[semantic-memory] Append-only general semantic ledger: arbitrary IR records, temporal/current views, conflict, supersession, and provenance

nsaspy commented

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

Owner

Parent: #4
Depends on: #1, #5
Upstream IR: lost-rob0t/prolog-rlm#392

Goal

Persist the full general semantic IR as immutable append-only knowledge history while exposing efficient namespace-scoped current/history/world-model views.

This is not a fact/rule-only table. Storage must preserve the semantic shape of every supported upstream record: entities/types, relations, rules, defaults/exceptions, procedures, events/states, temporal/causal/modal/constraint knowledge, attributed claims, hypotheses, preferences, examples, and extension vocabulary.

Core invariant

Never rewrite an old semantic record in place to make it look current.

Represent evolution explicitly:

projection_asserted(Record).
projection_supersedes(New, Old).
projection_retracts(Retraction, Old, Reason).
projection_conflicts(A, B).
projection_corroborates(A, B).
projection_invalidates(Event, Old, Reason).

Exact vocabulary follows repository conventions.

Required record identity

Every semantic record must bind to:

stable semantic record ID/fingerprint;
memory ID + exact source version/hash;
namespace/scope;
upstream semantic IR/schema version;
domain vocabulary/extension version where applicable;
semantic compiler version/run/fingerprint;
exact source span(s);
explicit-vs-inferred/derived classification;
trust/provenance class;
semantic scope: time/scenario/speaker/location/etc. where applicable;
projection generation;
append event identity/time.

Preserve semantic structure

Do not denormalize away distinctions needed for reasoning, especially:

assertion vs attributed claim
explicit false vs unknown
strict rule vs default/exception
possible cause vs asserted cause
historical vs current validity
hypothesis/scenario vs world assertion
procedure step/order/branch structure
quantity value + unit + dimension
modality: must/may/should/forbidden/etc.
quantifier/cardinality

Indexes/materialized tables may optimize these shapes but are rebuildable caches, not the source of truth.

Current/history/world-model views

Provide general semantics equivalent to:

memory_semantic_history(+SubjectOrScope, +Options, -Outcome).
memory_semantic_current(+SubjectOrScope, +Options, -Outcome).
memory_semantic_record(+RecordRef, -Outcome).
memory_semantic_explain(+RecordRef, -Outcome).

Current views resolve retraction/supersession and temporal validity while preserving explicit conflict/uncertainty. A conflict does not automatically pick a winner.

Entity and relation indexing

Index for efficient lookup by at least:

entity/concept;
relation/predicate;
type/taxonomy;
event;
rule head/body concepts;
procedure goal/action;
time/validity;
source/provenance;
semantic class;
namespace/project;
domain vocabulary.

Do not require every extension relation to get a new storage migration.

Temporal/current semantics

A semantic record may be historically valid but no longer current. Current views must use upstream temporal/validity semantics rather than assuming latest append wins.

Examples:

role(A, maintainer) valid Jan-May
role(B, maintainer) valid Jun-current

Both remain in history; current query returns B for current time while a May query returns A.

Conflict and plurality

Preserve:

direct contradictions;
competing attributed claims;
ambiguous interpretations;
mutually exclusive hypotheses;
source disagreement;
incompatible entity resolutions.

Conflict links are semantic events/relations, not destructive reconciliation.

Trust semantics

Source trust and semantic interpretation trust are independent. Model-derived semantics remain model-derived even when attached to a trusted user's source memory.

Remembered normative/policy statements remain epistemic records and never become execution authority.

MACHINE-SPIRIT #400A epistemic ledger requirements

Depth 3A establishes that Symbolic Memory must not persist one mutable authoritative truth value for each proposition.

Persist/reconstruct a canonical Epistemic Support Graph (ESG) over the semantic ledger:

proposition
  ├─ positive support events
  ├─ negative support events
  ├─ independent justifications
  ├─ assumptions/context environments
  ├─ lifecycle events
  └─ provenance/validity

Required durable distinctions:

withdraw(source_assertion(P)) != assert(not(P))
source correction != world-state change
temporal expiry != epistemic retraction
default defeat != explicit negative evidence
supersession != deletion
trust != probability != compiler confidence

At minimum, preserve enough immutable data to rebuild proposition support states equivalent to:

neither
supported_only
refuted_only
both/conflicted

Those are evidence/support summaries, not universal truth values.

Durable epistemic records should include, where applicable:

support ID + proposition + polarity;
originating assertion/derivation;
context/theory + valid time;
justification dependency refs;
assumption/environment refs where promoted as durable knowledge;
explicit incompatibility/nogood evidence where promoted;
lifecycle events such as withdrawal, correction, supersession and derivation invalidation;
epistemic profile/version on durable promoted conclusion or acceptance receipts;
source/compiler/logical/lifecycle/profile provenance as separate classes.

Ordinary current support indexes, warranted-result caches, WFS tables, answer-set results, argumentation acceptance sets and revision projections are rebuildable derived state unless a promoted result requires a replay receipt.

A source withdrawal deactivates that source's current positive support but does not synthesize negative support. An explicit correction to ¬P may create a new negative support assertion in addition to the withdrawal/supersession relation.

Two independent derivations of P must remain independently addressable so invalidating one premise removes only the dependent derivation.

ATMS-style assumption environments are useful but can grow exponentially; do not require global environment-label materialization. Storage must support query-scoped/bounded environment reconstruction, nogood indexing and subsumption.

Extension durability

Persist unknown/new domain vocabulary using stable versioned references without requiring core code changes. Unknown extensions remain queryable data but cannot become executable host predicates.

Acceptance

Persist and reload at least one record from every major #392 semantic class.
Mixed semantic package preserves links among entities/events/rules/procedures/claims.
Correction creates new records + supersession, never mutation.
Retraction leaves old record in history but excludes it from ordinary current view.
Historical temporal view differs correctly from current view.
Contradictory memories preserve both records plus explicit conflict.
Attributed claims remain attributed after reopen.
Default/exception structure survives persistence exactly enough for upstream reasoning.
Quantities/units/cardinalities survive round trip.
Duplicate identical projection commit is idempotent.
Unknown/incompatible IR schema never becomes active current knowledge.
New domain vocabulary can be persisted without database-schema redesign.
Current/history views are namespace-isolated.
Every active semantic record explains exact source/compiler provenance.
Positive and negative support for one proposition can coexist durably.
Withdrawal does not create an explicit-negative support record unless the correction actually asserts one.
Independent derivations survive unrelated premise invalidation.
Current epistemic support/acceptance views can be rebuilt from immutable ledger history.
Epistemic profile/version is recoverable for any persisted promoted acceptance receipt.

Refs #4 #5 and prolog-rlm#392-#394 #397 #400.

Parent: #4 Depends on: #1, #5 Upstream IR: lost-rob0t/prolog-rlm#392 ## Goal Persist the **full general semantic IR** as immutable append-only knowledge history while exposing efficient namespace-scoped current/history/world-model views. This is not a fact/rule-only table. Storage must preserve the semantic shape of every supported upstream record: entities/types, relations, rules, defaults/exceptions, procedures, events/states, temporal/causal/modal/constraint knowledge, attributed claims, hypotheses, preferences, examples, and extension vocabulary. ## Core invariant Never rewrite an old semantic record in place to make it look current. Represent evolution explicitly: ```prolog projection_asserted(Record). projection_supersedes(New, Old). projection_retracts(Retraction, Old, Reason). projection_conflicts(A, B). projection_corroborates(A, B). projection_invalidates(Event, Old, Reason). ``` Exact vocabulary follows repository conventions. ## Required record identity Every semantic record must bind to: - stable semantic record ID/fingerprint; - memory ID + exact source version/hash; - namespace/scope; - upstream semantic IR/schema version; - domain vocabulary/extension version where applicable; - semantic compiler version/run/fingerprint; - exact source span(s); - explicit-vs-inferred/derived classification; - trust/provenance class; - semantic scope: time/scenario/speaker/location/etc. where applicable; - projection generation; - append event identity/time. ## Preserve semantic structure Do not denormalize away distinctions needed for reasoning, especially: ```text assertion vs attributed claim explicit false vs unknown strict rule vs default/exception possible cause vs asserted cause historical vs current validity hypothesis/scenario vs world assertion procedure step/order/branch structure quantity value + unit + dimension modality: must/may/should/forbidden/etc. quantifier/cardinality ``` Indexes/materialized tables may optimize these shapes but are rebuildable caches, not the source of truth. ## Current/history/world-model views Provide general semantics equivalent to: ```prolog memory_semantic_history(+SubjectOrScope, +Options, -Outcome). memory_semantic_current(+SubjectOrScope, +Options, -Outcome). memory_semantic_record(+RecordRef, -Outcome). memory_semantic_explain(+RecordRef, -Outcome). ``` Current views resolve retraction/supersession and temporal validity while preserving explicit conflict/uncertainty. A conflict does not automatically pick a winner. ## Entity and relation indexing Index for efficient lookup by at least: - entity/concept; - relation/predicate; - type/taxonomy; - event; - rule head/body concepts; - procedure goal/action; - time/validity; - source/provenance; - semantic class; - namespace/project; - domain vocabulary. Do not require every extension relation to get a new storage migration. ## Temporal/current semantics A semantic record may be historically valid but no longer current. Current views must use upstream temporal/validity semantics rather than assuming latest append wins. Examples: ```text role(A, maintainer) valid Jan-May role(B, maintainer) valid Jun-current ``` Both remain in history; current query returns B for current time while a May query returns A. ## Conflict and plurality Preserve: - direct contradictions; - competing attributed claims; - ambiguous interpretations; - mutually exclusive hypotheses; - source disagreement; - incompatible entity resolutions. Conflict links are semantic events/relations, not destructive reconciliation. ## Trust semantics Source trust and semantic interpretation trust are independent. Model-derived semantics remain model-derived even when attached to a trusted user's source memory. Remembered normative/policy statements remain epistemic records and never become execution authority. ## MACHINE-SPIRIT #400A epistemic ledger requirements Depth 3A establishes that Symbolic Memory must not persist one mutable authoritative `truth` value for each proposition. Persist/reconstruct a canonical **Epistemic Support Graph (ESG)** over the semantic ledger: ```text proposition ├─ positive support events ├─ negative support events ├─ independent justifications ├─ assumptions/context environments ├─ lifecycle events └─ provenance/validity ``` Required durable distinctions: ```text withdraw(source_assertion(P)) != assert(not(P)) source correction != world-state change temporal expiry != epistemic retraction default defeat != explicit negative evidence supersession != deletion trust != probability != compiler confidence ``` At minimum, preserve enough immutable data to rebuild proposition support states equivalent to: ```text neither supported_only refuted_only both/conflicted ``` Those are evidence/support summaries, not universal truth values. Durable epistemic records should include, where applicable: - support ID + proposition + polarity; - originating assertion/derivation; - context/theory + valid time; - justification dependency refs; - assumption/environment refs where promoted as durable knowledge; - explicit incompatibility/nogood evidence where promoted; - lifecycle events such as withdrawal, correction, supersession and derivation invalidation; - epistemic profile/version on durable promoted conclusion or acceptance receipts; - source/compiler/logical/lifecycle/profile provenance as separate classes. Ordinary current support indexes, warranted-result caches, WFS tables, answer-set results, argumentation acceptance sets and revision projections are **rebuildable derived state** unless a promoted result requires a replay receipt. A source withdrawal deactivates that source's current positive support but does not synthesize negative support. An explicit correction to ¬P may create a new negative support assertion in addition to the withdrawal/supersession relation. Two independent derivations of P must remain independently addressable so invalidating one premise removes only the dependent derivation. ATMS-style assumption environments are useful but can grow exponentially; do not require global environment-label materialization. Storage must support query-scoped/bounded environment reconstruction, nogood indexing and subsumption. ## Extension durability Persist unknown/new domain vocabulary using stable versioned references without requiring core code changes. Unknown extensions remain queryable data but cannot become executable host predicates. ## Acceptance - [ ] Persist and reload at least one record from every major #392 semantic class. - [ ] Mixed semantic package preserves links among entities/events/rules/procedures/claims. - [ ] Correction creates new records + supersession, never mutation. - [ ] Retraction leaves old record in history but excludes it from ordinary current view. - [ ] Historical temporal view differs correctly from current view. - [ ] Contradictory memories preserve both records plus explicit conflict. - [ ] Attributed claims remain attributed after reopen. - [ ] Default/exception structure survives persistence exactly enough for upstream reasoning. - [ ] Quantities/units/cardinalities survive round trip. - [ ] Duplicate identical projection commit is idempotent. - [ ] Unknown/incompatible IR schema never becomes active current knowledge. - [ ] New domain vocabulary can be persisted without database-schema redesign. - [ ] Current/history views are namespace-isolated. - [ ] Every active semantic record explains exact source/compiler provenance. - [ ] Positive and negative support for one proposition can coexist durably. - [ ] Withdrawal does not create an explicit-negative support record unless the correction actually asserts one. - [ ] Independent derivations survive unrelated premise invalidation. - [ ] Current epistemic support/acceptance views can be rebuilt from immutable ledger history. - [ ] Epistemic profile/version is recoverable for any persisted promoted acceptance receipt. Refs #4 #5 and prolog-rlm#392-#394 #397 #400.

nsaspy commented

2026-09-08 01:57:18 +00:00

Author

Owner

#398B storage consequence: the append-only semantic ledger should not flatten context/theory, existential witness, or justification structures into generic assertion metadata. Index current/history views by theory/context/profile; preserve source assertions independently from domain-theory promotions; persist witness scope/origin without promoting labelled unknowns to global entities; keep canonical dependency/justification records sufficient for explanation/replay while ordinary materialized consequences remain rebuildable caches. See upstream prolog-rlm#392 updated Machine Spirit #398B section.

#398B storage consequence: the append-only semantic ledger should not flatten `context/theory`, existential witness, or justification structures into generic assertion metadata. Index current/history views by theory/context/profile; preserve source assertions independently from domain-theory promotions; persist witness scope/origin without promoting labelled unknowns to global entities; keep canonical dependency/justification records sufficient for explanation/replay while ordinary materialized consequences remain rebuildable caches. See upstream `prolog-rlm#392` updated Machine Spirit #398B section.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit depth-1 storage consequences (#398C/#398D)

The general semantic ledger needs to persist the semantic theory graph while keeping compiled execution state rebuildable.

Persist canonically

source/context/theory identities
theory export signatures
vocabulary/schema versions
typed theory edges/import/bridge/mapping contracts
mapping status/provenance
canonical assertions/rules/procedures/events/witnesses
lifecycle events and promoted derivation receipts where policy requires them

Keep rebuildable/materialized

Horn/Datalog closures
tabled answers
transitive/current views
query-specific deductive plans
magic/demand-transformed rule plans
constraint solver caches
relevance/query indexes
ordinary intermediate proof structures

Materialized state should be keyed to the semantic/theory generations it depends on and report freshness. Appending/retracting/superseding a theory mapping, bridge, assertion or rule invalidates dependent derived state without mutating historical semantic records.

A mapping/alignment change must also remain historical knowledge: do not rewrite old source records into the latest ontology vocabulary.

This directly supports replay of old LLM logs/web/news/wiki semantic projections under the theory/schema version with which they were originally interpreted.

Refs upstream prolog-rlm #397/#398/#392/#394/#396.

## Machine Spirit depth-1 storage consequences (#398C/#398D) The general semantic ledger needs to persist the *semantic theory graph* while keeping compiled execution state rebuildable. ### Persist canonically ```text source/context/theory identities theory export signatures vocabulary/schema versions typed theory edges/import/bridge/mapping contracts mapping status/provenance canonical assertions/rules/procedures/events/witnesses lifecycle events and promoted derivation receipts where policy requires them ``` ### Keep rebuildable/materialized ```text Horn/Datalog closures tabled answers transitive/current views query-specific deductive plans magic/demand-transformed rule plans constraint solver caches relevance/query indexes ordinary intermediate proof structures ``` Materialized state should be keyed to the semantic/theory generations it depends on and report freshness. Appending/retracting/superseding a theory mapping, bridge, assertion or rule invalidates dependent derived state without mutating historical semantic records. A mapping/alignment change must also remain historical knowledge: do not rewrite old source records into the latest ontology vocabulary. This directly supports replay of old LLM logs/web/news/wiki semantic projections under the theory/schema version with which they were originally interpreted. Refs upstream prolog-rlm #397/#398/#392/#394/#396.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit #399B ledger consequence — preserve semantic force, not just proposition payload

The append-only ledger must be able to retain exported semantic records whose proposition payloads may look similar while their linguistic force and derivation are materially different.

Examples that must remain distinguishable forever:

asserted(P)
presupposed_by(Source,P)
pragmatic_candidate(P)
reported_content(Speaker,P)
generic(Pattern)
habitual(Pattern)
reconstructed_from_ellipsis(P)
metonymically_coerced(P)

Corrections/new compiler interpretations append new projection lineage/supersession rather than mutating older records. Current/query views may choose which classes participate, but the ledger cannot flatten them into one fact(P, confidence) representation. In particular, pragmatic candidates and generics must never silently become ordinary world facts merely because they were durably stored.

## Machine Spirit #399B ledger consequence — preserve semantic force, not just proposition payload The append-only ledger must be able to retain exported semantic records whose proposition payloads may look similar while their **linguistic force and derivation are materially different**. Examples that must remain distinguishable forever: ```text asserted(P) presupposed_by(Source,P) pragmatic_candidate(P) reported_content(Speaker,P) generic(Pattern) habitual(Pattern) reconstructed_from_ellipsis(P) metonymically_coerced(P) ``` Corrections/new compiler interpretations append new projection lineage/supersession rather than mutating older records. Current/query views may choose which classes participate, but the ledger cannot flatten them into one `fact(P, confidence)` representation. In particular, pragmatic candidates and generics must never silently become ordinary world facts merely because they were durably stored.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

MACHINE-SPIRIT #400B storage handoff — typed defeat, dependence, bitemporal epistemics

Depth 3B refines the #400A ESG persistence contract. Persist enough immutable history to rebuild, when applicable:

typed epistemic attacks: rebut, undermine, undercut, deterministic invalidate, with target refs and provenance;
typed/partial preference assertions and named policy/version refs rather than one global priority score;
evidence-origin/source-dependence links such as copied_from, syndicates, derived_from, shared_origin, plus evidence and certainty/status for the dependence claim;
distinct valid time and transaction/knowledge time, so backdated corrections change current belief about a past interval without rewriting what Machine Spirit knew before the correction arrived;
revision-state/operator lineage where a promoted iterated-revision projection needs replayability;
support/attack dependency structure sufficient to detect self-support/cycles while ordinary SCC/materialized acceptance caches remain rebuildable.

Hard invariants:

undercut(J->P) != support(not(P))
source copy != independent corroboration
suspected dependence != proven dependence
backdated correction != historical rewrite
valid_at(T) != known_at(T)
resource-limited evaluation != epistemic uncertainty

Ten copied/syndicated articles may remain ten immutable assertion episodes while exposing one or a few independent evidence-origin components under an explicit dependence policy.

Full B research and adversarial fixtures are on prolog-rlm#400. #400C/#400D still own cross-profile composition and execution/scaling details.

## MACHINE-SPIRIT #400B storage handoff — typed defeat, dependence, bitemporal epistemics Depth 3B refines the #400A ESG persistence contract. Persist enough immutable history to rebuild, when applicable: - typed epistemic attacks: `rebut`, `undermine`, `undercut`, deterministic `invalidate`, with target refs and provenance; - typed/partial preference assertions and named policy/version refs rather than one global priority score; - evidence-origin/source-dependence links such as `copied_from`, `syndicates`, `derived_from`, `shared_origin`, plus evidence and certainty/status for the dependence claim; - distinct **valid time** and **transaction/knowledge time**, so backdated corrections change current belief about a past interval without rewriting what Machine Spirit knew before the correction arrived; - revision-state/operator lineage where a promoted iterated-revision projection needs replayability; - support/attack dependency structure sufficient to detect self-support/cycles while ordinary SCC/materialized acceptance caches remain rebuildable. Hard invariants: ```text undercut(J->P) != support(not(P)) source copy != independent corroboration suspected dependence != proven dependence backdated correction != historical rewrite valid_at(T) != known_at(T) resource-limited evaluation != epistemic uncertainty ``` Ten copied/syndicated articles may remain ten immutable assertion episodes while exposing one or a few independent evidence-origin components under an explicit dependence policy. Full B research and adversarial fixtures are on prolog-rlm#400. #400C/#400D still own cross-profile composition and execution/scaling details.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

MACHINE-SPIRIT #400C durability handoff — federated epistemic provenance

Depth 3C adds a durable interoperability requirement: Symbolic Memory must preserve enough information to replay cross-theory epistemic federation without ever materializing one merged global truth graph.

Persist/reconstruct, where promoted/durable:

epistemic bridge ID/version/status;
local source theory + local epistemic profile/version;
exported local support state / acceptance / evaluation-completeness / model-scope receipt;
recipient import mode (evidence_only, skeptical-warrant-as-support, credulous candidate, attributed acceptance, revision input, argument import, inert reference, etc.);
bridge compatibility/loss classification (exact_preserving, sound_but_incomplete, lossy_declared, status_only, noncomposable, unknown_compatibility);
closure/negation contract when any absence-sensitive reasoning crosses a boundary;
priority/default lifting/admission policy when present;
source/evidence-dependence lineage across theory boundaries;
identity/ontology mapping assumptions + provenance + lifecycle;
federation profile/version for promoted cross-theory warrant;
bitemporal bridge/mapping/profile state so known_at(T) replays the mappings and bridge policies actually available at T;
cross-context attack/defeat type where promoted.

Hard invariants:

context boundary != independent evidence reset
local NAF/default absence != durable explicit negative evidence
credulous export != skeptical fact
identity mapping != destructive proposition merge
bridge revision projection != canonical history rewrite
noncomposable != dropped silently

Ordinary local/federated warrant caches, bridge closures, MCS equilibria, model products, and argumentation acceptance sets remain rebuildable derived state unless a promoted result requires a replay receipt.

Acceptance should include: two memories derived from one original source remain one independent evidence component after federation; identity-induced conflict can be removed by changing the identity assumption without deleting either original support; historic known_at uses historic mapping/bridge state; local no-equilibrium/failure does not corrupt unrelated ledger state.

## MACHINE-SPIRIT #400C durability handoff — federated epistemic provenance Depth 3C adds a durable interoperability requirement: Symbolic Memory must preserve enough information to replay **cross-theory epistemic federation** without ever materializing one merged global truth graph. Persist/reconstruct, where promoted/durable: - epistemic bridge ID/version/status; - local source theory + local epistemic profile/version; - exported local support state / acceptance / evaluation-completeness / model-scope receipt; - recipient import mode (`evidence_only`, skeptical-warrant-as-support, credulous candidate, attributed acceptance, revision input, argument import, inert reference, etc.); - bridge compatibility/loss classification (`exact_preserving`, `sound_but_incomplete`, `lossy_declared`, `status_only`, `noncomposable`, `unknown_compatibility`); - closure/negation contract when any absence-sensitive reasoning crosses a boundary; - priority/default lifting/admission policy when present; - source/evidence-dependence lineage **across** theory boundaries; - identity/ontology mapping assumptions + provenance + lifecycle; - federation profile/version for promoted cross-theory warrant; - bitemporal bridge/mapping/profile state so `known_at(T)` replays the mappings and bridge policies actually available at T; - cross-context attack/defeat type where promoted. Hard invariants: ```text context boundary != independent evidence reset local NAF/default absence != durable explicit negative evidence credulous export != skeptical fact identity mapping != destructive proposition merge bridge revision projection != canonical history rewrite noncomposable != dropped silently ``` Ordinary local/federated warrant caches, bridge closures, MCS equilibria, model products, and argumentation acceptance sets remain rebuildable derived state unless a promoted result requires a replay receipt. Acceptance should include: two memories derived from one original source remain one independent evidence component after federation; identity-induced conflict can be removed by changing the identity assumption without deleting either original support; historic `known_at` uses historic mapping/bridge state; local no-equilibrium/failure does not corrupt unrelated ledger state.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

MACHINE-SPIRIT #400D handoff — derived epistemic cache/frontier requirements

Depth 3D makes Symbolic Memory the durable source for canonical epistemic history while keeping federation execution state rebuildable.

Persist enough immutable lineage/index data to rebuild query-local epistemic frontiers and invalidate derived state by dependency:

evidence/lifecycle -> local epistemic summaries
local export -> bridge application
bridge/mapping/identity/dependence/closure contract -> SCC result
SCC result -> downstream query/warrant cache
profile/translator version -> summaries created under it

Do not persist one global model product or mandatory ATMS environment closure.

Derived/cache objects may include:

local status-bearing projected interface summaries;
query-local bridge SCC results;
factorized interface-choice/nogood graphs;
WFS/ASP/argumentation solver caches;
ordinary warrant/explanation synopsis caches.

They are disposable unless a promoted decision/action requires a replay receipt.

Cache/frontier identity must include theory/profile/evidence frontier, valid_at + known_at context, interface signature, identity/dependence generation, bridge/mapping/closure generation and translator/backend contract version where material.

Propagation-stop invariant: if a changed local theory recomputes to the same projected bridge-interface digest, downstream derived state remains valid; internal churn must not force global invalidation.

Historical replay must remain possible after later corrections/mappings/profile changes. known_at(Tpast) uses the historically available evidence/mapping/profile state, not today's cache.

Add acceptance for cold/warm/rebuilt and incremental/full differential equivalence, identity/bridge/profile scoped invalidation, same-interface propagation stop, and no destructive cache writes to canonical ledger history.

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

## MACHINE-SPIRIT #400D handoff — derived epistemic cache/frontier requirements Depth 3D makes Symbolic Memory the durable source for canonical epistemic history while keeping federation execution state rebuildable. Persist enough immutable lineage/index data to rebuild query-local epistemic frontiers and invalidate derived state by dependency: ```text evidence/lifecycle -> local epistemic summaries local export -> bridge application bridge/mapping/identity/dependence/closure contract -> SCC result SCC result -> downstream query/warrant cache profile/translator version -> summaries created under it ``` Do **not** persist one global model product or mandatory ATMS environment closure. Derived/cache objects may include: - local status-bearing projected interface summaries; - query-local bridge SCC results; - factorized interface-choice/nogood graphs; - WFS/ASP/argumentation solver caches; - ordinary warrant/explanation synopsis caches. They are disposable unless a promoted decision/action requires a replay receipt. Cache/frontier identity must include theory/profile/evidence frontier, `valid_at` + `known_at` context, interface signature, identity/dependence generation, bridge/mapping/closure generation and translator/backend contract version where material. **Propagation-stop invariant:** if a changed local theory recomputes to the same projected bridge-interface digest, downstream derived state remains valid; internal churn must not force global invalidation. Historical replay must remain possible after later corrections/mappings/profile changes. `known_at(Tpast)` uses the historically available evidence/mapping/profile state, not today's cache. Add acceptance for cold/warm/rebuilt and incremental/full differential equivalence, identity/bridge/profile scoped invalidation, same-interface propagation stop, and no destructive cache writes to canonical ledger history. Full design and sources are on prolog-rlm#400D.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit #401A procedural-memory handoff

Depth 4A distinguishes reusable procedural knowledge from episodic procedural experience and from run-time control state.

Durable ledger should be able to preserve/version, where upstream #392 exposes them:

admitted procedure/method knowledge + provenance;
task/network structure, partial order, conditions, expected effects, failure/recovery refs;
cases/precedents with situation signature, method/plan used, adaptations, actual outcome and VERIFY receipt;
diagnosis/repair receipts that are materially retained as experience;
procedure supersession/version relationships;
execution receipts linking method version -> plan instance -> expert/effect observations -> VERIFY outcome.

Hard storage distinctions:

procedure/method != run instance
expected postcondition != observed result
case succeeded once != trusted generalized rule
repair worked != diagnosis proven

Do not make transient Rete match memories, coordination-board agendas, planner frontier nodes, ordinary decomposition caches or active diagnostic search state canonical semantic truth; those are rebuildable/run-local unless a replay/promotion receipt specifically requires retention.

Verified cases can later support CBR retrieval and #404 candidate procedure/rule promotion, but Symbolic Memory must not itself turn successful cases into trusted procedural policy. Full design and A1-A20 fixtures are on prolog-rlm#401.

## Machine Spirit #401A procedural-memory handoff Depth 4A distinguishes **reusable procedural knowledge** from **episodic procedural experience** and from run-time control state. Durable ledger should be able to preserve/version, where upstream #392 exposes them: - admitted procedure/method knowledge + provenance; - task/network structure, partial order, conditions, expected effects, failure/recovery refs; - cases/precedents with situation signature, method/plan used, adaptations, actual outcome and VERIFY receipt; - diagnosis/repair receipts that are materially retained as experience; - procedure supersession/version relationships; - execution receipts linking method version -> plan instance -> expert/effect observations -> VERIFY outcome. Hard storage distinctions: ```text procedure/method != run instance expected postcondition != observed result case succeeded once != trusted generalized rule repair worked != diagnosis proven ``` Do **not** make transient Rete match memories, coordination-board agendas, planner frontier nodes, ordinary decomposition caches or active diagnostic search state canonical semantic truth; those are rebuildable/run-local unless a replay/promotion receipt specifically requires retention. Verified cases can later support CBR retrieval and #404 candidate procedure/rule promotion, but Symbolic Memory must not itself turn successful cases into trusted procedural policy. Full design and A1-A20 fixtures are on prolog-rlm#401.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit #401B downstream durability handoff — procedural policy history

Depth 4B refines durable procedural memory. Preserve append-only history sufficient to reconstruct why a monitored procedure was selected, how it behaved, and whether later evidence suggests a method defect, without storing one mutable current best procedure truth value.

Durable records should support, where material:

immutable method/policy version + provenance;
epistemic guard/assumption profile references;
declared observation gates/branch policy and nondeterministic guarantee/fairness assumptions;
temporal/resource contract versions;
verified run/case receipt;
actual observed primitive outcomes distinct from predicted outcomes;
exogenous events relevant to the run;
plan-instance repair vs method reselection decisions;
inert reusable-method counterexample/defect/supersession candidates;
CBR source-case refs, semantic differences, compatibility/adaptation rationale and contraindications;
authority/capability context sufficient to prevent a privileged historical case from transferring privilege to a new run.

Ordinary live execution frontiers, schedules, guard-match caches and current resource leases are run-local/rebuildable unless retained in a final execution/audit receipt.

Hard durability rules:

failed plan instance != invalid method
successful repair != proven diagnosis
case similarity != method applicability
case authority != current authority
method-defect candidate != trusted method mutation

Promotion/supersession of trusted procedural knowledge remains a later governed operation (#404 upstream), never an automatic consequence of case retention.

Full rationale/sources and B1-B28 fixtures: lost-rob0t/prolog-rlm#401 Depth-4B.

## Machine Spirit #401B downstream durability handoff — procedural policy history Depth 4B refines durable procedural memory. Preserve append-only history sufficient to reconstruct why a monitored procedure was selected, how it behaved, and whether later evidence suggests a method defect, without storing one mutable `current best procedure` truth value. Durable records should support, where material: - immutable method/policy version + provenance; - epistemic guard/assumption profile references; - declared observation gates/branch policy and nondeterministic guarantee/fairness assumptions; - temporal/resource contract versions; - verified run/case receipt; - actual observed primitive outcomes distinct from predicted outcomes; - exogenous events relevant to the run; - plan-instance repair vs method reselection decisions; - inert reusable-method counterexample/defect/supersession candidates; - CBR source-case refs, semantic differences, compatibility/adaptation rationale and contraindications; - authority/capability context sufficient to prevent a privileged historical case from transferring privilege to a new run. Ordinary live execution frontiers, schedules, guard-match caches and current resource leases are run-local/rebuildable unless retained in a final execution/audit receipt. Hard durability rules: ```text failed plan instance != invalid method successful repair != proven diagnosis case similarity != method applicability case authority != current authority method-defect candidate != trusted method mutation ``` Promotion/supersession of trusted procedural knowledge remains a later governed operation (#404 upstream), never an automatic consequence of case retention. Full rationale/sources and B1-B28 fixtures: lost-rob0t/prolog-rlm#401 Depth-4B.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit #401C handoff — durable procedural interoperability lineage

Depth 4C (prolog-rlm#401) selects PIFF/PSC for heterogeneous procedure federation. The append-only semantic ledger should preserve enough history to reconstruct procedure interoperability without rewriting source-native meaning:

native procedure artifact ID + formalism/version + source provenance;
Procedural Semantic Contract version/fingerprint;
procedure bridge/mapping candidate/admission/rejection + relation (exact_on, refine/abstract/project/lossy/incompatible/etc.);
mapping assumptions, preserved/lost dimensions, source/projected guarantee and fairness receipt;
host action-binding identity/version where retained for replay, clearly separate from semantic procedure identity and live authority;
host-specific projection/execution receipt + exact artifact/PSC/bridge/binding versions;
observed execution outcome/case provenance.

Hard invariant: mapping/adapter upgrades create new projection lineage; old executions still resolve the exact historical mapping. Concrete current capability/authority remains runtime/host state, not durable memory truth. See prolog-rlm#401 C1-C28; #401D will settle cache/incremental execution details.

## Machine Spirit #401C handoff — durable procedural interoperability lineage Depth 4C (prolog-rlm#401) selects PIFF/PSC for heterogeneous procedure federation. The append-only semantic ledger should preserve enough history to reconstruct procedure interoperability without rewriting source-native meaning: - native procedure artifact ID + formalism/version + source provenance; - Procedural Semantic Contract version/fingerprint; - procedure bridge/mapping candidate/admission/rejection + relation (`exact_on`, refine/abstract/project/lossy/incompatible/etc.); - mapping assumptions, preserved/lost dimensions, source/projected guarantee and fairness receipt; - host action-binding identity/version **where retained for replay**, clearly separate from semantic procedure identity and live authority; - host-specific projection/execution receipt + exact artifact/PSC/bridge/binding versions; - observed execution outcome/case provenance. Hard invariant: mapping/adapter upgrades create new projection lineage; old executions still resolve the exact historical mapping. Concrete current capability/authority remains runtime/host state, not durable memory truth. See prolog-rlm#401 C1-C28; #401D will settle cache/incremental execution details.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Upstream Machine Spirit #401D handoff — procedural execution/replay durability

Prolog-RLM Depth 4 is now design-complete with DPEC — Demand-Driven Procedural Execution & Conformance Graph.

Durability boundary for this ledger:

Durable / append-only where material

native procedure/method artifact + version identity;
PSC version and PIFF bridge/mapping versions used by a projection;
source/native procedure provenance;
host-projection receipt identity/fingerprint when a run/case/result materially depends on it;
observed execution/effect/monitor events needed to explain a completed/failed case;
plan/replan generation lineage;
expert/binding contract versions referenced by a historical run (not executable closures);
cancellation/external-effect outcome lineage;
final VERIFY evidence refs;
compact conformance/replay receipt for retained cases where exact replay is promised.

Rebuildable derived state by default

DPEC method/applicability indexes;
Rete/tabling caches;
packed decomposition forests/search frontiers;
current host-projection caches;
monitor indexes/current monitor state once run history has durable events;
temporal/resource solver caches;
native reference-validator caches;
query-local diagnosis hitting-set structures;
CBR similarity indexes.

Hard invariant:

persisted procedure semantics != persisted executable authority
historical host projection    != currently admitted host projection
historical authority decision != current authority

Changing PSC/bridge/binding/expert versions creates new projection/run lineage and never rewrites prior history. If DPEC recomputation yields the same selected observable-interface digest, downstream derived invalidation may stop, but original version lineage remains inspectable.

Full #401D research and D1-D34 conformance fixtures are on lost-rob0t/prolog-rlm#401.

## Upstream Machine Spirit #401D handoff — procedural execution/replay durability Prolog-RLM Depth 4 is now design-complete with **DPEC — Demand-Driven Procedural Execution & Conformance Graph**. Durability boundary for this ledger: ### Durable / append-only where material - native procedure/method artifact + version identity; - PSC version and PIFF bridge/mapping versions used by a projection; - source/native procedure provenance; - host-projection *receipt identity/fingerprint* when a run/case/result materially depends on it; - observed execution/effect/monitor events needed to explain a completed/failed case; - plan/replan generation lineage; - expert/binding contract versions referenced by a historical run (not executable closures); - cancellation/external-effect outcome lineage; - final VERIFY evidence refs; - compact conformance/replay receipt for retained cases where exact replay is promised. ### Rebuildable derived state by default - DPEC method/applicability indexes; - Rete/tabling caches; - packed decomposition forests/search frontiers; - current host-projection caches; - monitor indexes/current monitor state once run history has durable events; - temporal/resource solver caches; - native reference-validator caches; - query-local diagnosis hitting-set structures; - CBR similarity indexes. Hard invariant: ```text persisted procedure semantics != persisted executable authority historical host projection != currently admitted host projection historical authority decision != current authority ``` Changing PSC/bridge/binding/expert versions creates new projection/run lineage and never rewrites prior history. If DPEC recomputation yields the same selected observable-interface digest, downstream derived invalidation may stop, but original version lineage remains inspectable. Full #401D research and D1-D34 conformance fixtures are on lost-rob0t/prolog-rlm#401.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit #402A durable-ledger handoff

Depth 5A is now complete on prolog-rlm#402 and strengthens this issue's storage contract without changing repository ownership.

Adopt the BSLPF foundation: immutable/versioned semantic objects + an authoritative append-only bitemporal semantic ledger + rebuildable/materialized projections.

Concrete #6 deltas for later implementation/design reconciliation:

semantic-object identity, assertion/support episode identity, ledger-event identity, commit/batch identity and view-row identity are distinct;
durable commits are atomic ordered batches with stable logical frontier/sequence, commit identity, idempotency key + canonical batch digest, and optional expected-frontier/CAS precondition;
known_at / transaction time comes from immutable commit history; world valid_at remains an independent semantic time axis;
crash after durable commit but before view/index update is recovered by replay from a projection checkpoint/frontier;
projection checkpoints are keyed by ledger frontier and projection/schema/vocabulary/reasoning/mapping versions; stale views cannot satisfy current VERIFY/action admission;
current support/warrant/entity/procedure/retrieval/provenance indexes remain rebuildable derived state even when persisted for speed;
identity/ontology/schema evolution appends mapping/supersession lifecycle; it never rewrites original semantic IDs/history;
same idempotency key + same batch digest is one logical commit; same key + different digest is a hard conflict;
exact source/compiler/logical/lifecycle/profile/runtime provenance classes remain distinct;
the current whole-snapshot symbolic_memory_storage.pl implementation is a bootstrap backend only, not the semantic source-of-truth architecture.

This intentionally preserves the older SYMBOLIC-MEMORY-RESEARCH-006-persistence-and-concurrency.org as historical backend/bootstrap research while superseding its earlier position that complete replayable semantic event history could be optional at the world-model contract layer.

Refs: lost-rob0t/prolog-rlm#402 (#402A), #392, #400, #401, and this issue's parent #4.

## Machine Spirit #402A durable-ledger handoff Depth 5A is now complete on `prolog-rlm#402` and strengthens this issue's storage contract without changing repository ownership. Adopt the **BSLPF** foundation: immutable/versioned semantic objects + an **authoritative append-only bitemporal semantic ledger** + rebuildable/materialized projections. Concrete #6 deltas for later implementation/design reconciliation: - semantic-object identity, assertion/support episode identity, ledger-event identity, commit/batch identity and view-row identity are distinct; - durable commits are atomic ordered batches with stable logical frontier/sequence, commit identity, idempotency key + canonical batch digest, and optional expected-frontier/CAS precondition; - `known_at` / transaction time comes from immutable commit history; world `valid_at` remains an independent semantic time axis; - crash after durable commit but before view/index update is recovered by replay from a projection checkpoint/frontier; - projection checkpoints are keyed by ledger frontier **and** projection/schema/vocabulary/reasoning/mapping versions; stale views cannot satisfy current VERIFY/action admission; - current support/warrant/entity/procedure/retrieval/provenance indexes remain rebuildable derived state even when persisted for speed; - identity/ontology/schema evolution appends mapping/supersession lifecycle; it never rewrites original semantic IDs/history; - same idempotency key + same batch digest is one logical commit; same key + different digest is a hard conflict; - exact source/compiler/logical/lifecycle/profile/runtime provenance classes remain distinct; - the current whole-snapshot `symbolic_memory_storage.pl` implementation is a **bootstrap backend only**, not the semantic source-of-truth architecture. This intentionally preserves the older `SYMBOLIC-MEMORY-RESEARCH-006-persistence-and-concurrency.org` as historical backend/bootstrap research while superseding its earlier position that complete replayable semantic event history could be optional at the world-model contract layer. Refs: lost-rob0t/prolog-rlm#402 (#402A), #392, #400, #401, and this issue's parent #4.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

MACHINE-SPIRIT #402B canonical durability refinement — GRIE

Depth 5B refines this ledger contract with GRIE — Governed Redaction & Integrity Envelope. Treat this as a hard clarification of append-only:

append-only logical commitment/history
    != permanent recoverability of every payload byte

Required durable distinctions now include:

correction != withdrawal != invalidation != quarantine
redaction != semantic retraction/negation
archive/access_revoke != physical sanitization
current admissibility != historical commitment != payload availability
ledger receipt != issuer authentication != local authorization

#6 owns durable lifecycle/admission/integrity state sufficient to represent: retention/encryption-scoped payload envelopes; authenticated admission receipts; redaction/sanitization completion (complete|partial|pending|blocked|failed); replay capability (exact|semantic_only|commitment_only|redacted|corrupt|unavailable); versioned schema/upcaster interpretation receipts; integrity checkpoints/recovery lineage; and duplicate classes distinguishing transport retry from independent/correlated evidence.

A redaction receipt must not copy sensitive plaintext into the audit trail, and global content hashes/dedup may not leak equality across authority-isolated namespaces. A corrupted/stale projection is disposable and may never repair/replace canonical ledger authority. Canonical segment corruption is an explicit integrity failure requiring authenticated recovery lineage.

Physical erasure/sanitization remains a backend capability: a tombstone is never proof bytes are gone from replicas/backups/media. See upstream prolog-rlm#402B for the full primary research and B1-B30 conformance fixtures.

## MACHINE-SPIRIT #402B canonical durability refinement — GRIE Depth 5B refines this ledger contract with **GRIE — Governed Redaction & Integrity Envelope**. Treat this as a hard clarification of `append-only`: ```text append-only logical commitment/history != permanent recoverability of every payload byte ``` Required durable distinctions now include: ```text correction != withdrawal != invalidation != quarantine redaction != semantic retraction/negation archive/access_revoke != physical sanitization current admissibility != historical commitment != payload availability ledger receipt != issuer authentication != local authorization ``` #6 owns durable lifecycle/admission/integrity state sufficient to represent: retention/encryption-scoped payload envelopes; authenticated admission receipts; redaction/sanitization completion (`complete|partial|pending|blocked|failed`); replay capability (`exact|semantic_only|commitment_only|redacted|corrupt|unavailable`); versioned schema/upcaster interpretation receipts; integrity checkpoints/recovery lineage; and duplicate classes distinguishing transport retry from independent/correlated evidence. A redaction receipt must not copy sensitive plaintext into the audit trail, and global content hashes/dedup may not leak equality across authority-isolated namespaces. A corrupted/stale projection is disposable and may never repair/replace canonical ledger authority. Canonical segment corruption is an explicit integrity failure requiring authenticated recovery lineage. Physical erasure/sanitization remains a backend capability: a tombstone is never proof bytes are gone from replicas/backups/media. See upstream prolog-rlm#402B for the full primary research and B1-B30 conformance fixtures.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

MACHINE-SPIRIT #402C handoff — Causal Ledger Federation Mesh

Depth 5C requires #6's append-only semantic ledger contract to distinguish a logical ledger identity from physical replicas/backups/mirrors and from genuinely independent ledgers.

Add durable/reconstructible support for typed records equivalent to:

sem_ledger_profile(Ledger, AuthorityDomain, Namespace,
                   OrderProfile, RetentionProfile, IntegrityProfile,
                   RedactionProfile, ReplayProfile, Meta).
sem_federation_link(Link, LocalLedger, RemoteLedger, MappingBundle,
                    AdmissionPolicy, IntegrityContract,
                    RetentionContract, RedactionContract,
                    Status, Provenance).
sem_cross_ledger_edge(Edge, FromEvent, ToEvent,
                      causal_basis(Basis), EvidenceRefs, Status).
sem_federation_cut(Cut, MemberFrontiers, LinkVersions, MappingVersions,
                   AdmissionGeneration, CreatedAt, Completeness).
sem_redaction_obligation(Id, OriginLedger, OriginEvent, TargetLedger,
                         Scope, RequiredAction, PolicyRef, Status, Receipts).

Hard requirements:

local commit order != cross-ledger causality; wall-clock equality/order is insufficient;
replica/backup/mirror copies of one logical ledger never count as independent semantic corroboration;
divergent histories claiming one logical-ledger identity produce an integrity/fork condition, not last-writer-wins;
remote redaction is tracked per target ledger; sending a notice cannot claim physical remote sanitization;
federation cuts bind exact per-ledger frontiers/checkpoints plus link/mapping/admission/redaction/integrity generations;
current/federated indexes remain rebuildable projections, not cross-ledger canonical truth;
remote issuer/principal identity remains provenance data and does not acquire local authority.

CRDT/consensus replication may implement an explicit replica group but is not the semantic merge rule for independent ledgers. Full research + C1-C34 fixtures: prolog-rlm#402. Refs #4 #7 #9 #10.

### MACHINE-SPIRIT #402C handoff — Causal Ledger Federation Mesh Depth 5C requires #6's append-only semantic ledger contract to distinguish a **logical ledger identity** from physical replicas/backups/mirrors and from genuinely independent ledgers. Add durable/reconstructible support for typed records equivalent to: ```prolog sem_ledger_profile(Ledger, AuthorityDomain, Namespace, OrderProfile, RetentionProfile, IntegrityProfile, RedactionProfile, ReplayProfile, Meta). sem_federation_link(Link, LocalLedger, RemoteLedger, MappingBundle, AdmissionPolicy, IntegrityContract, RetentionContract, RedactionContract, Status, Provenance). sem_cross_ledger_edge(Edge, FromEvent, ToEvent, causal_basis(Basis), EvidenceRefs, Status). sem_federation_cut(Cut, MemberFrontiers, LinkVersions, MappingVersions, AdmissionGeneration, CreatedAt, Completeness). sem_redaction_obligation(Id, OriginLedger, OriginEvent, TargetLedger, Scope, RequiredAction, PolicyRef, Status, Receipts). ``` Hard requirements: - local commit order != cross-ledger causality; wall-clock equality/order is insufficient; - replica/backup/mirror copies of one logical ledger never count as independent semantic corroboration; - divergent histories claiming one logical-ledger identity produce an integrity/fork condition, not last-writer-wins; - remote redaction is tracked per target ledger; sending a notice cannot claim physical remote sanitization; - federation cuts bind exact per-ledger frontiers/checkpoints plus link/mapping/admission/redaction/integrity generations; - current/federated indexes remain rebuildable projections, not cross-ledger canonical truth; - remote issuer/principal identity remains provenance data and does not acquire local authority. CRDT/consensus replication may implement an explicit replica group but is not the semantic merge rule for independent ledgers. Full research + C1-C34 fixtures: prolog-rlm#402. Refs #4 #7 #9 #10.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit #402D handoff: adopt SAFIRE as the computational durability target beneath this issue. Canonical ledger state should become segmented/append-only with dense per-ledger sequence, sparse/hierarchical federation cuts, exact projection dependency generations, atomic projection checkpoints, explicit freshness, typed replica profiles, bounded causal traversal, anti-entropy inside replica/mirror relationships only, and cold/warm/rebuilt/incremental differential conformance. The current whole-snapshot rewrite backend remains bootstrap-only. See prolog-rlm#402 D1-D40. Hard boundary: projections never repair canonical ledger authority; Bloom/index hits never become semantic evidence; replicas never manufacture corroboration.

Machine Spirit #402D handoff: adopt **SAFIRE** as the computational durability target beneath this issue. Canonical ledger state should become segmented/append-only with dense per-ledger sequence, sparse/hierarchical federation cuts, exact projection dependency generations, atomic projection checkpoints, explicit freshness, typed replica profiles, bounded causal traversal, anti-entropy inside replica/mirror relationships only, and cold/warm/rebuilt/incremental differential conformance. The current whole-snapshot rewrite backend remains bootstrap-only. See prolog-rlm#402 D1-D40. Hard boundary: projections never repair canonical ledger authority; Bloom/index hits never become semantic evidence; replicas never manufacture corroboration.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Upstream Machine Spirit #403B handoff — retrieval inputs from durable provenance/dependence

prolog-rlm#403B/OATH-TAPS depends on the durable world model preserving enough canonical metadata for adversarially honest retrieval:

exact record/cut membership so lexical/vector/graph hits can be revalidated before reasoning;
source/origin dependence components so mirrors/copies cannot manufacture corroboration;
lifecycle/validity/supersession/invalidation lineage for freshness obligations;
trust/authentication/provenance as separate dimensions rather than one retrieval score;
canonical metadata fields must not be writable by source prose (trusted, pinned, priority, etc.).

Indexes remain rebuildable sensors. An index hit absent from the exact admitted ledger cut is an index inconsistency, not evidence. Refs lost-rob0t/prolog-rlm#403 #400 #402 and symbolic-memory#7/#10.

## Upstream Machine Spirit #403B handoff — retrieval inputs from durable provenance/dependence `prolog-rlm#403B`/OATH-TAPS depends on the durable world model preserving enough canonical metadata for adversarially honest retrieval: - exact record/cut membership so lexical/vector/graph hits can be revalidated before reasoning; - source/origin dependence components so mirrors/copies cannot manufacture corroboration; - lifecycle/validity/supersession/invalidation lineage for freshness obligations; - trust/authentication/provenance as separate dimensions rather than one retrieval score; - canonical metadata fields must not be writable by source prose (`trusted`, `pinned`, `priority`, etc.). Indexes remain rebuildable sensors. An index hit absent from the exact admitted ledger cut is an index inconsistency, not evidence. Refs lost-rob0t/prolog-rlm#403 #400 #402 and symbolic-memory#7/#10.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

MACHINE-SPIRIT #404A handoff — EPOCH learning ledger

Depth 7A requires the durable world model to persist learning history without collapsing experience into trusted rules.

Canonical durable classes should be able to represent:

reasoning/strategy experience envelope
candidate lesson/rule/method/control strategy
promotion-evaluation receipt
admit / suspend / deprecate / supersede / retract lifecycle
exact learned-object version used by historical runs
counterexample/regression evidence

Experience must bind to exact semantic/federation cut, epistemic profile, procedure/strategy/expert versions, observations, VERIFY/Review/diagnosis/resource/provider receipts and replay lineage. Derived strategy-performance aggregates/indexes may remain rebuildable; the source episodes and material promotion lifecycle are durable.

Hard distinction:

experience != candidate != admitted learned knowledge != host binding/capability/authority

Correlated/mirrored episodes must remain distinguishable for promotion-independence accounting. Ref upstream prolog-rlm#404A EPOCH.

## MACHINE-SPIRIT #404A handoff — EPOCH learning ledger Depth 7A requires the durable world model to persist **learning history without collapsing experience into trusted rules**. Canonical durable classes should be able to represent: ```text reasoning/strategy experience envelope candidate lesson/rule/method/control strategy promotion-evaluation receipt admit / suspend / deprecate / supersede / retract lifecycle exact learned-object version used by historical runs counterexample/regression evidence ``` Experience must bind to exact semantic/federation cut, epistemic profile, procedure/strategy/expert versions, observations, VERIFY/Review/diagnosis/resource/provider receipts and replay lineage. Derived strategy-performance aggregates/indexes may remain rebuildable; the source episodes and material promotion lifecycle are durable. Hard distinction: ```text experience != candidate != admitted learned knowledge != host binding/capability/authority ``` Correlated/mirrored episodes must remain distinguishable for promotion-independence accounting. Ref upstream prolog-rlm#404A EPOCH.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit #404B / CITADEL-EPOCH durable-learning handoff

The append-only semantic ledger must persist learning/promotion history as first-class semantic/control provenance rather than one mutable confidence field.

Persist/reconstruct, where applicable:

immutable experience IDs and exact execution/VERIFY/diagnosis lineage;
candidate-generation provenance and ancestor candidates;
candidate exposure closure over sources/semantic records/fixtures/metric+verifier outputs/model or learner artifacts;
evidence-use roles (generation, selection, validation, audit, post_admission) plus origin/source-dependence groups;
evaluator, metric, baseline and promotion-policy versions;
contamination/adaptive-holdout findings;
counterexample campaign receipts and bounded coverage;
promotion/hold/quarantine/admission/suspension/deprecation/supersession/retraction events;
drift and calibration epochs/receipts.

History remains append-only/bitemporal: confirmed drift or a new counterexample changes the current admission projection but does not rewrite what was promoted previously or why. Derived aggregate scores/current promotion views remain rebuildable state. Refs prolog-rlm#404B CITADEL-EPOCH.

## Machine Spirit #404B / CITADEL-EPOCH durable-learning handoff The append-only semantic ledger must persist learning/promotion history as first-class semantic/control provenance rather than one mutable `confidence` field. Persist/reconstruct, where applicable: - immutable experience IDs and exact execution/VERIFY/diagnosis lineage; - candidate-generation provenance and ancestor candidates; - **candidate exposure closure** over sources/semantic records/fixtures/metric+verifier outputs/model or learner artifacts; - evidence-use roles (`generation`, `selection`, `validation`, `audit`, `post_admission`) plus origin/source-dependence groups; - evaluator, metric, baseline and promotion-policy versions; - contamination/adaptive-holdout findings; - counterexample campaign receipts and bounded coverage; - promotion/hold/quarantine/admission/suspension/deprecation/supersession/retraction events; - drift and calibration epochs/receipts. History remains append-only/bitemporal: confirmed drift or a new counterexample changes the current admission projection but does not rewrite what was promoted previously or why. Derived aggregate scores/current promotion views remain rebuildable state. Refs prolog-rlm#404B CITADEL-EPOCH.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Machine Spirit #404C handoff — durable learned-artifact federation

PACT-EPOCH requires #6 to persist imported Learning Artifact Contracts (LACs) and their lifecycle append-only, including source/local promotion states side by side, learner kind, source environment/applicability assumptions, evidence-use summaries, source promotion/evaluation attestations, transfer assessments, drift/calibration epochs and opaque origin/dependence claims.

Hard storage invariant:

source admitted != locally admitted
source retracted/suspended != historical deletion
signed attestation != semantic truth

Imported source receipts remain immutable evidence; target promotion creates a distinct local receipt/version. Retraction/drift/mapping changes invalidate affected transfer projections without rewriting source history. Full C design: prolog-rlm#404C.

## Machine Spirit #404C handoff — durable learned-artifact federation PACT-EPOCH requires #6 to persist imported **Learning Artifact Contracts (LACs)** and their lifecycle append-only, including source/local promotion states side by side, learner kind, source environment/applicability assumptions, evidence-use summaries, source promotion/evaluation attestations, transfer assessments, drift/calibration epochs and opaque origin/dependence claims. Hard storage invariant: ```text source admitted != locally admitted source retracted/suspended != historical deletion signed attestation != semantic truth ``` Imported source receipts remain immutable evidence; target promotion creates a distinct local receipt/version. Retraction/drift/mapping changes invalidate affected transfer projections without rewriting source history. Full C design: prolog-rlm#404C.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

Implementation decomposition

#6 remains the canonical semantic-ledger/current-history requirement. It is now executed through:

#23 append-only full-IR commit protocol
 -> #24 bitemporal current/history projections
 -> #25 epistemic support/provenance DAG
 -> #26 general semantic indexes
 -> #37 SAFIRE production backend/frontiers/checkpoints
 -> #38 projection DAG/incremental invalidation
 -> #39 GRIE redaction/integrity lifecycle

#6 should not absorb the full scalable backend or query stack in one PR. Canonical history remains authoritative; all current/support/query indexes are rebuildable derived state.

## Implementation decomposition #6 remains the canonical semantic-ledger/current-history requirement. It is now executed through: ```text #23 append-only full-IR commit protocol -> #24 bitemporal current/history projections -> #25 epistemic support/provenance DAG -> #26 general semantic indexes -> #37 SAFIRE production backend/frontiers/checkpoints -> #38 projection DAG/incremental invalidation -> #39 GRIE redaction/integrity lifecycle ``` #6 should not absorb the full scalable backend or query stack in one PR. Canonical history remains authoritative; all current/support/query indexes are rebuildable derived state.

nsaspy commented

2026-09-08 01:57:19 +00:00

Author

Owner

MACHINE-SPIRIT #404D durable-ledger refinement

Depth 7D extends the append-only ledger contract to learned-state history. Do not add one mutable learned=true/admitted=true property to semantic records.

Persist direct lifecycle/evidence events sufficient to reconstruct:

immutable experiences and candidate versions;
LAC/attestation source identity;
exact EvaluationSnapshot and evaluation evidence;
promotion/hold/quarantine/admission events;
drift/suspension/supersession/retraction;
direct dependence/exposure edges and relevant dependency generations.

Current learned status/exposure closure/validation eligibility remain rebuildable views over canonical history. Historical queries must resolve the exact candidate/policy/verifier/attestation versions active at the requested frontier.

A promotion transition is valid only against its frozen snapshot/dependency generations; a later source withdrawal or mapping correction appends invalidation/suspension rather than rewriting the old decision. Refs prolog-rlm#404D and symbolic-memory#16/#35/#36.

## MACHINE-SPIRIT #404D durable-ledger refinement Depth 7D extends the append-only ledger contract to learned-state history. Do not add one mutable `learned=true/admitted=true` property to semantic records. Persist direct lifecycle/evidence events sufficient to reconstruct: - immutable experiences and candidate versions; - LAC/attestation source identity; - exact EvaluationSnapshot and evaluation evidence; - promotion/hold/quarantine/admission events; - drift/suspension/supersession/retraction; - direct dependence/exposure edges and relevant dependency generations. Current learned status/exposure closure/validation eligibility remain rebuildable views over canonical history. Historical queries must resolve the exact candidate/policy/verifier/attestation versions active at the requested frontier. A promotion transition is valid only against its frozen snapshot/dependency generations; a later source withdrawal or mapping correction appends invalidation/suspension rather than rewriting the old decision. Refs prolog-rlm#404D and symbolic-memory#16/#35/#36.

nsaspy commented

2026-09-08 01:57:20 +00:00

Author

Owner

prolog-rlm#405B AEGIS-MS durable handoff: persist direct influence/control-generation/replay-lineage edges needed to reconstruct end-to-end acceptance history, alongside existing source/compiler/logical/lifecycle/profile provenance. Canonical state should store direct edges/events; transitive influence/acceptance closure remains query-scoped/rebuildable and generation-keyed. Material changes to identity, redaction, integrity, admission/control refs or source lifecycle must invalidate dependent derived acceptance/query artifacts. This is provenance/history storage only: Symbolic Memory does not become an authority policy engine or final verifier.

prolog-rlm#405B AEGIS-MS durable handoff: persist **direct influence/control-generation/replay-lineage edges** needed to reconstruct end-to-end acceptance history, alongside existing source/compiler/logical/lifecycle/profile provenance. Canonical state should store direct edges/events; transitive influence/acceptance closure remains query-scoped/rebuildable and generation-keyed. Material changes to identity, redaction, integrity, admission/control refs or source lifecycle must invalidate dependent derived acceptance/query artifacts. This is provenance/history storage only: Symbolic Memory does not become an authority policy engine or final verifier.

nsaspy commented

2026-09-08 01:57:20 +00:00

Author

Owner

Machine Spirit #405C / CONCORDAT durable-ledger handoff

Cross-domain semantic history must preserve the origin domain + source ledger/frontier + bridge/appraisal generations, not flatten imported records into indistinguishable local assertions.

Persist enough to reconstruct:

foreign assertion/support
 -> source domain/ledger/object/frontier
 -> federation contract + semantic/epistemic mapping
 -> authentication/trust-root generation
 -> target appraisal/import event
 -> local support/current-view consequences

Hard invariants from #405C:

imported evidence != locally originated evidence;
replica/mirror copies != independent corroboration;
foreign verifier/promotion status != target warrant/admission;
hidden provenance => independence unknown;
redaction ACK != verified erasure;
bridge/trust/appraisal generation changes invalidate derived current views without rewriting canonical foreign history.

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

## Machine Spirit #405C / CONCORDAT durable-ledger handoff Cross-domain semantic history must preserve the **origin domain + source ledger/frontier + bridge/appraisal generations**, not flatten imported records into indistinguishable local assertions. Persist enough to reconstruct: ```text foreign assertion/support -> source domain/ledger/object/frontier -> federation contract + semantic/epistemic mapping -> authentication/trust-root generation -> target appraisal/import event -> local support/current-view consequences ``` Hard invariants from #405C: - imported evidence != locally originated evidence; - replica/mirror copies != independent corroboration; - foreign verifier/promotion status != target warrant/admission; - hidden provenance => independence unknown; - redaction ACK != verified erasure; - bridge/trust/appraisal generation changes invalidate derived current views without rewriting canonical foreign history. Canonical research: lost-rob0t/prolog-rlm#405C.

nsaspy referenced this issue

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

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

nsaspy referenced this issue

2026-09-08 01:57:40 +00:00

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

nsaspy referenced this issue

2026-09-08 01:59:55 +00:00

[SLICE 27][CORE-H] Symbolic Memory v1 north-star integration: heterogeneous corpus → durable semantics → bounded zero-model reasoning → learned-history replay #4

Rows
Columns

[semantic-memory] Append-only general semantic ledger: arbitrary IR records, temporal/current views, conflict, supersession, and provenance #43

Goal

Core invariant

Required record identity

Preserve semantic structure

Current/history/world-model views

Entity and relation indexing

Temporal/current semantics

Conflict and plurality

Trust semantics

MACHINE-SPIRIT #400A epistemic ledger requirements

Extension durability

Acceptance

Machine Spirit depth-1 storage consequences (#398C/#398D)

Persist canonically

Keep rebuildable/materialized

Machine Spirit #399B ledger consequence — preserve semantic force, not just proposition payload

MACHINE-SPIRIT #400B storage handoff — typed defeat, dependence, bitemporal epistemics

MACHINE-SPIRIT #400C durability handoff — federated epistemic provenance

MACHINE-SPIRIT #400D handoff — derived epistemic cache/frontier requirements

Machine Spirit #401A procedural-memory handoff

Machine Spirit #401B downstream durability handoff — procedural policy history

Machine Spirit #401C handoff — durable procedural interoperability lineage

Upstream Machine Spirit #401D handoff — procedural execution/replay durability

Durable / append-only where material

Rebuildable derived state by default

Machine Spirit #402A durable-ledger handoff

MACHINE-SPIRIT #402B canonical durability refinement — GRIE

MACHINE-SPIRIT #402C handoff — Causal Ledger Federation Mesh

Upstream Machine Spirit #403B handoff — retrieval inputs from durable provenance/dependence

MACHINE-SPIRIT #404A handoff — EPOCH learning ledger

Machine Spirit #404B / CITADEL-EPOCH durable-learning handoff

Machine Spirit #404C handoff — durable learned-artifact federation

Implementation decomposition

MACHINE-SPIRIT #404D durable-ledger refinement

Machine Spirit #405C / CONCORDAT durable-ledger handoff