Skip to content

LashSesh/pse

Repository files navigation

PSE — Post-Symbolic Exoskeleton

PSE is a post-symbolic cognitive substrate — a multi-layer formal architecture that sits between purely symbolic (GOFAI) and purely sub-symbolic (connectionist) computation, reducing to neither.

The base layer emits content-addressed, falsifiable, deterministic crystal artifacts from structured observation streams. Above it, a stack of formal traversal layers provides geometric and topological constraints that govern how probabilistic reasoning navigates a problem space. The complete architecture is constitutionally governed by the ADAMANT protocol (ADAMANT_v1.0.0.pdf), which enforces Artifact Supremacy, fail-closed gate semantics, deterministic replay identity, and Dissolution-Grundsatz across every layer.

For what post-symbolic computation means as a category, see docs/POST_SYMBOLIC.md.


Architecture

PSE is organised into three tiers:

Tier 1 — Core Engine

The observation-to-crystal pipeline. Feeds an observation stream into a 5D-embedded topological graph, tests resonance with a carrier helix-pair (Mandorla coherence κ), gates through eight conjunctive Kairos metrics, runs dual cascade-consensus, optionally falsifies against surrogate streams, and — when a configuration survives all of that — emits a Crystal: a content-addressed (SHA-256 / JCS) record of which graph region produced the resonance, under which constraints, with which provenance.

Crystals are byte-identical across replays, EU AI Act compliant (docs/COMPLIANCE.md), accumulate across sessions via the pattern-memory index, and compose through an algebra of operators (compose / dual / bridge / query / interpolate).

Tier 2 — Traversal Agent Stack

Eight formal cognitive layers that navigate problem spaces deterministically above the core engine. Every layer is fail-closed: gate failure never produces a commit. The PSE-Bridge is the only path to a SemanticCrystal.

Layer Spec ID Function
Traversal Agent PSE-TRAVERSE-v0.1 ProblemSpec → DoFGraph → CollapsePlan → Candidate pipeline
Signature PSE-TRAVERSE-SIGNATURE-01 Spectral blueprint search, Pareto-optimal frontier
Dynamics PSE-TRAVERSE-DYNAMICS-01 Morphodynamic tick stabilization over lifted state
Horizon PSE-TRAVERSE-HORIZON-03 Null-centered hypertoroidal phase geometry
Cognition PSE-TRAVERSE-COGNITION-01 Panoptic 720° solvability landscape
Phase Matrix PHASEMATRIX-HIVEMIND-03 Morphodynamic resonance cell substrate
Dual-Fabric Stitch PHASEMATRIX-HIVEMIND-03.1 Persistent field tensor / ephemeral fabric isolation
Topology (TPT-MTL) PSE-TRAVERSE-TPT-MTL-04 Topological panoptic triangulation + Möbius micro-lift
NCTCS PSE-NCTCS-CONFORMANCE-01 Null-centered toroidal control closure (C0–C4)
Metatron PSE-METATRON-MONOLITH-01 Holistic eigenmode closure, G_meta conjunction

Tier 3 — Governance

The ADAMANT protocol (ADAMANT_v1.0.0.pdf) is the constitutional architectural contract referenced by every layer. Key invariants it enforces:

  • Artifact Supremacy (Axiom 6.1.1) — machine-readable artifacts override documentation summaries; a MaterializationAudit record is authoritative over a conformance report.
  • Fail-closed gates — no output when G = 0, regardless of partial satisfaction.
  • Dissolution-Grundsatz — trace, evidence, and gate history are preserved across compaction; removal is a spec violation.
  • Deterministic replay — two runs over byte-identical inputs produce byte-identical outputs at every layer.

The eval matrix (PSE-EVAL-MATRIX-01) is the structured validation instrument. It runs diagnostic cases across all layers and scores PSE against a raw-LLM baseline. Current status: PSE 100% recall, Raw LLM 52–63% recall, across 50+ diagnostic cases spanning 14 layers. All results are diagnostic_only=true pending productive-task validation.


Status

Aspect State
Engine architecture (Strands E–N) Complete
Operator algebra (compose/dual/bridge/query/interpolate) Complete
Falsification (Shuffle, BlockBootstrap, PhaseRandomize) Complete
EU AI Act compliance proof Drafted
Throughput on commodity hardware Verified
Adaptive Kairos calibration Opt-in, working
Diagnostic surface (state.last_gate, pse-demo) Live
PSE Traversal Agent v0.1 Complete
Signature layer (PSE-TRAVERSE-SIGNATURE-01) Shipped
Dynamics layer (PSE-TRAVERSE-DYNAMICS-01) Shipped
Horizon layer (PSE-TRAVERSE-HORIZON-03) Shipped
Cognition layer (PSE-TRAVERSE-COGNITION-01) Shipped
Phase Matrix layer (PHASEMATRIX-HIVEMIND-03) Shipped
Dual-Fabric Stitch layer (PHASEMATRIX-HIVEMIND-03.1) Shipped
Topology layer (PSE-TRAVERSE-TPT-MTL-04) Shipped
NCTCS closure layer (PSE-NCTCS-CONFORMANCE-01) Shipped
Holistic eigenmode layer (PSE-METATRON-MONOLITH-01) Shipped
Domain validation layer (PSE-VALIDATION-RUNNER-DOMAIN-01) Shipped
Eval matrix (PSE-EVAL-MATRIX-01) Shipped
ADAMANT protocol (constitutional governance contract) v1.0.0
Productive-task validation (productive_agent_validated) Open frontier

Verified throughput, single-thread, release build, Xeon @ 2.10 GHz:

Bench Value
B01a observe-only ingest up to 2.07 M obs/sec
B01b full pipeline (gate path) up to 659 K obs/sec
B15 macro_step end-to-end 43–110 µs
B05 determinism check PASS (bit-identical replay)
Workspace test suite 928 / 928 passing

The engine produces 0 crystals on default thresholds for unconfigured synthetic workloads — by design. Crystal formation requires either real data the metrics were calibrated for, or the opt-in adaptive calibrator (see pse-demo). Calibration on a real domain is the work that turns this from "well-built engine" into "deployed product".


Quick start

# Build the workspace
cargo build --release

# Run the 30-second demo (synthetic stream, full diagnostics)
cargo run --release -p pse-demo

# Run the full benchmark suite
cargo run --release --example bench_full -p pse-core

# Run the ground-truth benchmark (PSE vs STL-z-score vs Isolation Forest)
cargo run --release -p pse-bench-gt --bin bench_gt

# Inspect kairos rejections in real time
RUST_LOG=pse_core=debug cargo run --release -p pse-demo

Embed PSE in your own program:

use pse_core::{macro_step, GlobalState};
use pse_core::adaptive::AdaptiveCalibrator;
use pse_graph::PassthroughAdapter;
use pse_types::Config;

let config = Config::default();
let mut state = GlobalState::new(&config);

// Optional: self-calibrating Kairos thresholds.
state.adaptive = Some(AdaptiveCalibrator::new(0.05, 200, 100));

let adapter = PassthroughAdapter::new("my_source");
let batch: Vec<Vec<u8>> = vec![serde_json::to_vec(&my_event)?];

if let Ok(Some(crystal)) = macro_step(&mut state, &batch, &config, &adapter) {
    // crystal.crystal_id is the SHA-256 content address
    // crystal.region is the set of graph vertices that produced the resonance
    // crystal.commit_proof carries the falsification p-value (if enabled)
    println!("crystal: {}", hex::encode(crystal.crystal_id));
}

// state.last_gate carries the full GateSnapshot (all 8 metrics) for every tick,
// pass or fail — read it to diagnose why a tick did or didn't crystallize.

Layer detail: Traversal Agent

The pse-traverse crate is the agent layer above the core engine: a deterministic controller that turns a structured problem space into a fail-closed traversal of degrees of freedom, then binds successful candidates to PSE crystals via the bridge — never fabricating crystals on its own.

Ein post-symbolischer Traversierungsagent ist ein Agent, der nicht nur Antworten produziert, sondern Problemräume topologisch erschließt, Lösungspfade kontrolliert kollabiert und jeden stabilen Erkenntnis- oder Implementierungsschritt überprüfbar historisiert.

The pipeline:

ProblemSpec  →  FieldCube         (dimensions, constraints, couplings, paths,
                                   carriers, evidence, topology summary)
             →  DoFGraph           (degree-of-freedom graph)
             →  PathExcision[]     (formal options without admissible paths)
             →  CollapsePlan       (deterministic step ordering: detect → hard
                                   → soft → resolve → verify → commit)
             →  StructuralOperator (Laplacian/matrix profile from DoFGraph)
             →  Signature          (sorted spectral values, content-addressed)
             →  SignatureDiagnostics (gap / degeneracy / rigidity /
                                   asymmetry / fragmentation ∈ [0,1], RegimeHint)
             →  SignatureGateOutcome (advisory or fail-closed gate check)
             →  Candidate          (solver-emitted, with assignments + payloads)
             →  GateReport         (Dual-Fabric: primal + mirror + MCI;
                                   + SignatureGate diagnostic channel)
             →  CommitOutcome      (Crystal | NoCrystal | EvidenceOnly | GateFailed)

The signature layer exposes a blueprint search surface for multi-cycle traversal optimisation:

BlueprintSearch  →  TraversalBlueprint[]  (DeterministicGrid: quantization_scale
                                           × lambda_hard combos, content-addressed)
NonDominatedFrontier                       (Pareto tracker: gap↑, fragmentation↓,
                                           degeneracy↓)
SearchLedger                               (append-only hash-chained evaluation log)
SearchAutopilot                            (Exploration → Exploitation → Refinement
                                           → Validation → Complete)

Layer detail: Dynamics (PSE-TRAVERSE-DYNAMICS-01)

Stabilises problem spaces over ticked state transitions before final candidate generation:

BaseState[]  →  LiftedState[]    (N → N+1 lift, auxiliary = logical tick)
             →  FieldSignal      (alignment / dispersion / pressure, quantized)
             →  GuidanceField    (relax nodes, prune low-weight transitions)
             →  MorphodynamicCompressor (Hebbian update, Split, Merge, Prune)
             →  TransitionProof  (path_delta, energy_delta, density_delta)
             →  DynamicGateReport (Fire / Hold, fail-closed, GATE-01)
             →  DynamicTickReport (content-addressed, replay-identical)

dynamic_run() executes ticks until a DynamicStopCondition is met and embeds the DynamicRunReport as an optional field in TraversalRunReport. DynamicPolicy supports Explore / Exploit / Homeostasis with deterministic density-based adaption (POLICY-01). No SemanticCrystals are produced by the dynamics layer — PSE-bridge remains the sole commit path.


Layer detail: Horizon (PSE-TRAVERSE-HORIZON-03)

Wraps a projection-capable state in a null-centered hypertoroidal phase-space geometry and only finalises when every conjunctive sub-gate (G_visible ∧ G_cone ∧ G_causal ∧ G_dual) plus the upstream Projection-v0.2 outcome is satisfied:

StableState   →  NullCenter         (canonical reference id)
              →  HorizonChart       (carriers, rays, cones, causal order)
              →  PhaseRays          (deterministic carrier traces, T^n)
              →  EventHorizonWindowV3
                                    (phase / epoch / amplitude / jitter / visibility)
              →  ProjectionCone     (angle, focus, dispersion bounds)
              →  CausalAdmissibility
                                    (declared vs observed carrier order)
              →  CollapseEmissionDuality
                                    (round-trip back to the same NullCenter)
              →  HorizonCrossingGate
                                    (G_visible ∧ G_cone ∧ G_causal ∧ G_dual)
              →  CombinedGate       (G_v0.2 ∧ G_v0.3 ∧ ReplayReady)
              →  FinalizedEmissionV3
                                    (emitted only on Pass)
              →  HorizonCertificate (chain hash over rd / chart / crossing /
                                     v0.2 cert / emission / replay)

Every gate-relevant scalar is a CanonicalNumber (no platform floats); every keyed structure is BTreeMap; every list is sorted before hashing. Two runs with byte-identical descriptors and inputs produce byte-identical reports and certificates. The horizon layer never produces a SemanticCrystal — the existing PSE-Bridge remains the only commit path. Failure modes resolve to a deterministic HorizonFailurePolicy (WaitForHorizon / RefineProjectionCone / MigrateCarrier / Recondense / Hold / Abort) and a corresponding HorizonV3Outcome variant.


Layer detail: Cognition (PSE-TRAVERSE-COGNITION-01)

Wraps the problem's reachable solvability landscape in a panoptic phase space. Instead of asking "may this state finalise?" (the projection layer's question) it asks "which states, constraints, phases, memory paths, attractors, carriers, non-local routes and finalisation candidates are visible, latent, blocked or dominant?":

ProblemSpec/Traversal/Projection inputs
  →  CanonicalCognitionState
  →  CognitiveState5D            (ψ, ρ, ω, χ, τ + potential / energy /
                                  entropy / stability)
  →  SpiralMemoryHitSet          (resonance addressing, sorted by
                                  resonance then hash)
  →  ConstraintLatticeCognition  (weighted mass, entropy reduction,
                                  feasible-set uniqueness, percolation)
  →  HypercubePuzzleState        (hidden singles, boundary contracts,
                                  EntropyCollapseCertificate)
  →  PhasePanorama               (current + counterfactual 360° horizons
                                  → 720° total)
  →  ScorpioPhaseScheduler       (deterministic activation windows,
                                  no steganography)
  →  GovernedWormhole            (admitted only with ReasonCode + TTL +
                                  budget + audit_trace)
  →  SelfModelTensor             (reflexive modulation, drift bound)
  →  DualTriggerFeedbackGate     (ordered (t1 ∧ t2 ∧ Δϕ) sequence)
  →  FixpointCalibrationShell    (PoR-acceptance triplet)
  →  CarrierMigrationPlan        (anti-drift, friction, shock)
  →  SingularityTriggerReport    (degenerate stability spectrum)
  →  CognitionHandoffGate        (G_perc ∧ G_panorama ∧ G_self ∧
                                  G_trigger ∧ ReplayReady)
  →  CognitionCandidateBundle    (pass) | CognitionHoldReport (fail)
  →  CognitionReport             (content-addressed, replay-stable)

The cognition kernel never finalises, never commits, and never constructs a SemanticCrystal or FinalizedEmission. It hands CognitionCandidateBundles off to projection-v0.2 only when every sub-gate fires; otherwise it emits a deterministic CognitionHoldReport with a CognitionFailurePolicy and a CognitionRecoveryAction.


Layer detail: Phase Matrix (PHASEMATRIX-HIVEMIND-03)

The morphodynamic resonance cell substrate that sits below the cognition kernel. Instantiates PhaseCells in a CellPool over a PhaseSubnet, runs each cell's LocalResonanceProcessor to emit ResonancePulses, forms ResonanceClusters through five fail-closed gates (G_cluster, G_morph, G_intent, G_dissolve plus matrix-boundary check), composes a four-edge FunnelGraph (Spatial / Temporal / Semantic / Resonance, validated acyclic by WHITE/GRAY/BLACK DFS), advances a MorphodynamicField (H = α · Φ + β · µ), produces a ConvergenceField and an IntentCandidate, and finally compacts the working state into a ClusterTrace + DissolutionReport while preserving trace, evidence and lifecycle history (the Dissolution-Grundsatz). The substrate emits CellToHandoffCandidates only — no SemanticCrystal, no FinalizedEmission — so the PSE-Bridge remains the only commit path.

The Dual-Fabric Field-Tensor Stitch Layer (PHASEMATRIX-HIVEMIND-03.1) extends this with a persistent FieldTensorState (Fabric-T) and an ephemeral ResonanceFabricState (Fabric-H). Fabric-H is derived deterministically from each CellSubstrateCycleReport and may never directly mutate Fabric-T (Invariant 1 / isolation guarantee). Fabric-T evolves exclusively through accepted CouplingUpdates, gated by the seven-sub-gate StitcherGate conjunction G_stitch = G_conv ∧ G_mci ∧ G_delta ∧ G_budget ∧ G_trace ∧ G_boundary ∧ G_evidence (fail-closed). Every accepted update increments tensor_revision and extends the FieldTensorTrace append-only audit log (the Dissolution-Grundsatz equivalent for the stitch layer).


Layer detail: Topology (PSE-TRAVERSE-TPT-MTL-04)

Topological Panoptic Triangulation and Möbius-Tripolar Micro-Lift (TPT-MTL). Instead of point-wise projections it triangulates a 5D phase-space window into a MeshHolo, lifts each point through a PrimaryPhaseDualAntiphase (MTL-D1 Möbius reflection) → SeamComponent triple, guards every mesh mutation with a TopologyGuard (Betti-shift ∈ AllowedShift ∧ W_p ≤ θ_PD), and gates the resulting TopologicalCrystalCandidate through thirteen fail-closed gates before materialising a TptMtlBundle:

PhaseSpaceWindow  →  AxisBridgeReport  (semantic / runtime / carrier separation)
                  →  MeshHolo          (seeded + evolved under TopologyGuard)
                  →  MicroFiber[]      (primary + dual + seam per point)
                  →  CarrierReport     (I-06: null_center is stateless)
                  →  ReinterpretationReport  (Betti numbers → claim candidates)
                  →  TptMtlGateReport  (13 gates: Adapter/Axis/Symmetry/Topology/
                                        Entropy/MicroLift/Carrier/Kairos/Truth/
                                        Boundary/Replay/Matrix/Emission)
                  →  TopologicalCrystalCandidate  (NOT a SemanticCrystal)
                  →  TptMtlBundle      (content-addressed, fully replayable)
                  →  ReplayManifest    (5-digest verification anchor)

Conformance class TPTM-5. Ten invariants (I-01 … I-10) enforced at every stage. The layer never produces a SemanticCrystal; the existing PSE-Bridge remains the sole commit path.


Layer detail: NCTCS (PSE-NCTCS-CONFORMANCE-01)

Null-Centered Toroidal Control Closure Layer. Classifies conformance C0–C4 from a deterministic pipeline:

NullCenterRef  →  NullProjectionAudit  (K0 ≠ π0(K0))
               →  PhaseVisibilityAudit
               →  CandidateFormationAudit
               →  MaterializationAudit  (no direct fabric→tensor mutation; fail-closed)
               →  TraceReplayContractReport
               →  classify_conformance  (C0–C4, cumulative prerequisite chain)
               →  MacroControlState     (C4 only; derived from null_center + tensor
                                         + trace, NEVER from resonance or coherence)
               →  NctcsClosureBundle    (content-addressed, byte-identical replay)

Conformance classes are cumulative: reached_class is the highest class whose prerequisite chain is fully satisfied. A MaterializationAudit artifact recording no_direct_fabric_to_tensor_mutation=false overrides a conformance report summary that records c2=trueArtifact Supremacy (ADAMANT Axiom 6.1.1) applies.


Layer detail: Metatron (PSE-METATRON-MONOLITH-01)

Holistic Eigenmode Closure Layer. Evaluates the fail-closed composite gate G_meta = G_nctcs ∧ G_trace ∧ G_replay ∧ G_iso ∧ G_gap ∧ G_eval ∧ G_drift over the full PSE stack. HolisticEigenmodeState (M0–M5 conformance, content-addressed, replayable) is produced only when G_meta = 1; gate failure yields a MetatronDiagnosticReport (fail-closed, no HolisticEigenmodeState under any circumstances).

G_iso requires ≥ 1 IsomorphicProjectionReport with passed = true; an empty list unconditionally fails G_iso. isomorphism_score is a computed metric and does not override the passed determination. MetatronOperator is not a controller — it derives state from the persistent field tensor and MacroControlState, never from the ephemeral resonance field.


Layer detail: Eval Matrix (PSE-EVAL-MATRIX-01)

The eval matrix wraps every post-symbolic layer into a structural research instrument — not a benchmark harness. It binds system variants (B0_Baseline … B7_FullStack), workload families, domain datasets, metric specs, calibration states, ablations, replay verification, and statistical aggregation into one reproducible tensor.

EvaluationSpec ─┐
                ├─ plan ──► EvaluationPlan
DatasetManifest┤            │
GroundTruthProf┤            ├─ run ──► EvaluationRunLedger (append-only,
SystemVariant  │            │            hash-chained)
WorkloadSpec   │            │
MetricSpec     │            └─ TrialReport[]  (content-addressed,
CalibrationProf┘                              JCS-canonical, replayable)
                                              │
                                              ├─ replay  ── byte-identity check
                                              ├─ score   ── CapabilityProfile,
                                              │            Safety-Adjusted Utility,
                                              │            Layer Marginal Utility
                                              ├─ ablate  ── B6 → noCognition, …
                                              └─ report  ── Markdown / JSON

Schlussformel (§23): a system counts as empirically improved only when ΔU_task > 0 ∧ ΔU_safety ≥ 0 ∧ ReplayIdentity = 1 ∧ InvalidRunRate ≤ ε ∧ LMU_target > 0 — otherwise the result is a diagnostic finding, surfaced as a ConclusionFlag.


CLI reference

# Inspect the structured problem space
cargo run --release -p pse-traverse-cli -- inspect \
    --problem crates/pse-traverse/examples/problem_minimal.json

# Generate a deterministic CollapsePlan
cargo run --release -p pse-traverse-cli -- plan \
    --problem crates/pse-traverse/examples/problem_minimal.json \
    --out target/traverse/plan.json

# Plan + full signature layer (operator → signature → diagnostics → gate)
cargo run --release -p pse-traverse-cli -- plan \
    --problem crates/pse-traverse/examples/problem_minimal.json \
    --signature

# Full run including a PSE-bridge commit attempt per required dimension
cargo run --release -p pse-traverse-cli -- run \
    --problem crates/pse-traverse/examples/problem_minimal.json \
    --out target/traverse/run.json

# Run with SignatureGate as a diagnostic channel on every GateReport
cargo run --release -p pse-traverse-cli -- run \
    --problem crates/pse-traverse/examples/problem_minimal.json \
    --signature-gate --out target/traverse/run_sig.json

# Verify byte-identical replay
cargo run --release -p pse-traverse-cli -- replay \
    --run target/traverse/run.json

# Generate traversal blueprints (DeterministicGrid search)
cargo run --release -p pse-traverse-cli -- search \
    --problem crates/pse-traverse/examples/problem_minimal.json \
    --n 8 --out target/traverse/blueprints.json

# Build a content-addressed HorizonChart (PSE-TRAVERSE-HORIZON-03)
cargo run --release -p pse-traverse-horizon-cli -- chart \
    tools/pse-traverse-horizon-cli/tests/fixtures/input_minimal.json \
    --rd tools/pse-traverse-horizon-cli/tests/fixtures/rd_minimal.json \
    --out target/horizon/chart.json

# Run the horizon pipeline and finalize (only when G_v0.2 ∧ G_v0.3 = 1)
cargo run --release -p pse-traverse-horizon-cli -- finalize \
    tools/pse-traverse-horizon-cli/tests/fixtures/input_minimal.json \
    --rd tools/pse-traverse-horizon-cli/tests/fixtures/rd_minimal.json \
    --out target/horizon/final.json

# Verify byte-identical replay of the certificate
cargo run --release -p pse-traverse-horizon-cli -- replay \
    target/horizon/final.json \
    --rd tools/pse-traverse-horizon-cli/tests/fixtures/rd_minimal.json

# Run the panoptic phase cognition kernel (PSE-TRAVERSE-COGNITION-01)
cargo run --release -p pse-traverse-cognition-cli -- observe \
    tools/pse-traverse-cognition-cli/tests/fixtures/input_minimal.json \
    --rd tools/pse-traverse-cognition-cli/tests/fixtures/rd_minimal.json \
    --out target/cognition/run.json

# Inspect the 720° panorama (current + counterfactual horizons)
cargo run --release -p pse-traverse-cognition-cli -- panorama \
    tools/pse-traverse-cognition-cli/tests/fixtures/input_minimal.json \
    --rd tools/pse-traverse-cognition-cli/tests/fixtures/rd_minimal.json \
    --out target/cognition/panorama.json

# Hand off the candidate bundle (only when every sub-gate fires)
cargo run --release -p pse-traverse-cognition-cli -- bundle \
    target/cognition/run.json \
    --out target/cognition/bundle.json

# Stamp an evaluation spec from a built-in preset (PSE-EVAL-MATRIX-01)
cargo run --release -p pse-eval-matrix-cli -- init \
    --template agent-cognition --out target/eval/spec.json

# Plan, run, replay, score and report the full pipeline
cargo run --release -p pse-eval-matrix-cli -- plan   --spec target/eval/spec.json   --out target/eval/plan.json
cargo run --release -p pse-eval-matrix-cli -- run    --spec target/eval/spec.json   --plan target/eval/plan.json   --out target/eval/bundle.json
cargo run --release -p pse-eval-matrix-cli -- replay --bundle target/eval/bundle.json
cargo run --release -p pse-eval-matrix-cli -- score  --spec target/eval/spec.json   --bundle target/eval/bundle.json --out target/eval/summary.json
cargo run --release -p pse-eval-matrix-cli -- report --summary target/eval/summary.json --format md --out target/eval/summary.md

# Run a full PHASEMATRIX-HIVEMIND-03 cell-substrate cycle and verify replay
cargo run --release -p pse-phase-matrix-cli -- cluster-cycle \
    target/phase/input.json --rd target/phase/rd.json \
    --out target/phase/cycle.json
cargo run --release -p pse-phase-matrix-cli -- cluster-replay \
    target/phase/cycle.json

# Run the NCTCS closure layer (PSE-NCTCS-CONFORMANCE-01)
cargo run --release -p pse-validation-runner-cli -- nctcs-close \
    target/run_dir/ --out target/run_dir/nctcs_closure_bundle.json
cargo run --release -p pse-validation-runner-cli -- nctcs-replay \
    target/run_dir/nctcs_closure_bundle.json
cargo run --release -p pse-validation-runner-cli -- nctcs-verify \
    target/run_dir/nctcs_closure_bundle.json

# Run the Metatron holistic eigenmode closure (PSE-METATRON-MONOLITH-01)
cargo run --release -p pse-metatron-cli -- inspect target/run_dir/
cargo run --release -p pse-metatron-cli -- close   target/run_dir/ \
    --out target/run_dir/metatron_closure_report.json
cargo run --release -p pse-metatron-cli -- verify  \
    target/run_dir/holistic_eigenmode_state.json
cargo run --release -p pse-metatron-cli -- replay  \
    target/run_dir/metatron_closure_report.json

# Run the domain validation layer (PSE-VALIDATION-RUNNER-DOMAIN-01)
cargo run --release -p pse-validation-runner-cli -- run \
    --profile domain \
    --domain-manifest validation_domains/embedded_ground_truth/manifest.json \
    --out validation_runs/domain_run

# Bench_gt JSON output for a single scenario:
cargo run --release -p pse-bench-gt --bin bench_gt -- \
    --scenario seismo --format json --out /tmp/seismo.json

Workspace layout

28 crates, 10 domain adapters, 11 tool binaries:

crates/
  pse-types       Schema (Observation, Crystal, GateSnapshot, …)
  pse-graph       Persistent graph + Observation → vertex projection
  pse-extract     Constraint program / inverse-weave / DoF analysis
  pse-cascade     Mandorla, helix-pair, cascade operators (DK/SW/PI/WT)
  pse-evidence    Crystal construction, evidence chain, content address
  pse-replay      Deterministic replay & verification
  pse-constraint  Intrinsic step, morphogenic update
  pse-registry    Crystal registry / lookup
  pse-manifest    Run descriptors / provenance
  pse-capsule     AES-256-GCM sealed transport (counter-reuse detector)
  pse-scheduler   Tick orchestration
  pse-topology    Laplacian, Fiedler, Betti, spectral gap
  pse-store       Persistent crystal store
  pse-scale       Multi-scale state (Micro/Meso/Macro)
  pse-pmhd        Path-Minimum Hierarchical Decomposition
  pse-navigator   TRITON spiral / SimplexMesh / singularity scan
  pse-swarm       Multi-agent crystal propagation
  pse-memory      Pattern-memory index (cross-session)
  pse-net         Network transport (feature-gated)
  pse-wasm        WebAssembly bindings
  pse-gateway     HTTP gateway
  pse-cli         CLI front-end
  pse             Meta-crate
  pse-core        Engine orchestrator (`macro_step`), DomainAdapter trait,
                  AdaptiveCalibrator, operator algebra, falsifier
  pse-metatron    Periodic Table of Graphs (Metatron Scan, n ≤ 8) +
                  PSE-METATRON-MONOLITH-01 Holistic Eigenmode Closure
                  Layer (closure/ submodule): LocalMonolithProjection →
                  IsomorphicProjectionReport → SpectralGapStitchReport →
                  MetatronGateReport (G_meta = G_nctcs ∧ G_trace ∧
                  G_replay ∧ G_iso ∧ G_gap ∧ G_eval ∧ G_drift,
                  fail-closed) → HolisticEigenmodeState (only on pass;
                  content-addressed, replayable, M0–M5 conformance class)
  phase-matrix    PHASEMATRIX-HIVEMIND-03 morphodynamic resonance cell
                  substrate + PHASEMATRIX-HIVEMIND-03.1 Dual-Fabric
                  Field-Tensor Stitch Layer
  pse-validation-runner
                  Validation runner + PSE-NCTCS-CONFORMANCE-01 closure
                  layer + PSE-VALIDATION-RUNNER-DOMAIN-01
  pse-traverse    PSE Traversal Agent v0.1 + Signature + Dynamics + Horizon
                  + Cognition (PSE-TRAVERSE-SIGNATURE-01,
                  PSE-TRAVERSE-DYNAMICS-01, PSE-TRAVERSE-HORIZON-03,
                  PSE-TRAVERSE-COGNITION-01)
  pse-eval-matrix PSE-EVAL-MATRIX-01 evaluation matrix

adapters/
  pse-adapter-binance     Crypto markets (Binance OHLCV)
  pse-adapter-vitals      Medical (ECG / vital signs)
  pse-adapter-seismo      Seismology (USGS earthquakes)
  pse-adapter-weather     Weather (Open-Meteo)
  pse-adapter-airquality  OpenAQ
  pse-adapter-entsoe      Energy grid (ENTSO-E)
  pse-adapter-iot         Predictive maintenance
  pse-adapter-syslog      Syslog / security ops
  pse-adapter-tabular     CSV / tabular
  pse-adapter-modelmon    ML model monitoring

tools/
  pse-bench-gt        Ground-truth precision/recall (PSE vs STL-z-score vs IsoForest)
  pse-bench-bbo       TRITON spiral vs Random vs Halton on BBO test functions
  pse-audit           Determinism / replay auditor
  pse-demo            30-second runnable showcase + gate diagnostics
  pse-traverse-cli    Traversal Agent CLI: inspect / plan [--signature] /
                      run [--signature-gate] / replay / search /
                      dynamics (init | tick | run | replay | inspect)
  pse-traverse-horizon-cli
                      PSE-TRAVERSE-HORIZON-03 CLI:
                      inspect / chart / rays / crossing / finalize /
                      replay / verify (binary: pse-traverse-horizon)
  pse-traverse-cognition-cli
                      PSE-TRAVERSE-COGNITION-01 CLI:
                      inspect / observe / state5 / memory-query /
                      lattice / puzzle / panorama / calibrate /
                      trigger / bundle / replay / verify
                      (binary: pse-traverse-cognition)
  pse-eval-matrix-cli  PSE-EVAL-MATRIX-01 CLI:
                      init / validate / plan / run / replay / score /
                      ablate / compare / report
                      (binary: pse-eval-matrix)
  pse-phase-matrix-cli PHASEMATRIX-HIVEMIND-03 CLI:
                      cell-pool / cluster-cycle / cluster-replay /
                      cluster-verify / stitch-fabric / stitch-candidates /
                      stitch-gate / stitch-apply / stitch-cycle /
                      stitch-replay / tensor-inspect
                      (binary: phase-matrix)
  pse-validation-runner-cli
                      Validation runner CLI:
                      nctcs-close / nctcs-replay / nctcs-verify
                      (binary: pse-validation-runner)
  pse-metatron-cli    PSE-METATRON-MONOLITH-01 CLI:
                      inspect / project-local / isomorphism /
                      spectral-gap / close / replay / verify
                      (binary: pse-metatron)

Specifications

Document Covers
ADAMANT_v1.0.0.pdf Constitutional architectural contract (all layers)
pse_traversal_agent_spec_v0_1_REUPLOAD.pdf Traversal Agent v0.1
pse_traverse_signature_spec.pdf PSE-TRAVERSE-SIGNATURE-01
pse_traverse_dynamics_spec_v0_1.pdf PSE-TRAVERSE-DYNAMICS-01
pse_traverse_horizon_spec_v0_3.pdf PSE-TRAVERSE-HORIZON-03
pse_traverse_cognition_spec_v0_1.pdf PSE-TRAVERSE-COGNITION-01
PHASEMATRIX_HIVEMIND_03.pdf PHASEMATRIX-HIVEMIND-03 / 03.1
pse_nctcs_conformance_spec_v0_1.pdf PSE-NCTCS-CONFORMANCE-01
PSE-METATRON-MONOLITH-01.pdf PSE-METATRON-MONOLITH-01
PSE_EVAL_MATRIX_01.pdf PSE-EVAL-MATRIX-01
topologisches_traversierungsframework_v3.pdf Underlying topological framework

Where to go next

  • docs/POST_SYMBOLIC.md — what post-symbolic computation is as a category, why it's distinct, what it can and cannot do.
  • docs/COMPLIANCE.md — EU AI Act formal compliance proof sketch.
  • cargo run --release -p pse-demo — see the core engine end-to-end, with gate diagnostics, in under a minute.
  • Calibrating — pick a domain (or use one of the ten shipped adapters), feed real observations through, watch state.last_gate, tune. The diagnostic surface is the lever.
  • Extending — implement ObservationAdapter for your data source; optionally implement DomainAdapter for domain-specific vocabulary. The trait surface is two methods.

Quality / production-readiness

Check Command Status
Compiler warnings RUSTFLAGS="-D warnings" cargo build --workspace --all-targets --locked clean
Format cargo fmt --all -- --check clean
Lints cargo clippy --workspace --all-targets --locked clean (default level)
Tests cargo test --workspace --locked 928 / 928 passing
Doc build RUSTDOCFLAGS="-D warnings" cargo doc --workspace --no-deps --locked clean
Reproducible builds Cargo.lock is committed; binaries are --locked enforced
CI GitHub Actions: fmt + clippy + build (Linux/macOS/Windows) + test + doc + audit .github/workflows/ci.yml
Dependency updates Dependabot (weekly Cargo, monthly Actions) .github/dependabot.yml
Security policy Private vulnerability reporting + threat model SECURITY.md

See CONTRIBUTING.md for the development loop and the determinism / replay rules every contribution must respect, and CHANGELOG.md for the release log.


Author

Sebastian Klemm

License

MIT

About

Post-Symbolic Exoskeleton (PSE) is a post-symbolic cognitive substrate — a multi-layer formal architecture that sits between purely symbolic (GOFAI) and purely sub-symbolic (connectionist) computation, reducing to neither.

Topics

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages