ] resource ceiling never becomes arbitrary top-1 truth. Refs #4 #6 and prolog-rlm#392 #394 #397 #400.

nsaspy commented

2026-09-08 01:57:23 +00:00

Author

Owner

MACHINE-SPIRIT #400B query handoff — defeat paths, independence, known-at vs valid-at

Depth 3B strengthens the epistemic query surface.

A nontrivial result should expose two orthogonal status axes:

semantic_acceptance:
  skeptically_warranted | credulously_warranted | rejected |
  undefined | undecided | profile_inconsistent

evaluation_completeness:
  complete | bounded_partial | resource_limited |
  solver_error | unsupported_query_class

Do not use resource_limited as another word for unknown, and do not turn WFS undefined or no-stable-model into ordinary false.

Explanations/results should additionally expose, where material:

rebut / undermine / undercut / invalidate paths;
typed preference basis + named composition policy/version;
raw support episodes vs independent evidence-origin/dependence components;
circular/self-support conditions and external grounding;
valid-world-time separately from transaction/knowledge-time;
revision operator/state refs for revision-projection queries.

Add an explicit distinction equivalent to:

memory_semantic_ask(..., valid_at(WorldTime), ...).
memory_semantic_ask(..., known_at(TransactionTime), ...).

For a correction learned in February about what was actually true in January, valid_at(January) under today's evidence may differ from known_at(January).

Profile changes may legitimately alter reinstatement/default/cycle warrant, but the canonical ESG + typed defeat/dependence history must remain identical and the profile/version must explain the disagreement.

Full design and 24 adversarial fixtures are preserved on prolog-rlm#400.

## MACHINE-SPIRIT #400B query handoff — defeat paths, independence, known-at vs valid-at Depth 3B strengthens the epistemic query surface. A nontrivial result should expose two orthogonal status axes: ```text semantic_acceptance: skeptically_warranted | credulously_warranted | rejected | undefined | undecided | profile_inconsistent evaluation_completeness: complete | bounded_partial | resource_limited | solver_error | unsupported_query_class ``` Do not use `resource_limited` as another word for `unknown`, and do not turn WFS `undefined` or no-stable-model into ordinary false. Explanations/results should additionally expose, where material: - rebut / undermine / undercut / invalidate paths; - typed preference basis + named composition policy/version; - raw support episodes vs independent evidence-origin/dependence components; - circular/self-support conditions and external grounding; - valid-world-time separately from transaction/knowledge-time; - revision operator/state refs for revision-projection queries. Add an explicit distinction equivalent to: ```prolog memory_semantic_ask(..., valid_at(WorldTime), ...). memory_semantic_ask(..., known_at(TransactionTime), ...). ``` For a correction learned in February about what was actually true in January, `valid_at(January)` under today's evidence may differ from `known_at(January)`. Profile changes may legitimately alter reinstatement/default/cycle warrant, but the canonical ESG + typed defeat/dependence history must remain identical and the profile/version must explain the disagreement. Full design and 24 adversarial fixtures are preserved on prolog-rlm#400.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

MACHINE-SPIRIT #400C query handoff — explicit epistemic federation

Cross-memory/cross-theory joins must no longer mean union all matching propositions and run one implicit reasoner.

Add query semantics equivalent to:

epistemic_federated_query(+Query,
                           +TheoryRefs,
                           +BridgePolicy,
                           +FederationProfile,
                           +Context,
                           +Options,
                           -Outcome).

Required explicit inputs include participating theories, bridge/admission policy, federation profile/version, context, valid_at and/or known_at where relevant.

Result must preserve:

local outcome vector (each local profile/version + support/acceptance/completeness);
bridge closure and bridge versions;
imported support/counterevidence and import mode;
skeptical/credulous/model plurality;
identity/ontology mapping assumptions;
source-dependence components across memories;
cross-context rebut/undermine/undercut/invalidate edges where applicable;
noncomposable/lossy bridge diagnostics;
semantic acceptance separate from evaluation completeness;
complete federation receipt/provenance.

Hard defaults:

all memories != one global classical/nonmonotonic theory
local NAF absence != explicit negative support in another memory
credulous local result != ordinary cross-memory warrant
priority namespace A != priority namespace B
failed/no-model context != failure of unrelated context
semantic federation != host authority federation

Recommended federation modes: federated_evidence as safest baseline, federated_skeptical, federated_credulous, federated_argumentation, optional bounded mcs_equilibrium, and managed/revision federation where explicitly requested.

Cross-memory query planning should be demand-driven. A query over one unaffected deterministic theory must not evaluate every unrelated ASP/default context merely because they share the same store.

## MACHINE-SPIRIT #400C query handoff — explicit epistemic federation Cross-memory/cross-theory joins must no longer mean `union all matching propositions and run one implicit reasoner`. Add query semantics equivalent to: ```prolog epistemic_federated_query(+Query, +TheoryRefs, +BridgePolicy, +FederationProfile, +Context, +Options, -Outcome). ``` Required explicit inputs include participating theories, bridge/admission policy, federation profile/version, context, `valid_at` and/or `known_at` where relevant. Result must preserve: - local outcome vector (each local profile/version + support/acceptance/completeness); - bridge closure and bridge versions; - imported support/counterevidence and import mode; - skeptical/credulous/model plurality; - identity/ontology mapping assumptions; - source-dependence components across memories; - cross-context rebut/undermine/undercut/invalidate edges where applicable; - noncomposable/lossy bridge diagnostics; - semantic acceptance separate from evaluation completeness; - complete federation receipt/provenance. Hard defaults: ```text all memories != one global classical/nonmonotonic theory local NAF absence != explicit negative support in another memory credulous local result != ordinary cross-memory warrant priority namespace A != priority namespace B failed/no-model context != failure of unrelated context semantic federation != host authority federation ``` Recommended federation modes: `federated_evidence` as safest baseline, `federated_skeptical`, `federated_credulous`, `federated_argumentation`, optional bounded `mcs_equilibrium`, and managed/revision federation where explicitly requested. Cross-memory query planning should be demand-driven. A query over one unaffected deterministic theory must not evaluate every unrelated ASP/default context merely because they share the same store.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

MACHINE-SPIRIT #400D handoff — federated query planning / projected-interface execution

Depth 3D refines memory_semantic_query/… into a demand-driven federated execution contract.

A query should compile to an inspectable Epistemic Query Plan that records:

selected theories and why they are relevant;
selected local/federation profile versions;
bridge dependency graph + SCC condensation;
local projected interface signatures;
evaluation strategy per SCC;
identity/dependence/closure assumptions;
hard resource ceilings;
exact evidence/bridge/profile frontier fingerprints;
explanation/replay policy.

Do not evaluate every memory/theory just because it is queryable. Do not enumerate full local model sets when multiple models have the same bridge-visible epistemic exports.

Federation execution should use:

acyclic relevant region -> topological delta propagation
monotone cyclic region  -> least fixed point / semi-naive
WFS region              -> tabled/alternating fixed point
ASP/equilibrium region  -> bounded projected-interface solver search
argumentation region    -> selected semantics, with bounded search where needed
noncomposable region    -> explicit partial/noncomposable diagnostic

Result must continue to split semantic acceptance from evaluation completeness, including noncomposable, no_equilibrium, resource_limited, backend_unavailable, and stale where applicable.

Explanations should be compressed receipt DAGs linking query plan -> local interface summaries -> bridge/SCC receipts -> local justifications -> source assertions, with expandable counterevidence rather than duplicated proof trees.

Required conformance now includes:

irrelevant-theory pruning;
projected-interface equivalence against bounded full-model oracle;
cold/warm/rebuilt cache semantic equality;
incremental/full semantic equality;
concurrency-interleaving determinism;
source-dependence preservation through federation;
explicit identity/closure/profile invalidation;
resource exhaustion never becoming top-1 truth;
provider/LLM call count exactly zero for all compiled-semantic epistemic fixtures.

Full design/sources are preserved on prolog-rlm#400D.

## MACHINE-SPIRIT #400D handoff — federated query planning / projected-interface execution Depth 3D refines `memory_semantic_query/…` into a demand-driven federated execution contract. A query should compile to an inspectable **Epistemic Query Plan** that records: - selected theories and why they are relevant; - selected local/federation profile versions; - bridge dependency graph + SCC condensation; - local projected interface signatures; - evaluation strategy per SCC; - identity/dependence/closure assumptions; - hard resource ceilings; - exact evidence/bridge/profile frontier fingerprints; - explanation/replay policy. Do not evaluate every memory/theory just because it is queryable. Do not enumerate full local model sets when multiple models have the same bridge-visible epistemic exports. Federation execution should use: ```text acyclic relevant region -> topological delta propagation monotone cyclic region -> least fixed point / semi-naive WFS region -> tabled/alternating fixed point ASP/equilibrium region -> bounded projected-interface solver search argumentation region -> selected semantics, with bounded search where needed noncomposable region -> explicit partial/noncomposable diagnostic ``` Result must continue to split semantic acceptance from evaluation completeness, including `noncomposable`, `no_equilibrium`, `resource_limited`, `backend_unavailable`, and `stale` where applicable. Explanations should be compressed receipt DAGs linking query plan -> local interface summaries -> bridge/SCC receipts -> local justifications -> source assertions, with expandable counterevidence rather than duplicated proof trees. Required conformance now includes: - irrelevant-theory pruning; - projected-interface equivalence against bounded full-model oracle; - cold/warm/rebuilt cache semantic equality; - incremental/full semantic equality; - concurrency-interleaving determinism; - source-dependence preservation through federation; - explicit identity/closure/profile invalidation; - resource exhaustion never becoming top-1 truth; - provider/LLM call count exactly zero for all compiled-semantic epistemic fixtures. Full design/sources are preserved on prolog-rlm#400D.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Machine Spirit #401C handoff — procedural portability/query status

General semantic recall should expose procedural interoperability/admission state without implying execution authority. For procedure-oriented results, preserve distinctions equivalent to:

represented
semantic_contract_known
mapping_available
semantic_compatible(context)
execution_projectable(context)
host_bound
capability_available
authority_admissible_now
plan_valid_now

Return PSC/bridge/mapping versions, declared losses/approximations, source/projected guarantee, fairness assumptions, unbound actions/tasks and relevant provenance in explanation/projection receipts. A remembered procedure may be semantically useful/queryable while still non-projectable or locally blocked. source says this works remains different from trusted/admitted here.

This query status is descriptive symbolic knowledge only; memory recall cannot create host action bindings or authority. See prolog-rlm#401 Depth-4C PIFF/PSC design.

## Machine Spirit #401C handoff — procedural portability/query status General semantic recall should expose procedural interoperability/admission state without implying execution authority. For procedure-oriented results, preserve distinctions equivalent to: ```text represented semantic_contract_known mapping_available semantic_compatible(context) execution_projectable(context) host_bound capability_available authority_admissible_now plan_valid_now ``` Return PSC/bridge/mapping versions, declared losses/approximations, source/projected guarantee, fairness assumptions, unbound actions/tasks and relevant provenance in explanation/projection receipts. A remembered procedure may be semantically useful/queryable while still non-projectable or locally blocked. `source says this works` remains different from `trusted/admitted here`. This query status is descriptive symbolic knowledge only; memory recall cannot create host action bindings or authority. See prolog-rlm#401 Depth-4C PIFF/PSC design.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Upstream Machine Spirit #401D handoff — procedure query/admission status

Depth 4D adds a computational distinction that the general query surface should expose without conflating recall with execution:

represented
compatible(Profile/PSC)
projectable(HostProfile)
bound(CurrentBindingGeneration)
currently_admissible(Evidence/Resource/AuthorityFrontier)

A how do I do X? query may return a represented/compatible procedure even when no local executable binding exists. currently_admissible is live derived state and must expose the evidence/policy generations it depends on; it is not a durable truth bit.

Procedure results should also carry, where applicable:

native formalism + version;
PSC/bridge versions;
guarantee/fairness class;
declared projection loss/noncomposability;
required observation/guard classes;
host binding status;
resource/authority prerequisites as requirements, not granted authority;
conformance receipt/profile for executable projections;
bounded/resource-limited status rather than arbitrary top-1 procedure.

Hard distinction:

queryable procedure != executable permission

Known symbolic procedure recall/planning remains zero-model once semantics are present. Full DPEC design is on prolog-rlm#401.

## Upstream Machine Spirit #401D handoff — procedure query/admission status Depth 4D adds a computational distinction that the general query surface should expose without conflating recall with execution: ```text represented compatible(Profile/PSC) projectable(HostProfile) bound(CurrentBindingGeneration) currently_admissible(Evidence/Resource/AuthorityFrontier) ``` A `how do I do X?` query may return a represented/compatible procedure even when no local executable binding exists. `currently_admissible` is live derived state and must expose the evidence/policy generations it depends on; it is not a durable truth bit. Procedure results should also carry, where applicable: - native formalism + version; - PSC/bridge versions; - guarantee/fairness class; - declared projection loss/noncomposability; - required observation/guard classes; - host binding status; - resource/authority prerequisites as **requirements**, not granted authority; - conformance receipt/profile for executable projections; - bounded/resource-limited status rather than arbitrary top-1 procedure. Hard distinction: ```text queryable procedure != executable permission ``` Known symbolic procedure recall/planning remains zero-model once semantics are present. Full DPEC design is on prolog-rlm#401.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

MACHINE-SPIRIT #402B handoff — historical commitment vs retrievability

General query results must now separate:

historically_committed
currently_admissible / currently_warranted
payload_retrievable
replay_capability

A valid historical answer after authorized erasure may be committed at frontier F; payload redacted; replay=commitment_only. It must not recover erased source text from a stale projection. Likewise quarantined, integrity_failure, and incompatible-schema states are not false or ordinary unknown.

Action/VERIFY-sensitive queries must fail closed on stale/integrity-failed projections and exact source expansion remains authority-gated. See prolog-rlm#402B.

## MACHINE-SPIRIT #402B handoff — historical commitment vs retrievability General query results must now separate: ```text historically_committed currently_admissible / currently_warranted payload_retrievable replay_capability ``` A valid historical answer after authorized erasure may be `committed at frontier F; payload redacted; replay=commitment_only`. It must not recover erased source text from a stale projection. Likewise `quarantined`, `integrity_failure`, and incompatible-schema states are not `false` or ordinary `unknown`. Action/VERIFY-sensitive queries must fail closed on stale/integrity-failed projections and exact source expansion remains authority-gated. See prolog-rlm#402B.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

MACHINE-SPIRIT #402C handoff — federated query/cut semantics

General query must not interpret all memories/all ledgers as one implicit latest snapshot. Federated queries select or construct an explicit federation cut over named logical ledgers and expose its completeness/integrity state.

Conceptual refinement:

memory_semantic_query(Query, Context, EpistemicProfile,
                      [federation_cut(Cut)|Options], Outcome).

memory_federation_cut(Federation, CutSpec, Cut).

Result/receipt should preserve:

exact per-ledger frontier/checkpoint;
federation-link + mapping versions;
local/remote admission and integrity status;
causal predecessors used;
replica/source-dependence classification;
per-ledger payload/replay/redaction state;
completeness: e.g. causally_closed, partial_remote_unavailable, incomplete_dependency, integrity_blocked.

Missing/offline remote knowledge is not false. Deterministic presentation order is not semantic causal order. A valid remote transparency receipt is evidence of registration/inclusion, not epistemic warrant; #400's epistemic profiles still decide warrant above the storage federation layer.

Exact same cut/frontiers/versions must be replay-addressable subject to GRIE replay capability (exact|semantic_only|commitment_only|redacted|...). Full C design and fixtures are on prolog-rlm#402.

### MACHINE-SPIRIT #402C handoff — federated query/cut semantics General query must not interpret `all memories/all ledgers` as one implicit latest snapshot. Federated queries select or construct an explicit **federation cut** over named logical ledgers and expose its completeness/integrity state. Conceptual refinement: ```prolog memory_semantic_query(Query, Context, EpistemicProfile, [federation_cut(Cut)|Options], Outcome). memory_federation_cut(Federation, CutSpec, Cut). ``` Result/receipt should preserve: - exact per-ledger frontier/checkpoint; - federation-link + mapping versions; - local/remote admission and integrity status; - causal predecessors used; - replica/source-dependence classification; - per-ledger payload/replay/redaction state; - completeness: e.g. `causally_closed`, `partial_remote_unavailable`, `incomplete_dependency`, `integrity_blocked`. Missing/offline remote knowledge is not `false`. Deterministic presentation order is not semantic causal order. A valid remote transparency receipt is evidence of registration/inclusion, not epistemic warrant; #400's epistemic profiles still decide warrant above the storage federation layer. Exact same cut/frontiers/versions must be replay-addressable subject to GRIE replay capability (`exact|semantic_only|commitment_only|redacted|...`). Full C design and fixtures are on prolog-rlm#402.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Machine Spirit #402D handoff: semantic query/history/current/explain surfaces must carry the exact requested cut, materialized cut, projection dependency generations, freshness/completeness and replay capability. lagging, stale_dependency, rebuilding, integrity_blocked, redaction_blocked, resource_limited, and partial federation cuts must not collapse into unknown or current. Exact-cut callers may catch up or demand-evaluate, but serving an older projection silently is forbidden. See prolog-rlm#402 SAFIRE D16-D21/D36.

Machine Spirit #402D handoff: semantic query/history/current/explain surfaces must carry the exact requested cut, materialized cut, projection dependency generations, freshness/completeness and replay capability. `lagging`, `stale_dependency`, `rebuilding`, `integrity_blocked`, `redaction_blocked`, `resource_limited`, and partial federation cuts must not collapse into `unknown` or `current`. Exact-cut callers may catch up or demand-evaluate, but serving an older projection silently is forbidden. See prolog-rlm#402 SAFIRE D16-D21/D36.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Machine Spirit prolog-rlm#403A handoff — TAPS bounded projection semantics

Bounded projection now needs two independent completeness dimensions rather than one implicit complete bit:

support_closure = closed | partial | resource_limited | stale
search_coverage = exhaustive_for_declared_fragment | bounded_structural | heuristic | partial_backend | unknown

A returned result can be support-closed (all admitted rules/premises/assumptions/counterevidence needed for that returned derivation are present) while search coverage remains heuristic (other answers may have been missed by top-k/activation/vector retrieval).

Recommended world-model projection flow over an exact #402/SAFIRE cut:

hard namespace/read-authority + time/scenario/epistemic-profile/freshness admission;
exact/structural seeds first;
lexical/case/graph activation for gaps;
vector/learned/model candidate generation only when needed;
mandatory semantic closure over rules/premises/guards/justifications/counterevidence/identity/profile assumptions/provenance refs;
soft budgeted periphery with relevance/novelty/source-dependence/recency/cost;
explicit frontier/gaps + replay receipt.

Hard distinctions:

retrieval score != warrant/truth
ranking != trust
similarity != identity
omitted != false

Working-memory residency is derived transient state; eviction never retracts ledger records and stable refs permit re-entry. Source expansion remains explicit and read-authorized.

Refs prolog-rlm#403 #381 #396 #400 #402.

## Machine Spirit prolog-rlm#403A handoff — TAPS bounded projection semantics Bounded projection now needs two independent completeness dimensions rather than one implicit `complete` bit: ```text support_closure = closed | partial | resource_limited | stale search_coverage = exhaustive_for_declared_fragment | bounded_structural | heuristic | partial_backend | unknown ``` A returned result can be support-closed (all admitted rules/premises/assumptions/counterevidence needed for that returned derivation are present) while search coverage remains heuristic (other answers may have been missed by top-k/activation/vector retrieval). Recommended world-model projection flow over an exact #402/SAFIRE cut: 1. hard namespace/read-authority + time/scenario/epistemic-profile/freshness admission; 2. exact/structural seeds first; 3. lexical/case/graph activation for gaps; 4. vector/learned/model candidate generation only when needed; 5. mandatory semantic closure over rules/premises/guards/justifications/counterevidence/identity/profile assumptions/provenance refs; 6. soft budgeted periphery with relevance/novelty/source-dependence/recency/cost; 7. explicit frontier/gaps + replay receipt. Hard distinctions: ```text retrieval score != warrant/truth ranking != trust similarity != identity omitted != false ``` Working-memory residency is derived transient state; eviction never retracts ledger records and stable refs permit re-entry. Source expansion remains explicit and read-authorized. Refs prolog-rlm#403 #381 #396 #400 #402.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Upstream Machine Spirit #403B handoff — OATH-TAPS projection/query semantics

prolog-rlm#403B is complete. B hardens bounded semantic projection against manipulation of candidate exposure.

memory_semantic_projection / query should preserve, where triggered by the attention/evidence profile:

support_closure
search_coverage
opposition_coverage
freshness coverage/gaps
identity-disambiguation coverage/gaps
source-independence/origin components
selection_integrity
sensor disagreement
exact world-model cut / receipt

Hard requirements:

top-k absence is never proof of absence;
current evidence requirements reject stale-only high-ranked candidates;
lexical/vector/graph hits are revalidated against the exact admitted durable cut before reasoning;
source mirrors/copies remain distinct records but do not manufacture independent corroboration;
retrieval may explicitly seek refutation, exceptions/defeaters, fresher state, identity counterexamples and independent origins;
similarity/rank/trust signals remain distinct from warrant and identity;
closure/resource exhaustion returns an explicit gap rather than a conclusion with silently truncated dependencies;
no universal poison_free=true result.

Refs lost-rob0t/prolog-rlm#403 #397 #400 #402 and this repo #6/#9/#10.

## Upstream Machine Spirit #403B handoff — OATH-TAPS projection/query semantics `prolog-rlm#403B` is complete. B hardens bounded semantic projection against manipulation of *candidate exposure*. `memory_semantic_projection` / query should preserve, where triggered by the attention/evidence profile: ```text support_closure search_coverage opposition_coverage freshness coverage/gaps identity-disambiguation coverage/gaps source-independence/origin components selection_integrity sensor disagreement exact world-model cut / receipt ``` Hard requirements: - top-k absence is never proof of absence; - current evidence requirements reject stale-only high-ranked candidates; - lexical/vector/graph hits are revalidated against the exact admitted durable cut before reasoning; - source mirrors/copies remain distinct records but do not manufacture independent corroboration; - retrieval may explicitly seek refutation, exceptions/defeaters, fresher state, identity counterexamples and independent origins; - similarity/rank/trust signals remain distinct from warrant and identity; - closure/resource exhaustion returns an explicit gap rather than a conclusion with silently truncated dependencies; - no universal `poison_free=true` result. Refs lost-rob0t/prolog-rlm#403 #397 #400 #402 and this repo #6/#9/#10.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

MACHINE-SPIRIT prolog-rlm#403C handoff — MOSAIC-TAPS

General semantic query/projection should preserve heterogeneous retriever semantics rather than flattening to one score/list. Add/retain fields sufficient for:

exact world-model cut/frontier or backend snapshot/generation;
backend-local ordering value (exact_match, rank, BM25, vector metric/model, graph activation, etc.);
QueryIR translation receipt: preserved/weakened/unsupported constraints and native-vs-post filters;
support/opposition/freshness/identity/source-independence coverage plus unsatisfied obligations;
external-vs-canonical admission state;
fusion/composition provenance; RRF/fusion score is never warrant;
scoped backend robustness/threat claims rather than one global robust=true.

Cross-memory/backend joins must use reversible #10 mappings and #6/#402 source-dependence lineage. An external retrieval result does not become a durable canonical memory record merely because it was returned.

Refs lost-rob0t/prolog-rlm#397 #403 #381 #396 and this repo #6 #9 #10.

### MACHINE-SPIRIT prolog-rlm#403C handoff — MOSAIC-TAPS General semantic query/projection should preserve heterogeneous retriever semantics rather than flattening to one score/list. Add/retain fields sufficient for: - exact world-model cut/frontier or backend snapshot/generation; - backend-local ordering value (`exact_match`, rank, BM25, vector metric/model, graph activation, etc.); - QueryIR translation receipt: preserved/weakened/unsupported constraints and native-vs-post filters; - support/opposition/freshness/identity/source-independence coverage plus unsatisfied obligations; - external-vs-canonical admission state; - fusion/composition provenance; RRF/fusion score is never warrant; - scoped backend robustness/threat claims rather than one global `robust=true`. Cross-memory/backend joins must use reversible #10 mappings and #6/#402 source-dependence lineage. An external retrieval result does not become a durable canonical memory record merely because it was returned. Refs lost-rob0t/prolog-rlm#397 #403 #381 #396 and this repo #6 #9 #10.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Machine Spirit #403D handoff — RACE-TAPS query/projection execution

Depth 6D closes the computational contract for bounded semantic recall.

memory_semantic_query / memory_semantic_projection should preserve a replayable retrieval receipt containing at least: exact SAFIRE cut, logical obligation plan, relevant RSC/translator/backend/index generations, backend observations/pages, physical-plan/revision refs where exposed, normalized candidate/fusion/closure receipts, support_closure, search_coverage, opposition_coverage, freshness, selection integrity, explicit gaps and budgets.

Hard requirements:

a cache hit never substitutes for cut/generation validation;
physical-plan/statistics choice cannot change epistemic warrant;
async backend completion order cannot change normalized output;
postfilter-only ANN cannot claim exhaustive filtered search;
snapshot/index generation drift during pagination is explicit stale/best-effort lineage;
remote timeout/cancel is a coverage gap, never semantic absence;
source copies found through multiple retrievers remain one dependence component;
retained remote responses may support exact historical replay; live requery is a new observation lineage;
byte/work limits cannot retain a derived conclusion while silently dropping mandatory dependencies;
already-known symbolic queries remain provider-free (model_calls = 0).

Physical durable indexes/cursors/generations and useful persistent caches belong in symbolic-memory; logical obligation/RSC semantics remain upstream in prolog-rlm. Full D design and fixtures are on prolog-rlm#403D.

## Machine Spirit #403D handoff — RACE-TAPS query/projection execution Depth 6D closes the computational contract for bounded semantic recall. `memory_semantic_query` / `memory_semantic_projection` should preserve a replayable retrieval receipt containing at least: exact SAFIRE cut, logical obligation plan, relevant RSC/translator/backend/index generations, backend observations/pages, physical-plan/revision refs where exposed, normalized candidate/fusion/closure receipts, `support_closure`, `search_coverage`, `opposition_coverage`, freshness, selection integrity, explicit gaps and budgets. Hard requirements: - a cache hit never substitutes for cut/generation validation; - physical-plan/statistics choice cannot change epistemic warrant; - async backend completion order cannot change normalized output; - postfilter-only ANN cannot claim exhaustive filtered search; - snapshot/index generation drift during pagination is explicit stale/best-effort lineage; - remote timeout/cancel is a coverage gap, never semantic absence; - source copies found through multiple retrievers remain one dependence component; - retained remote responses may support exact historical replay; live requery is a new observation lineage; - byte/work limits cannot retain a derived conclusion while silently dropping mandatory dependencies; - already-known symbolic queries remain provider-free (`model_calls = 0`). Physical durable indexes/cursors/generations and useful persistent caches belong in symbolic-memory; logical obligation/RSC semantics remain upstream in prolog-rlm. Full D design and fixtures are on prolog-rlm#403D.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

MACHINE-SPIRIT #404A handoff — learned-knowledge query/explanation

General semantic query should be able to expose learned knowledge without flattening lifecycle/evidence:

candidate | admitted | suspended | superseded | retracted
origin experience refs
learner/profile/hypothesis language
promotion policy/version + receipt
counterexamples/regressions
exact rule/method/control version

A query may use only currently admitted learned knowledge under the selected profile/policy; candidate/suspended material remains inspectable evidence but cannot silently participate as trusted control knowledge. Strategy-performance statistics are operational selection evidence, not proposition warrant. Historical replay must resolve the exact learned-object versions used then.

Ref upstream prolog-rlm#404A EPOCH.

## MACHINE-SPIRIT #404A handoff — learned-knowledge query/explanation General semantic query should be able to expose learned knowledge without flattening lifecycle/evidence: ```text candidate | admitted | suspended | superseded | retracted origin experience refs learner/profile/hypothesis language promotion policy/version + receipt counterexamples/regressions exact rule/method/control version ``` A query may use only currently admitted learned knowledge under the selected profile/policy; candidate/suspended material remains inspectable evidence but cannot silently participate as trusted control knowledge. Strategy-performance statistics are operational selection evidence, not proposition warrant. Historical replay must resolve the exact learned-object versions used then. Ref upstream prolog-rlm#404A EPOCH.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Machine Spirit #404B / CITADEL-EPOCH query/explanation handoff

Queries over learned rules/methods/strategies must expose more than learned=true or one confidence number. A learned-object result should be able to report:

candidate/admitted/held/quarantined/suspended/superseded status + version;
applicability context/validity epoch;
generating experiences;
independent validation vs contaminated/selection evidence;
source/origin-dependence groups;
baseline/evaluation profile;
counterexamples and bounded-search coverage;
evaluator/metric/promotion-policy versions;
drift and calibration state;
limitations/assumptions in the promotion receipt.

Hard query distinction: admitted learned knowledge is an operational/semantic status, not source-explicit truth, epistemic warrant, capability or authority. Historical queries must recover the exact promotion/evaluation state that applied at the requested knowledge frontier. Refs prolog-rlm#404B.

## Machine Spirit #404B / CITADEL-EPOCH query/explanation handoff Queries over learned rules/methods/strategies must expose more than `learned=true` or one confidence number. A learned-object result should be able to report: - candidate/admitted/held/quarantined/suspended/superseded status + version; - applicability context/validity epoch; - generating experiences; - independent validation vs contaminated/selection evidence; - source/origin-dependence groups; - baseline/evaluation profile; - counterexamples and bounded-search coverage; - evaluator/metric/promotion-policy versions; - drift and calibration state; - limitations/assumptions in the promotion receipt. Hard query distinction: `admitted learned knowledge` is an operational/semantic status, not source-explicit truth, epistemic warrant, capability or authority. Historical queries must recover the exact promotion/evaluation state that applied at the requested knowledge frontier. Refs prolog-rlm#404B.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Machine Spirit #404C handoff — query foreign vs local learned status

General semantic query/explanation should expose learned-object transfer state explicitly:

foreign_candidate | evidence_only | incompatible | quarantined | locally_admitted

and return source admission authority/policy, target local promotion receipt, transfer assumptions, guarantee status (preserved|translated|weakened|invalidated|unknown|noncomposable), ontology/identity mappings, origin-dependence status and drift/calibration epoch separately.

A query must not flatten admitted_by(source) into local warrant. If privacy prevents proving source independence, report independence unknown and do not count it as independent corroboration/validation. Full design: prolog-rlm#404C.

## Machine Spirit #404C handoff — query foreign vs local learned status General semantic query/explanation should expose learned-object transfer state explicitly: ```text foreign_candidate | evidence_only | incompatible | quarantined | locally_admitted ``` and return source admission authority/policy, target local promotion receipt, transfer assumptions, guarantee status (`preserved|translated|weakened|invalidated|unknown|noncomposable`), ontology/identity mappings, origin-dependence status and drift/calibration epoch separately. A query must not flatten `admitted_by(source)` into local warrant. If privacy prevents proving source independence, report independence `unknown` and do not count it as independent corroboration/validation. Full design: prolog-rlm#404C.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Implementation decomposition

#7 remains the canonical general symbolic query/reasoning requirement. Focused implementation is now:

#25 support/counterevidence/provenance persistence
 -> #26 semantic indexes
 -> #27 identity/source-dependence reconciliation
 -> #28 general semantic query API
 -> #29 TAPS/OATH bounded dependency-complete projection
 -> #30 RACE cache/cursor/generation/replay receipts

Prolog-RLM keeps epistemic consequence/reasoning semantics. Symbolic Memory supplies durable indexed evidence, exact cuts, identities, bounded projections, and replayable query receipts.

## Implementation decomposition #7 remains the canonical general symbolic query/reasoning requirement. Focused implementation is now: ```text #25 support/counterevidence/provenance persistence -> #26 semantic indexes -> #27 identity/source-dependence reconciliation -> #28 general semantic query API -> #29 TAPS/OATH bounded dependency-complete projection -> #30 RACE cache/cursor/generation/replay receipts ``` Prolog-RLM keeps epistemic consequence/reasoning semantics. Symbolic Memory supplies durable indexed evidence, exact cuts, identities, bounded projections, and replayable query receipts.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

MACHINE-SPIRIT #404D query-surface refinement

Learned-object queries must expose status and guarantees as structured, versioned results, not flatten them to one confidence/admitted flag.

For a candidate/admitted/foreign object, query/explain should be able to return: exact candidate/version; current lifecycle; origin experiences; exposure/dependence completeness (independent|dependent|unknown as justified); EvaluationSnapshot/profile; support/counterexample coverage; baseline/safe-improvement receipt where applicable; LAC/attestation/transfer chain; drift/calibration epoch; and historical suspension/supersession lineage.

Bounded counterexample coverage, statistical improvement and exact symbolic proof are different guarantee classes. Query APIs must not present them as interchangeable confidence. Refs prolog-rlm#404D and #16/#35/#36.

## MACHINE-SPIRIT #404D query-surface refinement Learned-object queries must expose **status and guarantees as structured, versioned results**, not flatten them to one confidence/admitted flag. For a candidate/admitted/foreign object, query/explain should be able to return: exact candidate/version; current lifecycle; origin experiences; exposure/dependence completeness (`independent|dependent|unknown` as justified); EvaluationSnapshot/profile; support/counterexample coverage; baseline/safe-improvement receipt where applicable; LAC/attestation/transfer chain; drift/calibration epoch; and historical suspension/supersession lineage. Bounded counterexample coverage, statistical improvement and exact symbolic proof are different guarantee classes. Query APIs must not present them as interchangeable confidence. Refs prolog-rlm#404D and #16/#35/#36.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

prolog-rlm#405B AEGIS-MS query handoff: strong query/answer receipts need more than (world_cut, epistemic_profile). Expose the material identity/admission/redaction/integrity/control generations plus direct influence/coverage gaps required by the caller's conformance profile. A result cannot claim current/exact if a mandatory dependency is stale/unavailable/redacted beyond the required evidence. Query provenance must keep source provenance, logical justification, and control/influence provenance distinct; none is host authority.

prolog-rlm#405B AEGIS-MS query handoff: strong query/answer receipts need more than `(world_cut, epistemic_profile)`. Expose the material identity/admission/redaction/integrity/control generations plus direct influence/coverage gaps required by the caller's conformance profile. A result cannot claim `current/exact` if a mandatory dependency is stale/unavailable/redacted beyond the required evidence. Query provenance must keep `source provenance`, `logical justification`, and `control/influence provenance` distinct; none is host authority.

nsaspy commented

2026-09-08 01:57:24 +00:00

Author

Owner

Machine Spirit #405C / CONCORDAT query handoff

General semantic query must make federation explicit in the query contract. all memories must not become an ambient union of foreign theories/trust domains.

A federated query/result should pin:

target domain + epistemic profile
exact federation cut
admitted directional bridge contracts
identity mappings
origin/dependence policy
foreign trust-root/verifier refs
target appraisal policy

Cross-domain results preserve status vectors/losses. In particular foreign credulous/undefined/conflicted states cannot be relabeled as target skeptical/false/winner; missing/private provenance cannot prove independence; multiple/no equilibrium remains explicit.

Known already-projected federated queries must remain eligible for model_calls = 0.

Canonical research + C8-C13/C45-C50: lost-rob0t/prolog-rlm#405C.

## Machine Spirit #405C / CONCORDAT query handoff General semantic query must make federation explicit in the query contract. `all memories` must not become an ambient union of foreign theories/trust domains. A federated query/result should pin: ```text target domain + epistemic profile exact federation cut admitted directional bridge contracts identity mappings origin/dependence policy foreign trust-root/verifier refs target appraisal policy ``` Cross-domain results preserve status vectors/losses. In particular foreign credulous/undefined/conflicted states cannot be relabeled as target skeptical/false/winner; missing/private provenance cannot prove independence; multiple/no equilibrium remains explicit. Known already-projected federated queries must remain eligible for `model_calls = 0`. Canonical research + C8-C13/C45-C50: lost-rob0t/prolog-rlm#405C.

nsaspy referenced this issue

2026-09-08 01:57:32 +00:00

[semantic-memory] Expose general semantic ingestion, query, provenance, and bounded world-model projections through native/MCP tools #40

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] General symbolic recall/query over facts, rules, events, time, causality, procedures, constraints, claims, and provenance #42

Goal

Query classes

Entity / definition / taxonomy

Rule / implication

Procedure / planning

Event / state / temporal

Causal / diagnostic

Constraint / quantitative

Normative / policy knowledge

Claims / beliefs / conflicting sources

Goals / plans / preferences

Hypothesis / scenario

General API direction

MACHINE-SPIRIT #400A epistemic query contract

Result model

Reasoning

Bounded projection

Acceptance

MACHINE-SPIRIT #400B query handoff — defeat paths, independence, known-at vs valid-at

MACHINE-SPIRIT #400C query handoff — explicit epistemic federation

MACHINE-SPIRIT #400D handoff — federated query planning / projected-interface execution

Machine Spirit #401C handoff — procedural portability/query status

Upstream Machine Spirit #401D handoff — procedure query/admission status

MACHINE-SPIRIT #402B handoff — historical commitment vs retrievability

MACHINE-SPIRIT #402C handoff — federated query/cut semantics

Machine Spirit prolog-rlm#403A handoff — TAPS bounded projection semantics

Upstream Machine Spirit #403B handoff — OATH-TAPS projection/query semantics

MACHINE-SPIRIT prolog-rlm#403C handoff — MOSAIC-TAPS

Machine Spirit #403D handoff — RACE-TAPS query/projection execution

MACHINE-SPIRIT #404A handoff — learned-knowledge query/explanation

Machine Spirit #404B / CITADEL-EPOCH query/explanation handoff

Machine Spirit #404C handoff — query foreign vs local learned status

Implementation decomposition

MACHINE-SPIRIT #404D query-surface refinement

Machine Spirit #405C / CONCORDAT query handoff