AI Anomaly Detection (M50020)
| Metric ID | M50020 |
|---|---|
| Category | AI |
| Access | public |
| Classification | proprietary |
| Computation | AI + Classical + Quantum (simulated) |
| Computed by | Black Belt Labs |
| Freshness | fresh |
| Last refreshed |
Calculation methods
| Method | Last refreshed |
|---|---|
| Classical | |
| AI | |
| Quantum (simulated) |
This is a proprietary Madjik metric. Its methodology is a trade secret; the values are available through the API.
Quantum configuration
Every quantum calculation of this metric, with the machine and circuit setup it runs on. Hardware values come from real IBM quantum computers; simulator values from a classical simulation of the same kind of circuit.
| Method | Quantum computing simulated (quant-sim) |
|---|---|
| Provider | Qiskit Aer simulator (IBM open-source software) on Google Cloud |
| Machine | aer_simulator |
| Machine type | classical simulator (CPU, noise-free) |
| Qubits used | 5 |
| Circuits per run | 10 |
| Shots per run | 20,480 |
| Circuit depth | 1227 |
| Two-qubit gates | 747 |
| Gate basis | u, cx (Aer, flattened) |
| Algorithm | amplitude estimation (MLAE) of the |z|>2 exceedance mass across live metrics |
| Profile | full |
| Error mitigation | none (noise-free simulation) |
| How measured | recorded circuits, compiled to u/cx |
Personal metrics: figures are the maximum across submitted portfolios (circuit size follows portfolio size). Machine qubits is the size of the whole chip; qubits used is what this calculation needs.
Method comparison (experimental)
The same metric computed several ways on the same frozen inputs, one change at a time, so a gap between two rows has a known cause. Each plus/minus is the standard error of that value. How to read this, and how it scales.
Daily simulator-only comparison
Snapshot 20261016T0003_444fcc8e · 2026-10-16 00:03 UTC
| Step | Method | Value |
|---|---|---|
| Classical, exact | classical | 16.7 |
| Production value (-5.8 h from the snapshot) | AI | 75 |
| Production value (-0.7 h from the snapshot) | classical | 18.5 |
| Quantum simulator, ideal | quantum simulated | 16.7 |
| Quantum simulator, sampled | quantum simulated | 16.7 ± 0.07355 |
Scored against what actually happened
Alert score 0-100. Outcome: 1 if BTC's next-24h move exceeds twice its trailing 30-day daily volatility. AUC: 0.5 means no skill, 1.0 is perfect.
| Method | Result |
|---|---|
| AI | AUC 0.44 over 259 forecasts (14 large moves) |
| classical | AUC 0.51 over 166 forecasts (6 large moves) |
| quantum simulated | AUC n/a over 63 forecasts (0 large moves) |
Quantum values are not expected to beat the classical value at these problem sizes. The purpose of this section is to show, with uncertainty, how and why the methods differ. Small samples: treat scores as indicative.
Hypotheses monitored by this metric
Access this metric
Query it via the Madjik API:
curl -H "X-API-Key: $MADJIK_API_KEY" \
https://api.madjik.io/v1/metrics/M50020