Validation Suite

Section 0 — Reversible Plasma Core (GPU‑Verified Micro‑Artifact)

Before the semantic and scaling validations, the substrate is demonstrated on a minimal reversible operator: a swirling plasma‑like field evolved forward and backward on a toroidal grid. This micro‑artifact is fully reproducible and validated on NVIDIA H100.

docker pull ji3434/plasma-core:latest
docker run --gpus all ji3434/plasma-core:latest

Deterministic · Reversible · Hardware‑verified · Digest‑locked

Metrics (128×128 grid, dt=0.02, 400 forward + 400 backward)

This test demonstrates the same reversible law that powers the 100M‑token semantic validation. The operator maintains coherent plasma‑like structure under forward and backward evolution, with extremely low drift. This dual‑domain invariance—symbolic and physical—is the signature of a true computational substrate.

Section A - Semantic World-Ledger Validation (100M GPT Tokens)

A full 100,000,000-token streaming pass over a GPT-tokenized world-ledger corpus. The operator maintained perfect semantic consistency: zero drift, zero hallucinations, zero contradictions, and stable world-state across entity facts, contradictions, balances, and violations.

Documented behaviors:
- Entity locations tracked across 10K -> 100M tokens
- Contradictions detected deterministically
- Ledger balances updated with no drift
- Violations flagged with perfect consistency
- Identical results across repeated runs

This is not a benchmark.
This is a category separation: deterministic long-range world-state reasoning that transformers are architecturally incapable of.

Section B - Fractal World-Ledger Operator

The world-ledger operator reconstructs entity state, contradictions, balances, and violations across 100M tokens with zero drift. This demonstrates deterministic long-range semantic reasoning.

Observed properties:
- Stable entity state across the entire 100M-token stream
- Contradictions surfaced exactly when ground truth changed
- No hallucinations or spurious transitions
- Stream vs in-memory results identical
- Fully deterministic across repeated runs

Section 1 - Hardware Invariance Trace (T4 GPU)

Hardware invariance test.
A billion-token streaming pass executed on a single Tesla T4 GPU. Validates constant-memory and linear-time scaling. (Not a semantic workload.)

Section 1.5 - NVIDIA Build Hardware Run

NVIDIA Build execution trace.
Validated inside NVIDIA's managed container environment. Confirms hardware-level invariants: deterministic execution, constant memory, and stable scaling. Complements the semantic 100M-token validation above.

Section 2 - Streaming & Throughput

Throughput tests validate the operator's linear-time behavior and constant-memory footprint. These tests measure hardware scaling, not semantic reasoning.

Section 3 - Absorption / Escape Scaling

Scaling tests across 128->2048 grids demonstrate stable absorption, escape behavior, and analytic scaling-law consistency.

Section 4 - Dynamo Stability

The operator stabilizes chaotic dynamo trajectories, producing consistent polarity reversals across 16 independent runs.

Section 5 - Resonance Sweep

Frequency-domain tests show coherent resonance peaks and stable integrated ratios.

Section 6 - Invariance Suite

The invariance suite verifies prefix stability, translation invariance, rotational consistency, and stream vs in-memory equivalence.

Section 7 - Plasma Reversibility Test (NVIDIA Build)

A swirling plasma-like field is evolved forward for 400 steps and then reversed for 400 steps using the fractal operator. The final state returns to the initial configuration with extremely low error, demonstrating reversible transport, energy stability, and shock-preserving dynamics on real NVIDIA hardware.

Traditional solvers accumulate drift, diffuse structure, or collapse under reversal. The fractal operator does not. It maintains coherent plasma-like behavior without global solves, diffusion terms, or corrective noise. This is a hardware-verified physical invariant emerging from the same reversible law that powers the semantic 100M-token world-ledger validation above.

Executing this test on an NVIDIA H100—the most demanding floating‑point environment in the NVIDIA stack—demonstrates architectural robustness at production scale. Stability on H100 is a strong indicator that the operator is not merely a faster transformer alternative, but a general computational substrate capable of stable, reversible evolution across both symbolic and physical domains. The same operator that maintains world-state consistency over 100M tokens also stabilizes plasma-like fields on real silicon. This dual-domain invariance is the foundation of a trillion-dollar compute primitive.

Evaluate Your Workload

See how the substrate performs on your real data. This evaluation path is designed for AI labs, government programs, financial systems, HPC centers, and enterprise engineering teams requiring deterministic behavior, constant-memory scaling, and long-horizon stability.

This workflow supports classified workloads, regulated financial data, long-context AI pipelines, and large-scale simulations. All evaluations are conducted privately and securely.

docker pull ji3434/benchmark:latest
docker run --gpus all ji3434/benchmark:latest

→ Begin an Evaluation