DiscretaEngine
Anomaly Detection/v1.1/MMXXVI

See what’s actually happening
in your signal.

Anomaly detection from first principles. No training. No tuning. Orders-of-magnitude faster. Deployable on workloads where the established methods aren’t practical.

Training Required
None
zero calibration, zero labels
Throughput
586K
samples/sec, single core
vs. Top Detectors
2,000×
faster than HTM
Compute Savings
90%+
vs. Isolation Forest at scale
Operating Principles

Three properties that make the engine different from everything else in the space.

§01

No training phase.

Every other method requires labeled examples, assumes a statistical distribution, or needs a training window on your data. The engine does none of these. It analyses the geometric structure of your signal directly. A fresh signal in a domain you’ve never seen can be analyzed in milliseconds with no calibration.

§02

Built from first principles.

The detection method was derived from a fifteen-year investigation into how physical reality organizes itself — not optimized against benchmarks. The math comes from the theory. The benchmarks confirm it works.

§03

Orders-of-magnitude efficient.

586,000 samples per second on a single CPU core. 2,000× faster than HTM. No GPU required. No cloud dependency. Monitoring 10,000 metrics with Isolation Forest costs ~$2,500/month in compute. With Discreta, the same workload runs on a single core for under $200.

Live Analysis

Pass any signal through the engine. Watch it separate structure from noise.

Choose a sample signal or upload your own CSV with an API key. No data is stored, no training occurs, no calibration is required.

Upload Your Data
Drop a CSV here
or click to browse
API Key (optional)
Don’t have one? Request access below.
— or try a sample —
Sensitivity3.0
lower = more events
higher = only extremes
Signal Analysis
Infrastructure (CPU) · 512 pts · db4 · sens 3.0
0n=512
Select a signal and click Run Analysis
Validation

Tested against the industry standard benchmark.

NAB (Numenta Anomaly Benchmark) is the primary standard for evaluating anomaly detectors. Getting on the leaderboard at all places Discreta among the most capable detection systems ever built. Every other detector on this list required training on the data before producing results. Discreta ran cold — no training, no tuning.

DetectorStandardLow FPLow FNThroughput
Numenta HTM
neural temporal memory
70.5
62.6
75.2
~300 samp/s
CAD OSE
conformity anomaly detection
69.9
67.0
73.2
earthgecko Skyline
rolling statistical tests
58.2
46.2
63.9
~2K samp/s
KNN-CAD
nearest-neighbor conformal
58.0
43.4
64.8
Relative Entropy
information-theoretic
54.6
47.6
58.8
Discreta
continuous-discrete decomposition · v1.1
51.9
39.3
61.5
586K samp/s
Random Cut Forest
Amazon · ensemble trees
51.7
38.4
59.7
~5K samp/s
Twitter ADVec
seasonal decomposition
47.1
33.6
53.5

NAB scores use Numenta’s native scoring methodology, which rewards early detection and penalizes false positives. On the throughput axis, Discreta occupies a region no other entry is near.

API Access

Request an API key and run the engine against your own data.

The engine is entering production this year. API access is opening to a select group of companies in infrastructure monitoring, industrial operations, and fintech observability.

Your information is used only for this conversation. Not shared. Not sold.