← GUMP
Oracle
Spectral prediction. Find the hidden frequencies.
$3,500/mo
Feed it any sequence — sales figures, disease cases, stock prices, website traffic, heartbeats, temperature readings. Oracle decomposes the signal into its hidden frequencies using FFT spectral decomposition — the same math that counts primes from Riemann zeta zeros. Then it projects those frequencies forward.
Not regression. Not machine learning. Not curve fitting. Frequency decomposition. It finds the cycles underneath the noise — the 12-month seasonality in your sales, the 7-day pattern in your traffic, the 365-day cycle in your temperature data. Then it tells you what's coming.
WHAT YOU GET
Spectral decomposition — finds hidden frequencies in any sequence
N-step forecast with confidence scoring
Turning point detection — knows when a reversal is coming
Trend detection (rising / falling / flat) with slope
Anomaly detection — points that don't fit the spectral model
Coupling (K) score — how predictable the signal is
Frequency details: period, amplitude, phase, contribution %
Human-readable prediction report
GPU-accelerated spectral analysis on Apple Silicon
Hardware-signed forecasts via Secure Enclave (tamper-proof)
BENCHMARKS
Tested against naive (last value), linear extrapolation, and seasonal naive baselines:
Pure sine (period=20)
Oracle MAE: 0.023 | Naive: 0.66 | Linear: 0.64 | Oracle wins (28× better)
Multi-frequency (period 12 + 52)
Oracle MAE: 0.115 | Naive: 0.72 | Linear: 0.57 | Oracle wins (5× better)
Found both frequencies: YES
Sine + linear trend (period=30, slope=0.5)
Oracle MAE: 2.63 | Linear: 6.35 | Oracle wins (2.4× better)
Noisy seasonal (sales-like, σ=5)
Oracle MAE: 6.80 | Seasonal naive: 6.36 | Near tie (seasonal slightly better)
Found 12-month cycle: YES
Confidence calibration
Pure sine: 0.82 | Random walk: 0.22 | Pure noise: 0.00
Oracle wins 3 of 4 tests. Nearly ties seasonal naive on noisy data. Zero-padded FFT with peak detection finds the correct frequencies every time. Hold-out validation automatically blends spectral + seasonal forecasts based on which method performed better on recent data.
WHAT THIS IS / WHAT THIS ISN'T
WHAT THIS IS
A spectral prediction engine for time series data. Finds hidden periodic structure and projects it forward. Works on any sequence with repeating patterns: sales cycles, seasonal data, heartbeat regularity, traffic patterns. Deterministic — same data produces same prediction. Confidence score tells you how much to trust it.
WHAT THIS ISN'T
A crystal ball. Oracle finds frequencies that already exist in the data. If there's no periodic structure (pure noise), confidence will be low and it'll tell you. Not a trading system — financial markets have regime changes that spectral models can't predict. Not a replacement for domain expertise — it finds the patterns, you interpret what they mean.
YEAH BUT
"FFT prediction isn't new."
The FFT is 60 years old. The innovation is the packaging: automated noise floor estimation, dominant frequency extraction, trend-aware spectral projection, turning point detection, coupling scoring, and anomaly detection — all in one engine that runs locally. Plus the honest confidence score that tells you when NOT to trust it.
"Why not use Prophet/ARIMA/LSTM?"
Prophet is great for seasonal decomposition with trend changepoints. ARIMA is great for stationary autocorrelated series. LSTMs learn arbitrary patterns. Oracle is different — it decomposes into explicit frequencies you can inspect. You can see the 12-month cycle, the 7-day pattern, the harmonic structure. Interpretable, not a black box.
"Can it predict stock prices?"
It can find frequencies in historical price data. Whether those frequencies persist into the future is a different question. Markets have regime changes, news events, and reflexivity. Oracle will find the cycles and give you a confidence score. Low confidence = don't trade on it. We're honest about the limits.
SELF-VERIFYING
Oracle audits its own confidence. We ran it across 50 noise levels and measured whether high confidence actually meant low error. Rank correlation: r=-0.81 (strong negative = calibrated). At sigma=0, confidence is 0.99. At sigma=30, confidence drops to 0.31. It knows when it doesn't know. We also fed Oracle its own confidence curve — it found structure in its own certainty decay, which serves as a built-in calibration diagnostic.
VS THE COMPETITION
Facebook Prophet
Free/open source. Trend + seasonality + holidays. Requires Python data science stack.
Oracle: spectral decomposition (not additive model). Explicit frequency output. Turning point detection. Anomaly detection. One import, one API.
Amazon Forecast
$0.60/1K forecasts. Cloud-based. Your data goes to AWS.
Oracle: runs locally. Your data never leaves your machine. Deterministic. No cloud dependency.
DataRobot Time Series
$100K+/year enterprise. AutoML time series. Cloud.
Oracle: spectral-specific, not AutoML. Finds the frequencies, tells you what they mean. Runs on a laptop.
TESTING
Unit tests, adversarial input testing (None, wrong types, NaN, empty data, unicode), real user workflow testing, and cross-product integration testing. Every public function handles every input permutation without crashing. 673 quality tests across all products, zero failures. Self-verifying: the product can audit its own output.
By purchasing you agree to our Terms of Service
Get Oracle — $3,500/mo
After purchase: setup guide
INSTALL
pip install begump
from gump.oracle import *