YANBOLU / QPU EVIDENCE INSTRUMENTCALIBRATED PROTOCOL · COMMERCIAL PILOT
Independent measurement / failure-preserving protocol

Reproduce.
Refuse.
Record.

Did the quantum claim survive a frozen independent test on real hardware? One claim enters. A checkable evidence record comes out, whether the answer is yes or no.

Request auditRead instrument output
CHANNEL RECONSTRUCTION / AUDIT TRACERUN 016
NONEMEASURE_MOREREFUSEEVIDENCE STRENGTH →
unsafe inverse0
boundary exactFAIL 11/12
failure retainedTRUE
01 / DEFINEExact claim and failure condition.
02 / QUALIFYQPU controls before main spend.
03 / FREEZECriteria before evidential run.
04 / EXECUTEHardware run with controls.
05 / VERIFYWorkloads, hashes and findings.
Decision system / 01

Not every quantum error should be “corrected.”

A credible system first asks whether inversion is physically justified. Refusal and uncertainty are valid outputs, not defects to hide.

Signal
Evidence
Action
Meaning
Clear unitary
Committed
INVERSE
Reversible under the agreed model.
Irreversible
Contraction
REFUSE
Do not force a false correction.
Boundary
Insufficient
MEASURE_MORE
Preserve uncertainty; do not guess.
Identity
Within control
NONE
No intervention justified.
Evidence record / 02

The output is a case file, not a victory screen.

Committed audit trace
qpu.qualifycontrol setPASS
criteria.freezepre-runCOMMITTED
analysis.truth_accesscode pathNONE
hardware.workloadsrecorded27
evidence.filesprivate130
missing.expectedmanifest0
boundary.exactfinal testFAIL 11/12
unsafe.inversetested sets0
Measured prototype results
20/20Blind classification

Before hidden truth was opened.

7/7Paired recovery

Corrected combined rotations.

0.77°Signed angle error

All tested signs correct.

12/12Arbitrary recovery

Committed 3D inverses.

2.51°Matrix error

Mean arbitrary-axis error.

9/9U / D / AD

Strong channels classified.

NEGATIVE RESULT
RETAINED

Two sensitivity experiments formally failed exact boundary predictions. In both, finite-shot variation caused conservative refusal rather than an unsafe inverse. The thresholds were not moved afterwards to manufacture a pass.

Audit method / 03

Five phases. One claim. No disappearing failures.

01

Claim intake

State exactly what another person must believe.

02

Protocol

Fix success, failure and controls before evidence.

03

Qualification

Check the target QPU before the main budget.

04

Independent run

Execute, preserve and bind the hardware record.

05

Evidence pack

Human finding plus workload-linked verification.

Founding engagement / 04

One claim. One frozen test. One answer you can show.

Founding reproducibility audit

For a runnable result approaching publication, investment review or a customer demonstration.

  • One workload and claimed result
  • Up to two supported QPUs
  • Pre-run qualification and cost estimate
  • Frozen success and failure criteria
  • Independent execution with controls
  • Positive and negative outcomes retained
  • Human report and evidence receipt
  • Ten-working-day target
Claim boundary / 05

What the instrument does not claim.

Not regulator-approved certification.
Not a guarantee of quantum advantage.
Not novel tomography or control physics.
Not an IBM product, partnership or endorsement.
Not commercially validated by a paying customer yet.
Not proof of universal safety from fixed test sets.

The prototype mainly used IBM Heron r2 processors and small fixed or stratified samples. It does not establish long-term drift robustness, universal thresholds, multi-qubit coverage or performance across all providers.

Input required

What exact quantum claim
must another person believe?

Send the claim, runnable environment, target QPU and deadline. We will tell you whether it fits the founding audit before an engagement begins.

Start with the claim