nsaspy/prolog-rlm

Fork 0

[ARADR][MS1-R1][RESEARCH] Knowledge-representation foundations evidence inventory #386

New issue

Open

opened 2026-09-08 01:29:17 +00:00 by nsaspy · 0 comments

nsaspy commented

2026-09-08 01:29:17 +00:00

Owner

Parent: #411
Topic: MS1 — Knowledge Representation
Mode: RESEARCH ONLY
Canonical prior inputs: #398 #392 #397 #400 #401 #405

Pass contract

This pass does not choose a canonical Machine Spirit representation, profile set, storage schema, or executable lowering architecture.

Earlier #398/#392 material is treated as prior research/design evidence, not automatically re-approved design, because the old Cycle-1 pass combined research and design in one run. This pass separates the research evidence so MS1-D1 can make decisions in a later design-only loop.

Questions investigated

What representation families are actually needed to preserve heterogeneous knowledge without conflating storage syntax with reasoning semantics?
What current standards/implementations demonstrate separation between representation, validation, entailment/query semantics, and execution?
What machinery is needed for propositions/assertions, n-ary relations, quantifiers/existentials, contexts/theories, provenance, events/time, rules, and derived state?
Which composition/modularity guarantees are known to be non-trivial?
What current executable support exists in prolog-rlm main?
Which questions remain research questions rather than design choices?

Research findings

R1-F01 — Representation syntax does not determine one entailment semantics

Current standards continue to provide multiple explicit semantic/query regimes over related data models rather than one universal reasoning behavior.

Evidence:

W3C RIF uses multiple rule dialects with different expressiveness/semantics: https://www.w3.org/TR/rif-overview/ and https://www.w3.org/TR/rif-core/
OWL 2 defines EL/QL/RL profiles with materially different reasoning/performance tradeoffs: https://www.w3.org/TR/owl2-profiles/
SPARQL 1.2 Entailment Regimes remains a separate 2026 specification defining query-answer semantics under different entailment relations: https://www.w3.org/TR/sparql12-entailment/

Research implication only: Machine Spirit's KR problem must account for semantic/query regime identity independently of surface record syntax. Which profiles v1 exposes is left to MS1-D1/later evidence.

R1-F02 — RDF 1.2 materially improves statement-as-data representation but remains a graph data model

RDF 1.2 Concepts is a W3C Candidate Recommendation Snapshot dated 2026-04-07. RDF graphs remain sets of subject-predicate-object triples, while RDF 1.2 adds triple terms, allowing an RDF triple to occur as a term and improving statement-level description/reification.

Evidence:

This is strong evidence for preserving statement/proposition identity and metadata about statements. It is not evidence that triples alone encode all required quantifier, procedural, default, epistemic, or temporal semantics.

R1-F03 — Validation and inference remain distinct concerns in current standards work

SHACL 1.2 Core describes structural constraints over RDF data graphs. SHACL 1.2 Rules separately defines rule-based inference. The Rules draft exposes distinct infer and query operations and produces an inference graph; unioning inferred output with the base graph is not inherently mandatory.

Evidence:

SHACL 1.2 Core: https://www.w3.org/TR/shacl12-core/
SHACL 1.2 Rules: https://www.w3.org/TR/shacl12-rules/

The current Rules work also has to define behavior around creation of new RDF terms and negation-as-failure, demonstrating that adding inference changes semantic and termination concerns beyond schema validation.

Research implication only: shape/type/schema validity cannot be treated as proof that a proposition is entailed or operationally executable.

R1-F04 — Source/assertion context cannot safely be reduced to a global graph union

Prior #398B research on McCarthy/Guha contexts and source revision theories remains relevant. Current SPARQL entailment work also evaluates query semantics relative to an active graph and an entailment regime; semantic query behavior is not equivalent to blindly unioning every graph/source.

Evidence:

prior #398B context/microtheory survey and stress tests;
https://www.w3.org/TR/sparql12-entailment/

Inputs such as LLM logs, news, Wikipedia revisions, policies, scientific passages, hypotheticals, and quoted claims can make the same proposition usable under different source/speaker/time/scenario conditions.

Open research remains on exact context composition and bridge guarantees; this pass does not choose a context implementation.

R1-F05 — Existential knowledge requires explicit unknown-witness semantics

Plain function-free Datalog cannot faithfully express every statement of the form forall X ... exists Y .... Existential-rule systems use null/witness-like terms plus syntactic restrictions to recover useful computational properties.

Current implementation evidence:

Vadalog is based on Warded Datalog± and documents existential quantification, marked nulls, wardedness, EGDs and practical rule extensions: https://vadalog.org/vadalog-handbook/latest/vadalog-language.html

Prior #398B literature also covered guarded/warded existential rules and chase restrictions.

Research invariant:

unknown/existential witness != globally identified real-world entity

Exact admitted existential fragment, witness canonicalization, and benchmark envelope remain unresolved.

R1-F06 — Modular theory composition has stronger semantics than generic import/merge

Prior #398C found that composition may need distinct relations such as include/reference/translate/bridge/query-preserve/conservative-extension rather than one generic imports edge.

Description-logic modularity literature reinforces that conservative extension and query inseparability are distinct logical guarantees, and many forms are computationally hard or undecidable outside restricted fragments.

Evidence:

Botoeva et al., Inseparability and Conservative Extensions of Description Logic Ontologies: A Survey, Reasoning Web 2016: https://link.springer.com/book/10.1007/978-3-319-49493-7
broader ontology/data-management survey: https://link.springer.com/article/10.1007/s13218-020-00686-3

Research implication only: syntactic compatibility/union is insufficient evidence of semantic substitutability or query preservation.

R1-F07 — Demand-driven deductive evaluation is practical but fragment-dependent

Soufflé currently implements Magic-Set transformation to specialize positive Datalog evaluation and avoid irrelevant tuples. Its current documentation explicitly notes limitations: relations involving negation, functors, or aggregates are skipped by the implemented transformation.

Evidence:

This is implementation evidence that query-directed compilation/materialization can reduce work, while also showing that optimization guarantees depend on the admitted rule fragment.

R1-F08 — Provenance can be demand-produced from derivation metadata rather than eagerly storing every proof tree

Soufflé's current provenance implementation keeps rule/height annotations during evaluation and lazily constructs proof trees when explanations are requested.

Evidence:

https://souffle-lang.github.io/provenance

Prior #398B research on TMS/ATMS and provenance semirings independently establishes the value of explicit dependency/justification structure. This does not settle which proof/justification artifacts belong in canonical history versus rebuildable state.

R1-F09 — Incremental derived-state maintenance is executable today in SWI-Prolog, with semantic restrictions

SWI-Prolog currently documents:

incremental tabling that invalidates dependent tables after incremental dynamic changes and lazily recomputes them while preserving normal tabling properties including well-founded semantics;
monotonic tabling that propagates additions incrementally without full recomputation but applies only to monotonic programs and does not handle negation.

Evidence:

This supports treating derived/materialized state as dependency-tracked/rebuildable in at least some fragments. It does not establish Machine Spirit-scale performance or a universal incremental algorithm.

R1-F10 — Semantic mapping and epistemic acceptance are distinct research layers

The #400C handoff into #392 records a key cross-depth distinction:

semantic mapping/bridge
!= epistemic warrant transfer

A mapping may preserve proposition meaning while source/destination theories still disagree on negation, defaults, priorities, model plurality, trust or acceptance policy.

This is retained as a cross-topic dependency on MS3 rather than solved inside MS1-R1.

R1-F11 — Procedural representation must preserve distinctions before execution is considered

The #401A handoff into #392 records these distinct artifacts:

method knowledge
!= decomposition decision
!= plan instance
!= expert invocation
!= effect attempt
!= observed transition
!= VERIFY result

MS1 therefore has representation requirements originating in MS4, but this R1 does not select procedural constructors or an executable procedural profile.

R1-F12 — Canonical semantic records, deductive planning, and derived state are separate evidence classes in the prior corpus

#398C/D and #392 distinguish:

immutable/versioned semantic records and theory graph;
compiled/query-specific deductive plan state;
tables/materializations/current views/caches/intermediate derivations.

The current implementation evidence from SWI/Soufflé makes the distinction technically plausible, but no canonical Machine Spirit API implementing those planes exists today.

R1-F13 — Current `prolog-rlm` main does not implement the designed Machine Spirit semantic surface

Current GitHub code search on lost-rob0t/prolog-rlm main returned no canonical implementation hits for:

semantic_compile
sem_context

This matches #397/#405's executable acceptance audit. Therefore all sem_* examples in #392/#398 remain design-shaped research artifacts, not evidence of an implemented Machine Spirit semantic kernel.

Prior-cycle evidence normalization

The old #398 pass cannot be copied directly into ARADR MS1-R1 because it contains both literature research and explicit architecture selection (preferred, decision, adopt, design changes). Its source/evidence survey is retained here as prior evidence. Its architectural conclusions become inputs to the later design-only MS1-D1, not conclusions of this pass.

Likewise later #392 handoffs from #400/#401 are recorded as cross-topic constraints/evidence, not silently re-approved MS1 design.

Contradictions / tensions found

Universal representation vs profile-specific semantics — broad formula preservation is compatible with limited execution only if support/entailment guarantees are explicit; the exact boundary remains a design question.
First-class proposition identity vs context-sensitive meaning — structural sameness across theory/vocabulary versions may not imply interchangeable semantic identity.
Append-only canonical history vs efficient explanations/materializations — keeping all proofs eagerly is expensive; keeping none loses lineage unless derivation dependencies/receipts are reconstructible.
Open vocabulary vs safe lowering — extension names are necessary for arbitrary domains but cannot acquire host predicate/execution authority by lexical identity.
Existential expressiveness vs bounded reasoning — unrestricted chase behavior can be non-terminating; useful restricted fragments exist but v1 boundaries are not yet benchmarked.
Modular composition vs tractable compatibility checks — strong logical guarantees may be infeasible/undecidable for expressive fragments, requiring explicit approximation or restricted profiles.

Research gaps created for later RESEARCH passes

These remain RESEARCH REQUIRED unless another existing topic/pass already closes them with adequate evidence:

MS1-G01 — Canonical proposition/formula equivalence and fingerprinting

Research exact normalization/equivalence choices for:

alpha-renamed bound variables;
conjunction/disjunction ordering;
associative/commutative structure;
literals/numeric/unit normalization;
vocabulary/schema/theory version in fingerprints;
treatment of logically equivalent but structurally different formulas.

Goal is evidence, not a design decision.

MS1-G02 — Context/microtheory composition semantics

Compare explicit contextual logics/microtheories, RDF datasets/named graphs, local/contextual description logics and practical scoped-query systems. Determine which guarantees can be made without context-product explosion or implicit global union.

MS1-G03 — Executable existential fragment envelope

Benchmark/research candidate guarded/warded/acyclic existential-rule fragments against Machine Spirit query patterns. Determine expected chase termination/resource behavior, witness identity requirements and interaction with recursion/negation.

MS1-G04 — Theory bridge/module guarantee envelope

Research which query-inseparability/conservative-extension/mapping guarantees are decidable or practical for likely v1 fragments and what weaker contracts exist when exact checking is intractable.

MS1-G05 — Provenance/justification retention boundary

Research storage/complexity tradeoffs among canonical direct justification edges, durable promoted receipts, lazy proof reconstruction, provenance polynomials/semirings and rebuildable intermediate proofs.

MS1-G06 — Cross-profile composition semantics

Research safe composition between Horn/Datalog, temporal/event, constraints, existential rules, DL/ontology mappings, and preservation-only quantified formulas without laundering one profile's guarantee into another.

MS1-G07 — Scale evidence

Existing implementation examples prove techniques exist but do not establish Machine Spirit's scale envelope. Gather measured data for relevant rule fragments, incremental invalidation, explanation, context selection and existential reasoning on representative corpus sizes.

Design inputs now ready for MS1-D1

The following are evidence packages, not decisions:

current standards demonstrate profile/regime plurality;
RDF 1.2 demonstrates first-class statement terms while remaining graph-based;
SHACL 1.2 demonstrates structural-validation/inference separation;
existential-rule implementations demonstrate explicit witnesses under restricted fragments;
modularity literature demonstrates import/merge is weaker than semantic/query preservation;
SWI/Soufflé demonstrate fragment-specific incremental, demand-driven and explanation techniques;
prior Machine Spirit research requires separation among semantics, epistemic warrant, procedure/planning/execution, and derived state;
current prolog-rlm does not implement the target semantic API.

MS1-D1 may consume these findings and the full prior #398/#392 corpus. It must not perform new external research while designing.

Pass checkpoint

PASS: MS1-R1
MODE: RESEARCH
STATUS: COMPLETE
CANONICAL DESIGN CHANGED: NO
IMPLEMENTATION CHANGED: NO
CURRENT EXECUTABLE MACHINE-SPIRIT KR: NOT PRESENT
NEXT NORMAL PASS: MS1-D1 (DESIGN ONLY)
OPEN RESEARCH NODES: MS1-G01..MS1-G07

Parent: #411 Topic: MS1 — Knowledge Representation Mode: **RESEARCH ONLY** Canonical prior inputs: #398 #392 #397 #400 #401 #405 ## Pass contract This pass does **not** choose a canonical Machine Spirit representation, profile set, storage schema, or executable lowering architecture. Earlier #398/#392 material is treated as prior research/design evidence, not automatically re-approved design, because the old Cycle-1 pass combined research and design in one run. This pass separates the research evidence so MS1-D1 can make decisions in a later design-only loop. ## Questions investigated 1. What representation families are actually needed to preserve heterogeneous knowledge without conflating storage syntax with reasoning semantics? 2. What current standards/implementations demonstrate separation between representation, validation, entailment/query semantics, and execution? 3. What machinery is needed for propositions/assertions, n-ary relations, quantifiers/existentials, contexts/theories, provenance, events/time, rules, and derived state? 4. Which composition/modularity guarantees are known to be non-trivial? 5. What current executable support exists in `prolog-rlm` main? 6. Which questions remain research questions rather than design choices? # Research findings ## R1-F01 — Representation syntax does not determine one entailment semantics Current standards continue to provide multiple explicit semantic/query regimes over related data models rather than one universal reasoning behavior. Evidence: - W3C RIF uses multiple rule dialects with different expressiveness/semantics: https://www.w3.org/TR/rif-overview/ and https://www.w3.org/TR/rif-core/ - OWL 2 defines EL/QL/RL profiles with materially different reasoning/performance tradeoffs: https://www.w3.org/TR/owl2-profiles/ - SPARQL 1.2 Entailment Regimes remains a separate 2026 specification defining query-answer semantics under different entailment relations: https://www.w3.org/TR/sparql12-entailment/ Research implication only: Machine Spirit's KR problem must account for semantic/query regime identity independently of surface record syntax. Which profiles v1 exposes is left to MS1-D1/later evidence. ## R1-F02 — RDF 1.2 materially improves statement-as-data representation but remains a graph data model RDF 1.2 Concepts is a W3C Candidate Recommendation Snapshot dated 2026-04-07. RDF graphs remain sets of subject-predicate-object triples, while RDF 1.2 adds **triple terms**, allowing an RDF triple to occur as a term and improving statement-level description/reification. Evidence: - https://www.w3.org/TR/rdf12-concepts/ - https://www.w3.org/news/2026/w3c-invites-implementations-of-rdf-1-2-concepts-and-abstract-data-model-and-rdf-1-2-semantics/ This is strong evidence for preserving statement/proposition identity and metadata about statements. It is not evidence that triples alone encode all required quantifier, procedural, default, epistemic, or temporal semantics. ## R1-F03 — Validation and inference remain distinct concerns in current standards work SHACL 1.2 Core describes structural constraints over RDF data graphs. SHACL 1.2 Rules separately defines rule-based inference. The Rules draft exposes distinct `infer` and `query` operations and produces an inference graph; unioning inferred output with the base graph is not inherently mandatory. Evidence: - SHACL 1.2 Core: https://www.w3.org/TR/shacl12-core/ - SHACL 1.2 Rules: https://www.w3.org/TR/shacl12-rules/ The current Rules work also has to define behavior around creation of new RDF terms and negation-as-failure, demonstrating that adding inference changes semantic and termination concerns beyond schema validation. Research implication only: shape/type/schema validity cannot be treated as proof that a proposition is entailed or operationally executable. ## R1-F04 — Source/assertion context cannot safely be reduced to a global graph union Prior #398B research on McCarthy/Guha contexts and source revision theories remains relevant. Current SPARQL entailment work also evaluates query semantics relative to an active graph and an entailment regime; semantic query behavior is not equivalent to blindly unioning every graph/source. Evidence: - prior #398B context/microtheory survey and stress tests; - https://www.w3.org/TR/sparql12-entailment/ Inputs such as LLM logs, news, Wikipedia revisions, policies, scientific passages, hypotheticals, and quoted claims can make the same proposition usable under different source/speaker/time/scenario conditions. Open research remains on exact context composition and bridge guarantees; this pass does not choose a context implementation. ## R1-F05 — Existential knowledge requires explicit unknown-witness semantics Plain function-free Datalog cannot faithfully express every statement of the form `forall X ... exists Y ...`. Existential-rule systems use null/witness-like terms plus syntactic restrictions to recover useful computational properties. Current implementation evidence: - Vadalog is based on Warded Datalog± and documents existential quantification, marked nulls, wardedness, EGDs and practical rule extensions: https://vadalog.org/vadalog-handbook/latest/vadalog-language.html Prior #398B literature also covered guarded/warded existential rules and chase restrictions. Research invariant: ```text unknown/existential witness != globally identified real-world entity ``` Exact admitted existential fragment, witness canonicalization, and benchmark envelope remain unresolved. ## R1-F06 — Modular theory composition has stronger semantics than generic import/merge Prior #398C found that composition may need distinct relations such as include/reference/translate/bridge/query-preserve/conservative-extension rather than one generic `imports` edge. Description-logic modularity literature reinforces that conservative extension and query inseparability are distinct logical guarantees, and many forms are computationally hard or undecidable outside restricted fragments. Evidence: - Botoeva et al., *Inseparability and Conservative Extensions of Description Logic Ontologies: A Survey*, Reasoning Web 2016: https://link.springer.com/book/10.1007/978-3-319-49493-7 - broader ontology/data-management survey: https://link.springer.com/article/10.1007/s13218-020-00686-3 Research implication only: syntactic compatibility/union is insufficient evidence of semantic substitutability or query preservation. ## R1-F07 — Demand-driven deductive evaluation is practical but fragment-dependent Soufflé currently implements Magic-Set transformation to specialize positive Datalog evaluation and avoid irrelevant tuples. Its current documentation explicitly notes limitations: relations involving negation, functors, or aggregates are skipped by the implemented transformation. Evidence: - https://souffle-lang.github.io/magicset - https://souffle-lang.github.io/relations This is implementation evidence that query-directed compilation/materialization can reduce work, while also showing that optimization guarantees depend on the admitted rule fragment. ## R1-F08 — Provenance can be demand-produced from derivation metadata rather than eagerly storing every proof tree Soufflé's current provenance implementation keeps rule/height annotations during evaluation and lazily constructs proof trees when explanations are requested. Evidence: - https://souffle-lang.github.io/provenance Prior #398B research on TMS/ATMS and provenance semirings independently establishes the value of explicit dependency/justification structure. This does not settle which proof/justification artifacts belong in canonical history versus rebuildable state. ## R1-F09 — Incremental derived-state maintenance is executable today in SWI-Prolog, with semantic restrictions SWI-Prolog currently documents: - incremental tabling that invalidates dependent tables after incremental dynamic changes and lazily recomputes them while preserving normal tabling properties including well-founded semantics; - monotonic tabling that propagates additions incrementally without full recomputation but applies only to monotonic programs and does not handle negation. Evidence: - https://www.swi-prolog.org/pldoc/man?section=tabling-incremental - https://www.swi-prolog.org/pldoc/man?section=tabling-monotonic This supports treating derived/materialized state as dependency-tracked/rebuildable in at least some fragments. It does not establish Machine Spirit-scale performance or a universal incremental algorithm. ## R1-F10 — Semantic mapping and epistemic acceptance are distinct research layers The #400C handoff into #392 records a key cross-depth distinction: ```text semantic mapping/bridge != epistemic warrant transfer ``` A mapping may preserve proposition meaning while source/destination theories still disagree on negation, defaults, priorities, model plurality, trust or acceptance policy. This is retained as a cross-topic dependency on MS3 rather than solved inside MS1-R1. ## R1-F11 — Procedural representation must preserve distinctions before execution is considered The #401A handoff into #392 records these distinct artifacts: ```text method knowledge != decomposition decision != plan instance != expert invocation != effect attempt != observed transition != VERIFY result ``` MS1 therefore has representation requirements originating in MS4, but this R1 does not select procedural constructors or an executable procedural profile. ## R1-F12 — Canonical semantic records, deductive planning, and derived state are separate evidence classes in the prior corpus #398C/D and #392 distinguish: 1. immutable/versioned semantic records and theory graph; 2. compiled/query-specific deductive plan state; 3. tables/materializations/current views/caches/intermediate derivations. The current implementation evidence from SWI/Soufflé makes the distinction technically plausible, but no canonical Machine Spirit API implementing those planes exists today. ## R1-F13 — Current `prolog-rlm` main does not implement the designed Machine Spirit semantic surface Current GitHub code search on `lost-rob0t/prolog-rlm` main returned no canonical implementation hits for: ```text semantic_compile sem_context ``` This matches #397/#405's executable acceptance audit. Therefore all `sem_*` examples in #392/#398 remain design-shaped research artifacts, not evidence of an implemented Machine Spirit semantic kernel. # Prior-cycle evidence normalization The old #398 pass cannot be copied directly into ARADR `MS1-R1` because it contains both literature research and explicit architecture selection (`preferred`, `decision`, `adopt`, design changes). Its source/evidence survey is retained here as prior evidence. Its architectural conclusions become **inputs to the later design-only MS1-D1**, not conclusions of this pass. Likewise later #392 handoffs from #400/#401 are recorded as cross-topic constraints/evidence, not silently re-approved MS1 design. # Contradictions / tensions found 1. **Universal representation vs profile-specific semantics** — broad formula preservation is compatible with limited execution only if support/entailment guarantees are explicit; the exact boundary remains a design question. 2. **First-class proposition identity vs context-sensitive meaning** — structural sameness across theory/vocabulary versions may not imply interchangeable semantic identity. 3. **Append-only canonical history vs efficient explanations/materializations** — keeping all proofs eagerly is expensive; keeping none loses lineage unless derivation dependencies/receipts are reconstructible. 4. **Open vocabulary vs safe lowering** — extension names are necessary for arbitrary domains but cannot acquire host predicate/execution authority by lexical identity. 5. **Existential expressiveness vs bounded reasoning** — unrestricted chase behavior can be non-terminating; useful restricted fragments exist but v1 boundaries are not yet benchmarked. 6. **Modular composition vs tractable compatibility checks** — strong logical guarantees may be infeasible/undecidable for expressive fragments, requiring explicit approximation or restricted profiles. # Research gaps created for later RESEARCH passes These remain `RESEARCH REQUIRED` unless another existing topic/pass already closes them with adequate evidence: ### MS1-G01 — Canonical proposition/formula equivalence and fingerprinting Research exact normalization/equivalence choices for: - alpha-renamed bound variables; - conjunction/disjunction ordering; - associative/commutative structure; - literals/numeric/unit normalization; - vocabulary/schema/theory version in fingerprints; - treatment of logically equivalent but structurally different formulas. Goal is evidence, not a design decision. ### MS1-G02 — Context/microtheory composition semantics Compare explicit contextual logics/microtheories, RDF datasets/named graphs, local/contextual description logics and practical scoped-query systems. Determine which guarantees can be made without context-product explosion or implicit global union. ### MS1-G03 — Executable existential fragment envelope Benchmark/research candidate guarded/warded/acyclic existential-rule fragments against Machine Spirit query patterns. Determine expected chase termination/resource behavior, witness identity requirements and interaction with recursion/negation. ### MS1-G04 — Theory bridge/module guarantee envelope Research which query-inseparability/conservative-extension/mapping guarantees are decidable or practical for likely v1 fragments and what weaker contracts exist when exact checking is intractable. ### MS1-G05 — Provenance/justification retention boundary Research storage/complexity tradeoffs among canonical direct justification edges, durable promoted receipts, lazy proof reconstruction, provenance polynomials/semirings and rebuildable intermediate proofs. ### MS1-G06 — Cross-profile composition semantics Research safe composition between Horn/Datalog, temporal/event, constraints, existential rules, DL/ontology mappings, and preservation-only quantified formulas without laundering one profile's guarantee into another. ### MS1-G07 — Scale evidence Existing implementation examples prove techniques exist but do not establish Machine Spirit's scale envelope. Gather measured data for relevant rule fragments, incremental invalidation, explanation, context selection and existential reasoning on representative corpus sizes. # Design inputs now ready for MS1-D1 The following are **evidence packages**, not decisions: - current standards demonstrate profile/regime plurality; - RDF 1.2 demonstrates first-class statement terms while remaining graph-based; - SHACL 1.2 demonstrates structural-validation/inference separation; - existential-rule implementations demonstrate explicit witnesses under restricted fragments; - modularity literature demonstrates import/merge is weaker than semantic/query preservation; - SWI/Soufflé demonstrate fragment-specific incremental, demand-driven and explanation techniques; - prior Machine Spirit research requires separation among semantics, epistemic warrant, procedure/planning/execution, and derived state; - current `prolog-rlm` does not implement the target semantic API. MS1-D1 may consume these findings and the full prior #398/#392 corpus. It must not perform new external research while designing. # Pass checkpoint ```text PASS: MS1-R1 MODE: RESEARCH STATUS: COMPLETE CANONICAL DESIGN CHANGED: NO IMPLEMENTATION CHANGED: NO CURRENT EXECUTABLE MACHINE-SPIRIT KR: NOT PRESENT NEXT NORMAL PASS: MS1-D1 (DESIGN ONLY) OPEN RESEARCH NODES: MS1-G01..MS1-G07 ```

nsaspy referenced this issue

2026-09-08 01:29:18 +00:00

[MACHINE-SPIRIT 7/8] Meta-reasoning and bounded self-improvement: failure learning, strategy evolution, rule promotion, self-model and expert synthesis #394

nsaspy referenced this issue

2026-09-08 01:29:20 +00:00

[EPIC][experts] Closed-loop integration: Requirement→Knowledge→Plan→Experts→VERIFY→Repair→Review→completion #411

nsaspy referenced this issue

2026-09-08 01:46:19 +00:00

[EPIC][P0] First-class Prolog expert system: registry, routing, specialists, and closed-loop Spec→Plan→Expert→Verify→Repair #422

nsaspy referenced this issue

2026-09-08 01:46:51 +00:00

[EPIC][expert] Review / critic expert: adversarial change analysis, regression discovery, and evidence-backed findings #412