DSE0 Preflight and Removal Inventory - TsatsuAmable/nemosyne GitHub Wiki

Generated reference. This page is published automatically from the version-controlled repository. Canonical source: docs/review-plans/DSE0_PREFLIGHT_AND_REMOVAL_INVENTORY.md. Do not edit this wiki page directly.

DSE0 — Reduction Preflight, Caller Inventory, and Direct-Compile Plan

Status: Implementation Preflight / Authoritative Review Plan
Date: 2026-10-04
Governing authority: RFC 0012, ADR-0013
Execution authority: ROADMAP.md

1. Executive Summary and DSE0 Scope

Following the approval of RFC 0012 and ADR-0013, Nemosyne consolidates its production architecture into four core subsystems:

  1. Data and evidence runtime (Rust/WASM analytical authority)
  2. Investigation (single authoritative InvestigationAggregate / AtlasCore facade)
  3. Representation compiler (deterministic evidence-bound composition, admission, spatial plan compilation)
  4. Presentation and input (modality-independent scene/lifecycle owner, desktop/WebXR adapters, NIL boundary)

DSE0 delivers the mandatory preflight obligations before executable deletions begin in DSE1:

  • An exhaustive inventory of live callers, exports, and test suites for all candidate removals;
  • Explicit selection of two governed analytical phenomena for the minimum loop;
  • Direct deterministic compile path specification (bypassing compulsory neural advice and runtime search);
  • Production falsifiers establishing $N$-independent cardinality bounds and clean admission gates.

2. Removal and Consolidation Inventory

2.1 Target 1: SemanticEmbodimentGraphV1 (Duplicate Semantic Truth)

  • Status: RETIRE production duplication. Canonical SemanticSnapshotV1 owns grounded semantics.
  • Source Files:
    • src/moneta/representation/SemanticEmbodimentGraphV1.ts
    • src/moneta/representation/ComposedRepresentationValidation.ts
    • src/moneta/representation/index.ts (public barrel export)
  • Direct Callers:
    • ComposedRepresentationValidation.ts: validates node types (concept, observation, structure) and edge relationships.
  • Test Dependencies:
    • tests/composed-representation-validation.test.ts: exercises validation rules on graph instances.
    • tests/moneta-semantic-prerequisites.test.ts: checks graph construction prerequisites.
  • Migration Plan:
    • Retain read-only decoder for legacy package compatibility.
    • Remove mutable graph instantiation from active representation compilation paths.
    • Map validation rules directly onto SemanticSnapshotV1.body.

2.2 Target 2: Mandatory FullMonetaEngine.synthesizeOrAdapt Coordination

  • Status: MERGE into unified representation compiler request path; retire redundant coordinator.
  • Source Files:
    • src/moneta/adaptation/FullMonetaEngine.ts
    • src/moneta/adaptation/index.ts
  • Direct Callers:
    • src/atlas/domain/InvestigationAggregate.ts: lines 970 (synthesizeOrAdapt) and 1026 (explainFullMonetaDecision).
    • src/atlas/AtlasCore.ts: facades InvestigationAggregate.adaptRepresentation.
  • Test Dependencies:
    • tests/full-moneta-adaptive-intelligence.test.ts
    • tests/fm6-r6-human-refinement-qualification.test.ts
    • tests/fm7-r7-search-synthesis-qualification.test.ts
  • Migration Plan:
    • Introduce compileDirectEmbodimentPlan() in representation compiler subsystem.
    • Update InvestigationAggregate.adaptRepresentation() to support direct deterministic compilation mode as default.
    • Retain FullMonetaEngine methods as secondary research harness entry points until full caller migration.

2.3 Target 3: Compulsory FormaSystem1Proposer Inference on Default Path

  • Status: REMOVE from default execution; keep as optional, explicitly requested acceleration.
  • Source Files:
    • src/moneta/forma/FormaSystem1Proposer.ts
  • Direct Callers:
    • FullMonetaEngine.ts: line 172 (new FormaSystem1Proposer(...)), line 195 (generateProposals(...)).
  • Test Dependencies:
    • tests/forma-system1-proposer.test.ts
    • tests/fm5-system1-pairwise-benchmark.test.ts
  • Migration Plan:
    • When no advice artifact is supplied and advisory assistance is not requested by the investigator, the compiler skips FormaSystem1Proposer entirely.
    • Consumed advice identity is recorded in provenance only when advice was actually applied.

2.4 Target 4: Interactive Evolutionary Population Search (RepresentationSearchEngine.search)

  • Status: MOVE evolutionary search outside the interactive browser loop; retain deterministic baseline enumeration.
  • Source Files:
    • src/moneta/search/RepresentationSearchEngine.ts
    • src/moneta/search/RepresentationGenomeHandoff.ts
  • Direct Callers:
    • FullMonetaEngine.ts: line 205 (RepresentationSearchEngine.search(...)).
  • Test Dependencies:
    • tests/representation-search-engine.test.ts
    • tests/fm7-r7-search-synthesis-qualification.test.ts
  • Migration Plan:
    • Interactive compiler selects from the deterministic candidate grammar enumeration without spawning multi-generation populations.
    • Offline search engines output versioned RepresentationGraph or candidate packages which the product compiler validates through normal admission gates.

2.5 Target 5: Speculative DATASET Primitive and Separate Subject Registry

  • Status: DELETE from implementation proposals.
  • Rationale: The canonical dataset fingerprint (SHA-256) already uniquely and immutably identifies the subject. Adding a secondary DATASET visual primitive or DatasetSubjectRefV1 registry creates redundant indirection without improving scientific invariants.

3. Direct Deterministic Compilation Path (DSE1 Specification)

3.1 Selected Analytical Phenomena

For the initial complete DSE1 grounded loop, two complementary governed analytical phenomena are selected:

  1. Phenomenon 1 (Statistical Distribution): 1D continuous density / histogram estimation computed by Rust analytical kernel.
    • Grounded representation: Bounded density profile / frequency envelope.
    • Scientific purpose: Characterize variable dispersion, skewness, and modality.
  2. Phenomenon 2 (Topological Clustering): 2D/3D spatial cluster centroids and boundary hulls computed via Rust spatial reduction.
    • Grounded representation: Coordinated cluster boundary glyphs with semantic density anchors.
    • Scientific purpose: Identify partitioning, sub-population separation, and cluster relationships.

3.2 Direct Compilation Call Path

[Dataset] 
   │
   ▼ (Rust analytical execution via Worker port)
[AnalysisResult records + EvidenceLedger]
   │
   ▼ (Deterministic reduction & normalization)
[SemanticSnapshotV1 (Grounded facts only)]
   │
   ▼ (Evidence-bound deterministic composition)
[ComposedRepresentationPlan (Deterministic baseline)]
   │
   ▼ (Obligation & resource budget admission check)
[Admitted SpatialEmbodimentPlanV1 + Reverse Explanation]
   │
   ▼
[Desktop / WebXR Scene Presentation]

3.3 Core Invariants of the Direct Path

  1. Deterministic Execution: Given identical Dataset and CommittedInvestigationContextV2, the compilation produces bitwise identical SpatialEmbodimentPlanV1 and explanation text.
  2. Zero Neural Overhead: No ONNX models are instantiated; no network requests are dispatched; no random seeds are consumed.
  3. $N$-Independent Scene Cardinality: The number of rendered visual primitives (spatial meshes, coordinate anchors, text labels) is strictly bounded by the semantic resolution budget, regardless of whether $N=10^2$ or $N=10^6$ rows.

4. Production Falsifiers Protocol for DSE0/DSE1

ID Invariant Attacked Falsifying Condition Test Verification
FAL-DSE0-1 Direct compile autonomy Compilation fails or attempts to load ONNX/System-1 when advisory options are omitted. Direct compile unit test without model files in runtime path.
FAL-DSE0-2 Search population isolation Representation compilation execution time scales with evolutionary maxGenerations or populationSize. Timing assertion verifying $O(1)$ latency across generation parameters in direct mode.
FAL-DSE0-3 Bounded scene cardinality ($N$-independence) Rendered primitive count or vertex buffer transfer increases linearly with row count $N$. Benchmark test comparing $N=100$ vs. $N=50,000$ row datasets under identical budget.
FAL-DSE0-4 Reverse explanation accuracy Admitted plan fails to provide verifiable references to the originating Rust analytical result IDs. Assertion that every visual node cites its authoritative evidenceReference.
FAL-DSE0-5 Dual-epistemic separation Conjectural elements appear in admitted plan when epistemicPurpose is GROUNDED_ANALYSIS. Admission check rejects ungrounded primitives during exploratory mode mismatch.

5. Next Execution Steps (DSE1 Tranche)

  1. Implement compileDirectEmbodimentPlan() in representation compiler subsystem.
  2. Wire direct deterministic compile path into InvestigationAggregate / AtlasCore.
  3. Build DSE1 qualification battery exercising falsifiers FAL-DSE0-1 through FAL-DSE0-5.
  4. Validate that existing smoke and journey tests continue to pass with direct compilation active.