Horos Engineering · proof of code · quantum control

Run the detector yourself

The element of the quantum rail — what it is built from, what is gated, and the three records it stands on.

This is the drift test from the quantum control kernel, running in your browser against the published calibration history of ibm_fez — a 156-qubit IBM Heron processor. Move the parameters. Watch what fires and what it costs. Paste your own series if you have one.

167 daily requests over 2026-03-04 to 2026-08-17, all returned, resolving to 159 distinct served calibrations after de-duplicating repeated provider responses. Read-only: no circuits were executed and no control loop was touched. The archive is hashed and the pull is reproducible from the public record — the numbers below are recomputed in this page from the raw matrix, not quoted.

The instrument #

Detector form

Window

Qubit

Median filter width 5

Drop below baseline 50%

Persistence, consecutive 3

Baseline length 10

Your own series

this qubit

—

baseline median

—

floor

—

final

—

What it costs across the whole device #

qubits firing, of 156

—

at this setting

—

drop threshold	persistence 2	persistence 3	persistence 5	persistence 8

The declared row is highlighted. The record's stated result — exactly one qubit — is at that threshold over the June frame; over the full 159-snapshot record the same setting fires on more, and that widening is the cost of a longer window, stated rather than hidden. Loosen it and the device lights up. Tighten it and the brief excursions vanish. There is no setting that catches everything and stays quiet — that trade is the instrument, and it is why the kernel's constants are declared in code before the results they govern rather than tuned afterwards.

Reproduction check #

What this is not #

This is a demonstration of the detector family on real device metadata. It is not the estimator deployed on control hardware, and nothing here writes to any plant. The published record labels this stratum for exactly what it is.

It fires on change, not on badness. A qubit that has always been poor stays silent — try q72, whose baseline sits near 10 µs and which never fires however you set the threshold. That asymmetry is the point of the test, and an absolute threshold would destroy it.

The advisory tier chatters. On the simulated run of record every healthy run shows watchlist excursions; that tier is a trend surface, not a calibrated alarm, and its redesign is declared frontier work rather than claimed as solved.

Detector: a persistence-gated drift test on per-qubit history, applied read-only — a median filter over a moving window, a threshold set as a fractional drop below the baseline median, and onset only after P consecutive exceedances. (No MAD gate: that is the structural rail's statistic, not this one.) Data: ibm_fez served calibration history via qiskit-ibm-runtime, archived 2026-08-17. The record that states this result carries its own hash and cites the archive by SHA-256.

ArchitectureHoros ODC — observation, decision and control architectureFinite cognitionTry to get something past the governorStructuralRun the structural precursor detectorResearchThe programme — a premise, nine records, and what is still openVisionThe question the architecture was built to ask

Horos Engineering is Horos Labs Ltd, a company registered in England and Wales, number 17013272, Birmingham. Nine records published as one set, CC-BY 4.0 preprints, not peer-reviewed; each names its eight sisters by SHA-256 so the set can be checked rather than taken. Backed by a portfolio of filed UK patent applications (priority March 2026, pending), alongside trade-secret integration and calibration know-how. Not affiliated with or endorsed by IBM or NASA.

© 2026 Horos Labs Ltd · info@horosengineering.com