S1-GRT-JEPA (MatterWaveAI/s1-grt-jepa)
S1-GRT-JEPA is a System-1 Rotational JEPA World-Model Judge, Model-Predictive Control (MPC) Action Router & Agent Safety Guardrail (0 dense linear matrices, 0 convolutional layers) built on the HELMHOLTZ-1 rotational foundation architecture.
Conventional safety classifiers, reward models, and agent routers evaluate prompts or tool calls statically—inspecting surface tokens without simulating what will happen after an action is executed. As a result, static judges are brittle to multi-step side effects, subtle prerequisite violations, and indirect prompt injections hidden inside tool outputs.
S1-GRT-JEPA replaces static classification with latent world-model lookahead: before emitting a single-pass APPROVE / REJECT / ESCALATE decision or selecting the next agent tool call, it rolls candidate actions forward in orthogonal latent space across 1..H steps and evaluates the predicted outcome state against goal and safety invariants—with zero autoregressive token decoding.
Release Note (Initial Preview Card): This card reports held-out
XL-smolS1-GRT-JEPAlatent lookahead benchmarks (12,003held-outXL-smolrecords /29,609questions), collapse-resistance verification, and full 12-domainXL-smolSystem-1 evaluation (17,130held-out test records /44,287test questions). Model weights, architectural implementation details, and source code are withheld during the staged world-model release cycle to prevent white-box guardrail evasion.
Core Capabilities (S1-GRT-JEPA)
- Latent Trajectory Lookahead (
MPCAction Routing,0Autoregressive Tokens):- Given an agent state, goal, and
Kcandidate actions,S1-GRT-JEPApredicts the resulting latent state for each candidate action in a single parallel forward pass and selects the action that minimizes distance to the verified goal state.
- Given an agent state, goal, and
- Latent Prompt-Injection & Agent-Drift Defense:
- Detects adversarial hijack attempts embedded inside external tool observations (e.g., untrusted files, web pages, or logs) by measuring abrupt latent trajectory divergence away from the user's original goal—automatically routing the agent to safe recovery or human escalation (
77.17%top-1 /86.60%pass@2routing accuracy across803held-out trajectories /2,409test questions).
- Detects adversarial hijack attempts embedded inside external tool observations (e.g., untrusted files, web pages, or logs) by measuring abrupt latent trajectory divergence away from the user's original goal—automatically routing the agent to safe recovery or human escalation (
- 100% Option-Order Invariance (
Permutation Consistency = 1.000):- Unlike causal language models used as judges—which flip their verdict up to
65%of the time when multiple-choice options are reordered—S1-GRT-JEPAscores each candidate state transition independently in latent space, guaranteeing exact mathematical invariance (100.0%consistency) under any option permutation.
- Unlike causal language models used as judges—which flip their verdict up to
- Collapse-Free Orthogonal Dynamics (
0Linear Layers):- Every encoder and transition predictor block is a cascade of orthogonal transformations (
det = +1), maintaining99.4%effective rank retention and< 1e-6norm drift across 100 target updates without contrastive negatives.
- Every encoder and transition predictor block is a cascade of orthogonal transformations (
XL-smol Benchmark Results
1. S1-GRT-JEPA: XL-smol Latent Rollout Verification & MPC Agent Routing (12,003 Held-Out Records / 29,609 Questions)
S1-GRT-JEPA XL-smol Evaluation Suite |
Held-Out Records (n) |
Test Questions | Latent Loss Trajectory | Top-1 Accuracy | pass@2 Accuracy |
Option-Order Consistency |
|---|---|---|---|---|---|---|
Multi-Hop Relational Rollout Verification (2..3 Hops) |
2,500 |
5,000 |
3.2542 -> 0.1596 |
93.70% |
97.80% |
100.0% (PASS) |
| Causal Science & Physical Commonsense Verification | 2,500 |
5,000 |
3.1878 -> 0.1146 |
96.40% |
99.10% |
100.0% (PASS) |
| Single-Hop Factual World-Model Verification | 2,500 |
5,000 |
3.1045 -> 0.0982 |
91.20% |
96.20% |
100.0% (PASS) |
4-Mode Socratic & Epistemic Verification (SEE/SAY/DO/KNOW) |
2,500 |
5,000 |
Multi-View Latent | 68.60% |
94.80% |
100.0% (PASS) |
2D Visual-Spatial Grid Rule Induction (ARC-AGI-2 XL-smol) |
1,200 |
3,600 core (7,200 all) |
Latent Rule Induction | 62.67% |
87.50% |
100.0% (PASS) |
| Agent MPC Action Routing, Drift & Prompt-Injection Defense | 803 |
2,409 |
H = 1..2 Lookahead |
77.17% |
86.60% |
100.0% (PASS) |
Multi-Step Goal-Directed Beam Planning (H = 2 Hops) |
Held-Out Action Sequence | Multi-Step Chain | d_geo < 1e-5 |
94.80% |
98.40% |
100.0% (PASS) |
6-Suite XL-smol Aggregate (Macro / Micro) |
12,003 |
26,009 core (29,609 all) |
Orthogonal JEPA | 81.62% / 83.67% |
92.96% |
100.0% (PASS) |
2. Collapse-Free Rotational JEPA Invariants
| Invariant / Diagnostic | Target Gate | S1-GRT-JEPA Result |
Status |
|---|---|---|---|
| Dense Linear / Convolutional Layers | 0 / 0 |
0 / 0 (100% Orthogonal) |
PASS |
| Collapse Resistance (0 Contrastive Negatives) | >= 85.0% rank retention |
99.4% (23.62 / 23.76 erank) |
PASS |
| 100-Step Target Update Orthogonality Drift | < 1e-6 |
1.42e-14 (det = +1.000000) |
PASS |
Long-Horizon Rollout Stability (H = 10 Hops) |
< 1e-5 norm drift |
1.19e-7 (Zero explosion/collapse) |
PASS |
Analytic Inverse Dynamics (0 Gradient Steps) |
Held-out action recovery | 96.20% Top-1 (8.43e-8 analytic residual) |
PASS |
| Non-Commutative Action Sensitivity | Order-sensitive | 98.50% (2.65 divergence) |
PASS |
3. Consolidated 12-Domain XL-smol System-1 Judge Evaluation (n = 17,130 Held-Out Test Records / 44,287 Questions)
Evaluated across the complete 12-domain XL-smol System-1 verification benchmark (65,939 train / 10,563 validation / 17,130 held-out test records comprising 40,687 core test questions and 44,287 total test questions):
System-1 Judge Domain (12-Family XL-smol Suite) |
Held-Out Test Records (n) |
Core Test Questions | Primary Accuracy | Core Question Accuracy | Calibration Error (ECE) |
|---|---|---|---|---|---|
Multi-Hop Relational Rollouts (XL-smol) |
2,500 |
5,000 |
93.70% |
93.70% |
0.0883 |
Causal Science & Commonsense (XL-smol) |
2,500 |
5,000 |
96.40% |
96.40% |
0.0312 |
Single-Hop Factual Verification (XL-smol) |
2,500 |
5,000 |
91.20% |
91.20% |
0.0418 |
4-Mode Socratic Verification (XL-smol) |
2,500 |
5,000 |
68.60% |
68.60% |
0.0842 |
2D Spatial Grid Rule Induction (ARC-AGI-2 XL-smol) |
1,200 |
3,600 (7,200 all) |
62.67% |
62.67% (87.50% pass@2) |
0.0914 |
Agent MPC Action Routing & Drift (XL-smol) |
803 |
2,409 |
77.17% |
66.67% |
0.1840 |
| Epistemic Calibration & Abstention | 80 |
160 |
78.12% |
78.12% |
0.0764 |
| Program Loop Termination Verification | 150 |
600 |
74.00% |
74.00% |
0.0915 |
| Symbolic State Transition Tracking | 60 |
180 |
73.33% |
73.33% |
0.0890 |
Prerequisite Knowledge Graph Battery (v5) |
2,777 |
7,558 |
72.85% |
72.85% |
0.0881 |
| Resource & Concurrency Scheduling | 60 |
180 |
71.67% |
71.67% |
0.0940 |
| Formal Boolean & Truth-Table Verification | 2,000 |
6,000 |
68.45% |
68.45% |
0.1042 |
Full 12-Domain XL-smol Test Split (Aggregate) |
17,130 |
40,687 (44,287 all) |
77.35% Macro |
78.42% Micro (77.64% all) |
0.0887 Mean |
4. SLM-Bench Multiple-Choice & GSM8K Mathematical Verification (13,644 Examples)
Benchmark (30M Rung) |
Split (n) |
Random | Flat GRT-30M |
Rotational Baseline | S1-GRT-JEPA (24h Curriculum) |
S1-GRT-JEPA (10C Horizon) |
|---|---|---|---|---|---|---|
BoolQ (google/boolq) |
val (3,270) |
50.00% |
51.41% |
60.55% |
65.72% |
73.03% |
QNLI (nyu-mll/glue) |
val (1,000) |
50.00% |
51.60% |
54.00% |
77.90% |
86.14% |
ARC-Easy (allenai/ai2_arc) |
test (2,376) |
25.00% |
29.59% |
28.96% |
40.57% |
55.04% |
ARC-Challenge (allenai/ai2_arc) |
test (1,172) |
25.00% |
20.82% |
23.55% |
30.80% |
40.13% |
OpenBookQA (allenai/openbookqa) |
test (500) |
25.00% |
28.20% |
30.00% |
42.80% |
54.35% |
PIQA (baber/piqa) |
val (1,838) |
50.00% |
55.28% |
54.08% |
56.47% |
65.34% |
WinoGrande (allenai/winogrande) |
val (1,267) |
50.00% |
50.28% |
50.99% |
51.70% |
62.85% |
HellaSwag (Rowan/hellaswag) |
val (1,000) |
25.00% |
29.20% |
31.60% |
37.40% |
50.51% |
CommonsenseQA (tau/commonsense_qa) |
val (1,221) |
20.00% |
21.70% |
22.77% |
29.24% |
40.93% |
| 9-Task Macro Average | 13,144 total |
35.56% |
37.56% |
39.61% |
48.07% |
58.70% |
| GSM8K 4-Way Verification | test (500) |
25.00% |
34.20% (PPL 27.99) |
39.20% (PPL 10.65) |
99.20% (PPL 7.60) |
99.20% |
Coming Next: Scaled Rotational JEPA World Models (HELMHOLTZ-100M & HELMHOLTZ-350M)
Building directly on the 10M, 20M, and 30M HELMHOLTZ-1 and S1-GRT-JEPA verification ladder, our next scaled releases expand the collapse-free orthogonal JEPA world model and single-pass System-1 MPC judge:
HELMHOLTZ-100M(~195MDense-Equivalent Capacity — Scaled Rotational JEPA World Model &S1Trajectory Judge):- Scales the collapse-free orthogonal JEPA predictor (
helmholtz-1) and multi-step MPC trajectory judge (s1-grt-jepa) in a deep0-linear-layer rotational architecture—projected on our empirical scaling ladder to reach64.9%SLM-Benchmacro accuracy,89.2%structured knowledge accuracy, and>99%cognitive routing accuracy.
- Scales the collapse-free orthogonal JEPA predictor (
HELMHOLTZ-350M(~920MDense-Equivalent Capacity — Flagship Rotational JEPA Foundation World Model):- Scales multi-view Platonic alignment and long-horizon latent MPC trajectory verification in a flagship
0-linear-layer rotational world model (93.8%structured knowledge accuracy,97.0%cognitive telemetry routing).
- Scales multi-view Platonic alignment and long-horizon latent MPC trajectory verification in a flagship
MatterWaveAI Rotational JEPA Scaling Ladder (10M / 20M / 30M)
- Rotational JEPA & Causal World-Model Core:
MatterWaveAI/helmholtz-1(10M:helmholtz-b0.7.0-10m|20M:helmholtz-b0.7.0-20m) - System-1 Rotational JEPA Trajectory Judge & MPC Router:
MatterWaveAI/s1-grt-jepa(10M:s1-grt-jepa-b0.7.0-10m|20M:s1-grt-jepa-b0.7.0-20m) - 24-Hour Distilled Preview (
GRT-30M):MatterWaveAI/grt-30m - Whitepaper: The Geometric Rotation Transformer (GRT): A Purely Rotational Foundation Architecture (
GRT_Whitepaper.pdf)
License & Citation
Released under the Apache 2.0 license.
@techreport{householter2026grt,
title = {The Geometric Rotation Transformer: A Purely Rotational Foundation Architecture},
author = {Householter, David},
institution = {MatterWave AI},
year = {2026},
url = {https://matterwave.ai}
}