Nuviric Pilot — Verification

What we claim. What we don’t.

Nuviric Pilot is a receiver-side execution layer for IBM Heron processors: it characterizes the machine fresh, in the same sitting, plans against that live reading, and decodes accordingly. This page states exactly what the record supports — and exactly what it does not. The method is a trade secret and is not described here; the record is published so anyone can check what it does without taking our word for anything.

The record: pre-registered tests on IBM Quantum hardware, December 2025 – October 2026, across the Torino, Fez, Marrakesh, and Kingston backends. Every run is timestamped in IBM’s own job ledger. At the close of the October work, the full research program stood at 5,053 of 10,800 QPU seconds consumed on the account’s annual open-plan allocation.

Published record

Reading a Drifting Machine — Richard Vigslist, Nuvira Technologies. Preprint, published 15 October 2026, open access (CC BY 4.0). DOI: 10.5281/zenodo.23405654. The paper carries the fleet results, the two-machine exchange, the blind structure test, and the turnover census, with the IBM job IDs in its data-availability section. It is a timestamped record of what was done and when — the claims below are the same claims, sized the same way.

What is claimed

One receiver, three machines (2026-10-15). Under pre-registered success lines, on one identical fixed battery (six messages, 17 positions each, 36 message instances per arm), the finished receiver delivered 36 of 36 on ibm_kingston, 36 of 36 on ibm_marrakesh, and 35 of 36 on ibm_fez. Same-session textbook-plan controls on the identical battery delivered between 2 of 36 and 30 of 36, depending on machine and session. The single Fez miss failed in every receiver variant — one noisy flight, not a decode error — and the record is stated as 35 of 36, not rounded up.
Plans do not travel between machines. When the Kingston-built plan set was flown unchanged on ibm_marrakesh, both arms delivered 0 of 36 messages, even though 94% of individual positions still read correctly. What transfers is the method of reading the machine in front of you, never a map of a different machine.
Two machines exchanged symbol streams (2026-10-15). Kingston and Marrakesh each introduced themselves in a 17-symbol stream and answered across every hop of a six-stage relay: 204 of 204 positions delivered, zero substitutions. This is a delivered-content result under the stated protocol — it is not a claim of networking, entanglement, or machine understanding.
The symbol structure survives to a stranger (2026-10-15). An independent decoder given only raw counts from Marrakesh flights — no labels, no method — passed discreteness, passed the alphabet-size test on the second tier (said 8, truth 6), and strong-passed grouping at a pooled 0.971 against a 0.90 bar. Structure survives; meanings do not transfer without a teaching package.
The machine drifts, measurably and fast. Comparing three Marrakesh characterizations taken on 2026-10-15, only 2 of the 17 clearest wires survived between sittings a few hours apart, and only 1 from early morning to afternoon, while overall wire quality barely moved. Characterization is perishable; the receiver assumes same-sitting freshness because the record says it must.
The earlier compensation result stands as stated. The 2026-09-25 to 2026-10-13 work on ibm_kingston, physical qubit 0 — compensation beating textbook angles at shot-noise scale, driver-independent convergence on the 0.61–0.70 neighborhood, and measurable drift within a single night — remains part of this record and is graded in the preprint’s context sections.

What is NOT claimed

No speedup, no quantum advantage, no capacity or Holevo claim. Nothing here is a comparison against IBM’s circuit-level resilience stack or any commercial mitigation product; no published head-to-head under aged-characterization drift exists, including ours against them.
No claim the battery generalizes to arbitrary circuits. The battery is a fixed symbolic test set (product circuits, no entanglement). It measures delivered-message survival under drift; it is not a benchmark for general quantum workloads.
No claim of networking or understanding. The two-machine exchange moved symbol streams through a designed relay protocol. The machines did not network, and no claim is made that any system understood the content.
No peer review, and independent reproduction is offered, not completed. Verification is tiered: anyone may re-grade the published raw counts (no permission needed); named researchers, selected by the author, may reproduce on their own accounts under agreement, with verdicts publishable by either party.
Simulation is a gate, not evidence. Where simulation and hardware have disagreed in this program, this page follows hardware.

The language result is a separate artifact

The Nuviric language results — its codec, error-detection discipline, and confirm handshake, graded PASS under their own pre-registered tests on 2026-10-13 — are about the language system. They are not evidence for the receiver claims above, and they are not offered as such. The two bodies of work share hardware and a week; they do not share claims.

The bar for “works”

Nuviric Pilot works when an independent party, under a written agreement with success criteria fixed in advance, reproduces the delivered rates on their own backend — and reports it, pass or fail. Negative results are first-class results. Paid validation engagements are the route in; the method itself is delivered only as a hosted function, at no tier. Contact: richardvigslist@nuviratechnologies.io

Nuviric Pilot — Nuvira Technologies. Claims on this page are sized to the preserved record as of October 15, 2026. If the record changes, this page changes.