jc: Pillar 6 — EWA-Sandwich Σ-push-forward (cant-stop-thinking math foundation)#289
Merged
Merged
Conversation
Certifies the qualitative shift from Gaussian convolution (additive error
growth) to EWA-sandwich (geometric, bounded error). The mathematical
foundation for cant-stop-thinking loops where Frame n+1 = J·Frame n·Jᵀ
stays well-conditioned across arbitrary path depth.
# The mathematical claim
In Elliptical Weighted Average (EWA) splatting, a 3D Gaussian's covariance
propagates to image-space via the projection Jacobian J:
Σ_image = J · Σ_world · Jᵀ (sandwich, not addition)
For multi-hop edge propagation, the analogous operation along path
A → B → C is:
Σ_path = M_path · Σ_0 · M_pathᵀ
where M_path = M_n · M_{n-1} · ... · M_1
and M_k = sqrt(step_sigma_k)
The sandwich form preserves the geometric structure of the path. Plain
Gaussian convolution (Σ_AB ⊕ Σ_BC = Σ_AB + Σ_BC, addition in the Lie
algebra) gives O(n) arithmetic error growth; sandwich gives bounded
geometric (multiplicative) error iff the M_k are contractive on average.
# Result
[09/09] EWA-Sandwich: Σ-push-forward along multi-hop edge paths
✓ PASS measured CV=0.8800 predicted=0.6000 (2 ms)
PSD-preservation rate = 1.000000 (10000/10000 hops)
Worst path had 0 non-SPD intermediate states.
Concentration: mean ‖log(Σ_n)‖²_F = 0.8461, std = 0.7446
CV tightness = 1.467× (PASS if ≤ 1.75 in log-normal regime)
# Two claims certified jointly
1. PSD-preservation: 10000/10000 hops kept Σ in the SPD cone. This is
the qualitative break with naive convolution, which only preserves
semi-definiteness in the Lie algebra and can numerically degrade.
Sandwich preserves PSD by mathematical construction.
2. Bounded concentration: the coefficient of variation of ‖log(Σ_n)‖_F²
across 1000 independent paths sits at 1.467× the log-normal-corrected
KS bound. Naive chi-squared bound (√(2/n) ≈ 0.45) under-predicts
because sandwich has multiplicative log-normal tails; corrected bound
√(2/n)·√(1+2σ²n) accounts for this.
# Architectural significance
This is the missing link between the Σ-codebook probe (#288 — viable at
R²=0.9949) and Multi-Hop graph queries that don't lose signal at depth.
Three properties combine:
Pillar 5+ (KS): single-aggregation step is concentrated
Pillar 5++ (DZ): Hilbert-space-valued sums converge
Pillar 6 (EWA): path propagation preserves structure (THIS PR)
Together they certify the full pipeline:
Σ_codebook(input) → Σ_propagate(path) → Σ_aggregate(neighborhood)
^ ^
Pillar 6 Pillars 5+, 5++
The cant-stop-thinking loop now has its mathematical backbone:
Frame_{n+1} = J(edge_state) · Frame_n · J(edge_state)ᵀ
stays in the SPD cone for any depth, with bounded log-norm growth.
# Tests (7/7 grün)
- identity_sandwich_is_identity (basic algebra)
- sandwich_preserves_spd (1000 random samples — none fail)
- sandwich_with_identity_returns_input (sqrt(M)·I·sqrt(M) = M)
- path_propagation_returns_finite_results (no NaN/Inf)
- long_paths_dont_explode (length-50 paths with σ_step=0.1)
- deterministic_with_fixed_seed (RNG reproducibility)
- pillar_passes (full substrate-native claim)
# Files
- crates/jc/src/ewa_sandwich.rs (new, ~440 lines incl. 7 unit tests)
- crates/jc/src/lib.rs (mod decl + run_all_pillars list entry)
Pure Rust, zero deps, ~2ms runtime (1000 paths × 10 hops × matrix ops).
# Status
Implemented: 7/9 pillars (was 6/8 before this PR)
Passed: 7
Failed: 0
Deferred: 2 (coupled revival track unchanged)
# Run
cargo test --manifest-path crates/jc/Cargo.toml --release ewa_sandwich
cargo run --manifest-path crates/jc/Cargo.toml --release --example prove_it
This was referenced Apr 29, 2026
AdaWorldAPI
pushed a commit
that referenced
this pull request
May 6, 2026
…r VSA-scope correction pattern.md (NEW, 578 lines) — usability patterns for traversing the workspace's SoA/DTO surface as a graph (nodes = type defs, edges = producer-consumer + duplicate + seam, subgraphs = clusters). 15 named patterns covering canonical lookup, maturity scoring, Click-P-1 lens, register-laziness check, dual-tier writes, ingestion-commit, lineage- as-column, append-only governance, consult-first ordering, cross- session blackboard via ledger row IDs, source-vs-claim divergence, cluster-fix discipline, debug-as-API debt, scope-lock, and seam-naming. Plus 7 critical findings and append-only update protocol. ARCHITECTURE_ENTROPY_LEDGER.md (APPEND-only correction block) — VSA scope correction per CLAUDE.md I-VSA-IDENTITIES iron rule: - VSA-1 description tightened: Vsa16kF32 is for Markov chain over identity fingerprints exclusively. Provenance / JWT / RBAC / IDs are register territory, not VSA carriers. - PERMUTE-1 description tightened: vsa_permute is unitary as an operation but the braiding usage is NOT lossless; cross-talk shrinks unbinding margin with N. Bound: N <= sqrt(d)/4. New rows: - EWA-SANDWICH-1 (PR #289 was missing from initial snapshot): Stage 3 / Smart / entropy 2. Scope: SPD-bounded propagation of cognitive Vsa16kF32 across Markov rho^d cycles. NOT a lineage error model (corrected from initial framing). - SUBJECT-DTO-1: aspirational typed Subject struct with AuthSource enum (typed JwtClaims, not VSA). Implied by MedCareV2 #7+#8 wire shape. - MOCK-DRIVER-1: q2 PR #35 Phase-3 placeholder, Stage 2. Cross-repo resolution events: - THINK-1 partial: q2 PR #35 dropped thinking-engine + cognitive-shader-driver deps from cockpit-server, migrated to canonical contract::cognitive_shader::*. Wire compression 256x on cycle_fingerprint, 128x on color_acc. - TRUTH-1 partial: q2 PR #35 cockpit-server bridges to lance-graph-planner::nars::truth::TruthValue::deduction. - POLICY-1 + MEMBRANE-GATE-1: priority bump — MedCareV2 #8 now blocking on impl MembraneGate for Arc<rbac::Policy>. Section G — ingestion-vs-traversal axis added: Cypher-parser path (Option 1, ships now via PARSER-1 resolution) and splat-deposit path (Option 2, gated by SPLAT-1) both converge on E1 typed Action API. Retractions: E4 (VSA-bundled provenance) and E8 (geometrically- bounded provenance via Vsa16kF32+EWA) — register laziness; do NOT get appended to EPIPHANIES.md. E1, E2, E3, E5, E6, E7, E9 stand. https://claude.ai/code/session_012AUf5NFgeAAQa5aQAKwSgx
AdaWorldAPI
pushed a commit
that referenced
this pull request
May 6, 2026
…r VSA-scope correction pattern.md (NEW, 578 lines) — usability patterns for traversing the workspace's SoA/DTO surface as a graph (nodes = type defs, edges = producer-consumer + duplicate + seam, subgraphs = clusters). 15 named patterns covering canonical lookup, maturity scoring, Click-P-1 lens, register-laziness check, dual-tier writes, ingestion-commit, lineage- as-column, append-only governance, consult-first ordering, cross- session blackboard via ledger row IDs, source-vs-claim divergence, cluster-fix discipline, debug-as-API debt, scope-lock, and seam-naming. Plus 7 critical findings and append-only update protocol. ARCHITECTURE_ENTROPY_LEDGER.md (APPEND-only correction block) — VSA scope correction per CLAUDE.md I-VSA-IDENTITIES iron rule: - VSA-1 description tightened: Vsa16kF32 is for Markov chain over identity fingerprints exclusively. Provenance / JWT / RBAC / IDs are register territory, not VSA carriers. - PERMUTE-1 description tightened: vsa_permute is unitary as an operation but the braiding usage is NOT lossless; cross-talk shrinks unbinding margin with N. Bound: N <= sqrt(d)/4. New rows: - EWA-SANDWICH-1 (PR #289 was missing from initial snapshot): Stage 3 / Smart / entropy 2. Scope: SPD-bounded propagation of cognitive Vsa16kF32 across Markov rho^d cycles. NOT a lineage error model (corrected from initial framing). - SUBJECT-DTO-1: aspirational typed Subject struct with AuthSource enum (typed JwtClaims, not VSA). Implied by MedCareV2 #7+#8 wire shape. - MOCK-DRIVER-1: q2 PR #35 Phase-3 placeholder, Stage 2. Cross-repo resolution events: - THINK-1 partial: q2 PR #35 dropped thinking-engine + cognitive-shader-driver deps from cockpit-server, migrated to canonical contract::cognitive_shader::*. Wire compression 256x on cycle_fingerprint, 128x on color_acc. - TRUTH-1 partial: q2 PR #35 cockpit-server bridges to lance-graph-planner::nars::truth::TruthValue::deduction. - POLICY-1 + MEMBRANE-GATE-1: priority bump — MedCareV2 #8 now blocking on impl MembraneGate for Arc<rbac::Policy>. Section G — ingestion-vs-traversal axis added: Cypher-parser path (Option 1, ships now via PARSER-1 resolution) and splat-deposit path (Option 2, gated by SPLAT-1) both converge on E1 typed Action API. Retractions: E4 (VSA-bundled provenance) and E8 (geometrically- bounded provenance via Vsa16kF32+EWA) — register laziness; do NOT get appended to EPIPHANIES.md. E1, E2, E3, E5, E6, E7, E9 stand. https://claude.ai/code/session_012AUf5NFgeAAQa5aQAKwSgx
6 tasks
AdaWorldAPI
pushed a commit
that referenced
this pull request
May 6, 2026
…entries New: crates/jc/examples/splat_to_ewa_bridge.rs (~340 LOC) Closes the seam between SPLAT-1 (Stage 1, contract::splat shipped in PR #336/#344) and EWA-SANDWICH-1 (Stage 3, jc::ewa_sandwich shipped in PR #289). End-to-end Pillar-6-bounded propagation through the lance_graph_contract::splat surface. Empirical results (cargo run --release --example splat_to_ewa_bridge): Canonical 5-hop OSINT chain (Lavender → IDF → Israel → NSO → Pegasus → Khashoggi): - 5/5 hops SPD-preserved through witness_to_splat → splat_to_sigma → sandwich - ‖log Σ_5‖_F = 4.7159 (geometric shrinkage from sub-1.0 entries — expected given (eff_amp, width)/255 mapping) - 5 unique bit positions deposited into Support plane - 5 replay_refs preserved verbatim (identity preservation confirmed) - 12.4 KB total memory (12 KB SplatPlaneSet + 160 B per-splat ledger) - 107 µs runtime end-to-end 1000-path × 10-hop stress test (deterministic splitmix64 seed): - SPD-preservation rate: 1000/1000 (100%) — replicates Pillar 6's 10000/10000 result through the SPLAT contract - mean ‖log Σ_n‖_F = 13.07, std 2.82 - 395 µs total (0.4 µs/path) The L1-L4 BLAS framing is now empirically grounded: L1 popcount over plane → exact top-k of deposited bits L2 SplatPlaneSet::deposit → channel-routed splat into one row L3 sandwich along the chain → Σ_path Pillar-6-bounded L4 per-row L3 over SoA → "huge spatial BLAS" Provisionally named SplatShaderBlas per user 2026-05-06: "the godfather of needle-in-a-haystack". Distinct from existing blasgraph (adjacency- shaped CSR/CSC sparse semiring) — this substrate is plane-shaped (dense per-row, splat-as-deposit, Pillar-6 SPD bound). Both valid; SplatShaderBlas wins where fan-out is high and rows are sparse-deposit (the dense-row sparse-graph regime from nvgraph + GraphBLAS literature). Lab precedent: Gaussian splat tested at 20000 × 20000 with zero errors in lab condition (per user 2026-05-06). 16K production target sits well below the validated ceiling; bottleneck is the production seam (E1 BindSpace.apply Action API), not the algebra. Cargo.toml: lance-graph-contract added as [dev-dependencies] only. Production code stays zero-dep per JC's standalone-crate invariant. Files: - crates/jc/examples/splat_to_ewa_bridge.rs (NEW) - crates/jc/Cargo.toml (added [dev-dependencies] + [[example]]) - crates/jc/Cargo.lock (lockfile churn from new dev-dep) - .claude/board/ARCHITECTURE_ENTROPY_LEDGER.md (APPEND-only block: SPLAT-EWA-BRIDGE-1 row + L1-L4 BLAS picture + SplatShaderBlas naming + 20K × 20K lab precedent) https://claude.ai/code/session_012AUf5NFgeAAQa5aQAKwSgx
7 tasks
AdaWorldAPI
pushed a commit
that referenced
this pull request
May 13, 2026
…est anchor
Two more 2026-05-13 epiphanies + one TD row capturing the user's
"wire OGIT ↔ OSINT ↔ Palantir Gotham / Neo4j route" + "remember
FMA 75K human anatomy as on-screen rendering smoke test" directives.
EPIPHANIES.md (2 new entries, prepended):
1. FMA smoke-test anchor (75K-entity OWL ontology) — the canonical
dual-purpose demo for the integration arc:
- Neo4j-ish edges propagation test (EWA-Sandwich Σ-push-forward
replaces 5-hop Cypher; Pillar 6 PR #289 certified)
- Healthcare super-domain test (UnifiedBridge<MedcareBridge> with
SuperDomain::Healthcare audit chain at FMA-scale)
- Visual rendering test (q2 cockpit + notebook-render produces
3D anatomy with heart-click → edges propagation visible)
- Touches all three substrate paths (thinking-engine cognition,
ractor supervisor, ndarray::simd batch compute)
- 6-phase sequencing converging on demoable end-to-end FMA-heart
query (phase 4 = Q2-2.x Cypher console + Q2 3D view ready)
2. OGIT ↔ OSINT ↔ Palantir/Neo4j ↔ q2 route — q2 is the 8th consumer
subcrate slot alongside the 5 super-domain subcrates (medcare-rs,
smb-office-rs, woa-rs, hiro-rs, hubspot-rs) + 2 OSINT-related
crates (aiwar-ingest in q2 workspace, external aiwar/neo4j-rs
/aiwar-neo4j-harvest research repos). Verified q2 is accessible
via MCP (mcp__github__get_file_contents on AdaWorldAPI/q2).
End-to-end route mapped from external user opening q2 notebook
through PolicyRewriter chain → DataFusion ScanExec → arrow render.
Gaps identified: q2's local lance-graph + ndarray stubs need
re-exports; notebook-query polyglot dispatcher needs wiring;
Q2Bridge (D-ONTO-V5-5) needs TTL+bridge work.
TECH_DEBT.md (1 new TD row, prepended):
TD-Q2-STUBS-DEDUP-1 (P1) — q2's crates/stubs/lance-graph and
crates/stubs/q2-ndarray are placeholders that must be replaced
with re-exports from AdaWorldAPI/lance-graph + AdaWorldAPI/ndarray
before FMA smoke-test can compile against the canonical pipeline.
~60 LOC PR + 2 integration tests proving type-universe coherence
(q2_lance_graph_canonical_test, q2_ndarray_simd_dispatch_test).
https://claude.ai/code/session_01PjcbSTd9zdVPkta9qwkVKo
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Was
New Pillar 6 in the
jcproof-in-code harness: certifies that EWA-sandwich propagationΣ_path = M·Σ·Mᵀalong multi-hop edge paths preserves PSD by mathematical construction AND keeps log-norm growth bounded.This is the missing link between:
→ and the multi-hop path queries that dont lose signal at depth, which the cant-stop-thinking architecture requires.
Resultat
10000 von 10000 Hops haben Σ in der SPD-Mannigfaltigkeit gehalten. Das ist die qualitative Trennung zur naiven Konvolution, die nur Semi-Definitheit in der Lie-Algebra erhält und numerisch degradieren kann. Sandwich erhält PSD durch mathematische Konstruktion.
Die mathematische Behauptung
In Elliptical Weighted Average (EWA) Splatting wird die Kovarianz einer 3D-Gauss-Verteilung über die Projektions-Jacobian J in den Bildraum propagiert:
Übertragen auf Multi-Hop-Edge-Propagation entlang Pfad A→B→...→N:
Die Sandwich-Form erhält die geometrische Struktur des Pfades. Plain Gauss-Konvolution (Σ_AB ⊕ Σ_BC = Σ_AB + Σ_BC, Addition in der Lie-Algebra) gibt O(n) arithmetisches Fehler-Wachstum; Sandwich gibt bounded geometric (multiplicative) Fehler-Wachstum, sofern die M_k im Mittel kontraktiv sind.
Die zwei Behauptungen — gemeinsam zertifiziert
PSD-Erhaltung: 10000/10000 Hops. Qualitativer Bruch mit naiver Konvolution. Sandwich erhält PSD durch Konstruktion, nicht durch Toleranz.
Bounded Concentration: Coefficient-of-Variation von ‖log(Σ_n)‖_F² über 1000 unabhängige Pfade liegt bei 1.467× des log-normal-korrigierten KS-Bounds. Naive chi-quadrat Bound (√(2/n) ≈ 0.45) unter-vorhersagt weil Sandwich multiplikative log-normal Tails hat; korrigierte Bound √(2/n)·√(1+2σ²n) berücksichtigt das.
Architektur-Bedeutung
Die drei Säulen kombiniert:
Gesamte zertifizierte Pipeline:
Der cant-stop-thinking Loop hat jetzt sein mathematisches Rückgrat:
bleibt im SPD-Cone für jede Tiefe, mit bounded log-norm Wachstum.
Tests (7/7 grün)
identity_sandwich_is_identity— Basis-Algebrasandwich_preserves_spd— 1000 zufällige Samples, alle PSDsandwich_with_identity_returns_input— sqrt(M)·I·sqrt(M) = Mpath_propagation_returns_finite_results— keine NaN/Inflong_paths_dont_explode— Length-50-Pfade mit σ_step=0.1deterministic_with_fixed_seed— RNG-Reproduzierbarkeitpillar_passes— substrate-native AnspruchFiles
crates/jc/src/ewa_sandwich.rs(neu, ~440 Zeilen inkl. 7 Unit-Tests)crates/jc/src/lib.rs(Modul-Deklaration +run_all_pillarsEintrag)Pure Rust, zero deps, ~2ms Laufzeit.
Status nach diesem PR
Verifikation
Out of Scope für künftige PRs