Quantum · QCA · open technical brief · Root-1 · /quantum

Fail-closed architecture for quantum measurement and control.

QCA (Quantum Control Architecture) is a fail-closed architecture for observing, gating, and governing quantum control information — advisory detection and awareness, not a production controller, and not affiliated with or endorsed by IBM.

The demonstrated floor today is the Quantum Control Kernel (QCK): precursor / drift-anomaly detection on cryogenic control telemetry. TRL-4 in simulation and against public IBM Heron read-only calibration data. Self-verifiable.

D0 · Demonstrated TRL-4 D1 · Couple target D2 · Architecture — compile-ready

Horos Labs Ltd · Horos Engineering · info@horosengineering.com · not affiliated with or endorsed by IBM

Depth ladder

Proven floor. Coupling in motion. Architecture as destination.

Each rung carries its own honest evidence bar. Nothing on this page claims the rung above it.

D0

QCK — the sensor

Fail-closed estimator on control and calibration telemetry. Precursor detection of systematic bias drift. The numbers and events below.

● Demonstrated — TRL-4, published bounds
D1

The coupled loop

Non-perturbative observe + QCK + fail-closed gates + loop supervisor on live or partner-authorised feeds. Optional one instrumentation experiment. Success criteria fixed before the run.

● Couple target — in motion, not claimed done
D2

Full QCA

Complete fail-closed product architecture: coherent runtime, equilibrium, provenance memory, lifecycle — plus measurement and integrity planes.

● Architecture — compile-ready geometry, not live middleware

D0 evidence — two legs, one detector

The same governed detector: controlled adversary in simulation, then read-only check on public hardware calibration data we did not generate.

Simulation — Monte Carlo

Detection
96.7%
before saturation · collapse track
Median lead
12 s
INTERDICT to D = 1.0
False positives
0%
across 210 healthy runs
Coverage
7 × 30
environments × seeds · incl. 1/f

30 seeds × 7 sensor-environment conditions × 2 modes (collapse + healthy), including a 1/f dephasing spectrum. False positives count only healthy-signal runs reaching INTERDICT; WATCHLIST / HOLD on healthy tracks are not false positives.

Reference run (seed 0): INTERDICT at t = 226 s; saturation (D = 1.0) at t = 238 s — 12 s lead.

Real data — IBM public calibration (read-only)

Device
ibm_fez
156-qubit Heron · public cal
Event
q126
T1 88.0 → 19.1 µs (−78%)
Spurious flags
0
other 155 qubits silent

On IBM's public 156-qubit Heron calibration history, the observer flagged one real T1 collapse (q126, 88.0 → 19.1 µs, −78%) and stayed silent on all 155 other qubits — zero spurious flags.

Applied read-only to IBM’s published calibration history. Metadata only — no circuits executed, no queue, no cost. One real drift event found in stored history (n=1) — offered for replication, not as a measured false-alarm rate. The statistics live in the simulation; this is the existence proof on public data.

Not affiliated with or endorsed by IBM. Pull the same public telemetry and check the method yourself.

D0 behaviour classes — page-local vocabulary

Four regimes. Four responses. Abstain outside competence.

These names are the QCK page’s own behaviour vocabulary for D0 — not a claim that a full product OS is live.

REGIME 01

Sovereign

→ Deploy

Both gates pass — channel trustworthy. Proceed.

REGIME 02

Critical Fragile

→ Fix-Dampen

Viable but not truthfully responsive — damp before it propagates.

REGIME 03

Committed Stiff

→ Fix-Excite

Responsive but not viable — perturb to re-find truth.

REGIME 04

Drift / Chaos

→ Reset

Both gates fail — reset honestly rather than pretend.

Advisory outputs only. The operator owns the response. The observer’s job is to be right about the regime — and silent when it cannot know.

D1 — couple target: the bounded advisory experiment

Couple target — not claimed done

A bounded advisory experiment on a live or partner-authorised feed — not a deployment.

telemetry / live feeds │ ▼ NPO — non-perturbative observe (zero-write) │ ▼ QCK — detect · estimate · stay silent │ ▼ fail-closed gates — allow · abstain · refuse │ ▼ loop supervisor — stiction · re-evaluation │ ▼ advisory out · session evidence [optional] one instrumentation probe per run

Success criteria are fixed before the run: closed-loop spine on the agreed feed, fail-closed on insufficient evidence, QCK-class behaviour preserved, supervisor catches stiction, evidence pack, no production-controller claim. The partner keeps plant authority throughout.

Fixed before the window

  • Read-only feed class and access boundary
  • Run window and hardware-calibration method
  • Pass, fail, abstain, and stop conditions

Returned either way

  • Configuration, feed identifier, and agreed window
  • Advisory verdicts, refusals, and stiction re-evaluations
  • Hash-stamped evidence pack; no anecdote-only success

D2 — Quantum Control Architecture

Architecture product — compile-ready · not live on partner hardware

The destination geometry: measurement and control under one fail-closed law.

control / cal telemetry │ ▼ NPO ──► QCK ──► [measurement features] ──► WDU gates │ │ └───────── SCU / EHU / memory ◄────────┘ │ ▼ supervisor · LCU │ ▼ advisory outs · session evidence (the operator owns the plant at every depth)
QCKsense — fail-closed precursor detection
NPOobserve — non-perturbative read
WDUdiscriminate — allow / abstain / refuse
SCUhold — coherent runtime and provenance state
EHUbalance — equilibrium / commit discipline
LCUsustain — lifecycle, boot, maturity
Loopsupervise — stiction, stuck-output, re-evaluation
01Phase-space geometry — designed to reconstruct attractors from scalar feeds (Takens) and track chaotic divergence in T1/T2 drift
02Crosstalk & field coupling — designed to separate directed driving (transfer entropy) from noise (random-matrix ceilings)
03Control-loop margins — designed to compute return-difference clearance from pulse/reset instability
04Thermodynamic cost — designed to budget Landauer erasure heat against millikelvin limits
INTSelf-model honesty — audits its own calibration model and downgrades before it misleads
INTFail-closed portability — certified only when relocation leaks zero information; otherwise refuses to port

Architecture-register verbs only ("designed to"). Thresholds: calibrated per hardware during the coupling window — no universal numeric bound stated. External wording: precursor / drift-anomaly detection — not proven pre-collapse prediction.

The architecture is specified in full in HL-RP-QC-003DOI 10.5281/zenodo.21500237 ↗ · PDF ↗ (preprint, CC-BY 4.0).

Claim rails

A fail-closed company makes fail-closed claims.

Honest maturity

Demonstrated

  • Monte Carlo: 96.7% detection, 12 s median lead, 0% FP / 210 healthy runs
  • IBM public cal (read-only): q126 T1 −78%; 0 spurious flags on 155 others
  • Fail-closed / silent-on-stable discipline under published bounds

Not yet demonstrated

  • Operation on a live qubit-control loop inside a dilution refrigerator
  • Integration with an error-correction stack or real-time controller
  • False-positive characterisation across a full processor operational envelope
  • Certification pathway for safety-critical use

Research rail

Open-access preprints (not peer-reviewed), CC-BY 4.0 where stated. Verify DOIs live at use time.

HL-RP-QC-002
Precursor Detection of Systematic Bias Drift in Cryogenic Quantum Control The QCK method — simulation + public Heron read-only check. Quantum branch: instrument.
HL-RP-QC-003
The Quantum Control Architecture The governed layer the kernel belongs to — units, measurement and integrity planes, advisory boundary. Quantum branch: architecture.
HL-RP-FO-001
The Physical Limits of Finite Observation Shared finite-observer discipline beneath quantum, structural, and autonomous builds.
HL-RP-FC-002
Deviation as Located Missingness
HL-RP-FC-003
Forced Partition

Contact

For more information:

info@horosengineering.com

Horos Engineering · horosengineering.com