docs: Path 12 — DeepNSM × CausalEdge64 bridge + Epiphany 19 by AdaWorldAPI · Pull Request #62 · AdaWorldAPI/lance-graph

AdaWorldAPI · 2026-03-29T06:34:46Z

Path 12: Causal-semantic reasoning via shared palette algebra.

Pearl's 3-bit causal mask (SPO plane projection) and DeepNSM's
16 NsmCategory decomposition are the SAME operation:
Pearl: 2³ = 8 projections on causal roles
DeepNSM: 2¹⁶ projections on semantic categories (α selects useful ones)
Combined: "what CAUSED this, in the MENTAL category?"

DeepNsmCausalBridge:
- 4,096 words → 256 palette (k-means on 96D COCA)
- AttentionTable[256][256] (distributional similarity)
- ComposeTable[256][256] (multi-hop O(1))
- 16 category_tables (per-NsmCategory distance)
- 5 operations: word_to_edge, semantic_projection, cross_domain_analogy,
compose_chain, causal_semantic_query

CAM-PQ stroke alignment:
Stroke 1 = interventional (entity identity only)
Stroke 2 = observational (+ behavior)
Stroke 3 = full model (all 96 dimensions)

Base17 connection: 96D → 17D → 256 palette → 8-bit edge index
End-to-end ρ ≈ 0.957

Epiphany 19: DeepNSM, CausalEdge64, and bgz-tensor are three instances
of the same abstract machine — precomputed symmetric lookup with
plane-selective mask. The bridge makes them interoperable.

https://claude.ai/code/session_01Y69Vnw751w75iVSBRws7o7

Path 12: Causal-semantic reasoning via shared palette algebra. Pearl's 3-bit causal mask (SPO plane projection) and DeepNSM's 16 NsmCategory decomposition are the SAME operation: Pearl: 2³ = 8 projections on causal roles DeepNSM: 2¹⁶ projections on semantic categories (α selects useful ones) Combined: "what CAUSED this, in the MENTAL category?" DeepNsmCausalBridge: - 4,096 words → 256 palette (k-means on 96D COCA) - AttentionTable[256][256] (distributional similarity) - ComposeTable[256][256] (multi-hop O(1)) - 16 category_tables (per-NsmCategory distance) - 5 operations: word_to_edge, semantic_projection, cross_domain_analogy, compose_chain, causal_semantic_query CAM-PQ stroke alignment: Stroke 1 = interventional (entity identity only) Stroke 2 = observational (+ behavior) Stroke 3 = full model (all 96 dimensions) Base17 connection: 96D → 17D → 256 palette → 8-bit edge index End-to-end ρ ≈ 0.957 Epiphany 19: DeepNSM, CausalEdge64, and bgz-tensor are three instances of the same abstract machine — precomputed symmetric lookup with plane-selective mask. The bridge makes them interoperable. https://claude.ai/code/session_01Y69Vnw751w75iVSBRws7o7

OGAR#62 merged 2026-06-19T12:00:45Z (project_actor at CODEBOOK 0x0004). Collapses Redmine+OpenProject User/Principal/ApplicationRecord STI chain (4 source classes → one canonical), proving binary cross-curator convergence on real Rails sources. Updated SMOKE-4 entry's Cross-repo alignment paragraph to include #62 alongside #61. OGAR's project-management quartet (project, project_work_item, billable_work_entry, project_actor) is complementary to this commit's commerce/billing/erp sextet (CommercialLineItem, CommercialDocument, TaxPolicy, BillingParty, PaymentRecord, CurrencyPolicy). Different domain, same promotion mechanics. The STI-collapse pattern OGAR#62 demonstrated for ProjectActor is the same shape my BillingParty lexical detector applies on OSB::Client + Odoo::res_partner — same mechanism, different domain. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Xzyc27Nx3f8WC5KzwfWfjx

…esOrderLine / FulfillmentFlow / InventoryMovement / ProductOffering); 3-curator convergence on TaxPolicy + PaymentRecord Phase 2 of operator's "all of the above" 2026-06-19. Smoke target B (Spree↔Odoo commerce overlap) landed with real harvested triples. What landed: 1. Spree harvest fixture — pulled spree/spree upstream zipball (85 MB, no AdaWorldAPI fork available; canon's forks-only rule applies to Cargo deps, not read-only fixture harvests). Ran ruff_ruby_spo on spree/core: 289 models, 7954 triples, 909 KB ndjson → checked in as crates/lance-graph-ontology/tests/fixtures/spree_ruby_spo.ndjson. 2. CanonicalConcept enum 6 → 11 variants. New (all from smoke target B): - SalesOrder (Spree::Order + odoo:sale_order) - SalesOrderLine (Spree::LineItem + odoo:sale_order_line) - FulfillmentFlow (Spree::Shipment + odoo:stock_picking) - InventoryMovement (Spree::InventoryUnit + odoo:stock_move) - ProductOffering (Spree::Product / Variant + odoo:product_product / product_template) 3. CommercialDocument detector narrowed: drops the "order" ending so sale_order / Spree::Order route to SalesOrder (a distinct commerce-side concept), keeping CommercialDocument as the accounting-document concept (Invoice / account_move). This is the distinct-but-adjacent split the operator's smoke spec implied. 4. 5 new sibling detectors with the same lexical-shape pattern as the commerce sextet, each with its concept-specific lexical hint: - classes_matching_sales_order_shape_canonical (ends_with "order" AND NOT contains "line") - classes_matching_sales_order_line_shape_canonical (ends_with "lineitem" OR ends_with "order_line"/"orderline") - classes_matching_fulfillment_flow_shape_canonical (ends_with "shipment" OR ends_with "picking") - classes_matching_inventory_movement_shape_canonical (ends_with "inventoryunit" OR ends_with "stock_move" — the stock_ qualifier discriminates from account_move which is a CommercialDocument) - classes_matching_product_offering_shape_canonical (ends_with "product" OR ends_with "variant" OR ends_with "product_template") 5. TaxPolicy detector hardened: also matches contains("taxrate") so Spree::TaxRate surfaces (the strongest commerce-side tax-policy signal). OSB::Tax and odoo:account_tax stay matched via the existing ends_with("tax") arm. 6. 6 new corpus-driven tests, all green: - spree_order_and_odoo_sale_order_overlap_as_sales_order - spree_line_item_and_odoo_sale_order_line_overlap_as_sales_order_line - spree_shipment_and_odoo_stock_picking_overlap_as_fulfillment_flow - spree_inventory_unit_and_odoo_stock_move_overlap_as_inventory_movement (also asserts account_move does NOT promote here) - spree_product_variant_and_odoo_product_overlap_as_product_offering - spree_third_curator_convergence_on_tax_policy_and_payment_record (proves the existing TaxPolicy + PaymentRecord detectors generalize to a 3rd curator — Spree::TaxRate, Spree::Payment) Plus all 20 prior tests still green → 26/26 total. Scope reminder per operator discipline acceptance #4-7: each promotion is gated by ≥2-curator structural evidence on real corpora, not pre-emptive. The SMOKE-1 to SMOKE-4 + this commit's 5 concepts now cover smoke targets A (OSB↔Odoo accounting) and B (Spree↔Odoo commerce). Smoke target C (Redmine/OpenProject Project::TimeEntry materialization) is project-domain — already being shipped by the other session into OGAR's CODEBOOK (PRs #61/#62/#63 promoted project, project_work_item, billable_work_entry, project_actor, project_status, project_type). CanonicalConcept enum now 11 variants. The lance-graph-ontology CanonicalConcept set complements OGAR's project-domain 6-entry CODEBOOK with an 11-entry commerce/billing/erp candidate set (CommercialLineItem, CommercialDocument, TaxPolicy, BillingParty, PaymentRecord, CurrencyPolicy, SalesOrder, SalesOrderLine, FulfillmentFlow, InventoryMovement, ProductOffering). Each candidate is a future OGAR CODEBOOK row (Phase 3 of "all of the above" — the OGAR upstream promotion PR). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Xzyc27Nx3f8WC5KzwfWfjx