# P3-R1 Hardware Runbook

**Version:** 1.0.0  
**Date:** 15 September 2026  
**Status:** Executable, unactivated single-device pilot; no hardware data have been collected

## 1. Exact question

P3-R1 asks one bounded question:

> Does one of three predeclared eight-cycle coordinates predict a held-out three-qubit outcome beyond a conservative, calibrated Born/context envelope when the complete prediction vector is committed before a delayed random measurement setting?

It does **not** ask whether all quantum outcomes are deterministic, whether eight physical dimensions exist, whether consciousness causes collapse, or whether a local hidden-variable theory can reproduce Bell correlations.

The experiment is a one-processor pilot. Its qubits are not spacelike separated, so it cannot supply a physical no-signaling or Bell test. Its job is to test the chronology, provenance, calibration, and single-device outcome channel of Wallbreaker Protocol P3.

## 2. Frozen state and settings

The source state is

\[
|\psi\rangle=\frac1{\sqrt8}\sum_{x=0}^{7}e^{i\phi_x}|x\rangle,
\]

with the eight phases in `p3_run_config.json`. The two allowed measurement settings are:

- `Z_ALL`: direct computational-basis measurement;
- `X_ALL`: Hadamard on all three qubits, then computational-basis measurement.

Qiskit's displayed bitstring convention is frozen as \(q_2q_1q_0\). Every validation trial is a separate circuit/PUB with exactly one shot. Aggregated multi-shot counts are not admissible as individual trial records. The current provider result records job submission and result retrieval, not an authenticated device-level timestamp for each shot. P3-R1 therefore proves that every setting existed before job submission; it does not pretend to possess missing shot timestamps.

## 3. The three dimensional lifts

For trial \(t\), let \(w=t\bmod8\). Each candidate maps \(w\) through one predeclared permutation: forward, reverse, or binary-reflected Gray order. For setting \(c\), the candidate selects a target outcome \(a_{t,c}\) and its paired opposite \(d_{t,c}=a_{t,c}\oplus4\). Starting from the ideal Born reference \(b_c\), it transfers

\[
\delta_{t,c}=0.12\min\{b_c(a_{t,c}),b_c(d_{t,c})\}
\]

from \(d_{t,c}\) to \(a_{t,c}\). Thus \(q_{t,c}\) is normalized and positive. Over every complete eight-trial cycle, each transfer is reversed once, so

\[
\frac18\sum_{w=0}^{7}q_{w,c}=b_c.
\]

The probe therefore targets conditional information while preserving the cycle-averaged Born marginal in the mathematical reference model. This is a deliberately small falsifiable grammar, not a derived law of nature.

The candidates have prefix lengths 2, 3, and 3 bits. Their Kraft weights sum to \(1/2\); the unused half is retained as cash in the mixture e-process. No candidate, sign, phase, or code length may be changed after validation outcomes exist.

## 4. Composite null and calibration

The validation denominator is not an ideal simulator probability. For setting \(c\) and outcome \(y\), it is a frozen coordinatewise upper bound \(u_c(y)\) for the complete ordinary hardware model. A calibration artifact produces a simultaneous fixed-sample Hoeffding upper bound

\[
u_c(y)=\min\!\left\{1,
\widehat p_c(y)+
\sqrt{\frac{\log(K/\delta)}{2n_c}}+
ho_y
\right\},
\]

where \(K=16\) setting-outcome cells, \(\delta=10^{-3}\), and the predeclared future-drift allowance is \(\rho_y=0.02\). The calibration failure budget covers only its stated sampling assumptions. The 0.02 drift allowance is an empirical modeling assumption; interleaved checks, reversals, resets, and replication are still required before a physical interpretation.

If the actual conditional Born/context probability satisfies \(p_t(y\mid C_t)\le u_t(y)\), then for any precommitted normalized candidate \(q_t\),

\[
\mathbb E\!\left[
\frac{q_t(Y_t)}{u_t(Y_t)}\middle|\mathcal F_{t-1}
\right]
=\sum_y p_t(y\mid C_t)\frac{q_t(y)}{u_t(y)}\le1.
\]

The product is therefore a nonnegative supermartingale. A coordinatewise envelope may sum above one; it must not be renormalized.

## 5. Immutable chronology

The live sequence is:

```mermaid
flowchart TD
    A["Independent calibration"] --> B["Freeze envelope and config hash"]
    B --> C["Commit predictions for every setting"]
    C --> D["Generate delayed settings"]
    D --> E["Submit one-shot PUBs"]
    E --> F["Append raw outcomes"]
    F --> G["Independent replay and score"]
```

The commitment ledger is a SHA-256 chain. Every setting record points to the corresponding prediction hash. The manifest hashes the frozen config, commitment ledger, and setting ledger. The scoring program rejects missing trials, duplicates, hash changes, setting mismatches, outcomes preceding their settings, or a backend different from the frozen backend.

Local SHA-256 commitments establish detectable alteration relative to the supplied manifest; they do not independently prove wall-clock publication. A court-grade run should externally timestamp or publicly anchor the manifest before hardware submission.

## 6. Evidence and gates

For candidate \(j\),

\[
E_{j,T}^{\rm out}=\prod_{t=1}^{T}
\frac{q_{j,t}(Y_t\mid C_t,\Lambda_t)}{u_t(Y_t\mid C_t)}.
\]

The combined process is

\[
E_T^{\rm mix}=\frac12+
2^{-2}E_{1,T}+2^{-3}E_{2,T}+2^{-3}E_{3,T}.
\]

The primary outcome threshold is \(1/(2.87\times10^{-7})\). An outcome crossing is invalidated if the lift also crosses the frozen delayed-setting leakage test at \(\alpha=10^{-3}\), if chronology or hashes fail, or if the calibrated envelope and drift assumptions fail their controls.

Because this pilot has no remote station, `signal_channel` is reported as `NOT_TESTED_SINGLE_DEVICE`; it is never recorded as a pass.

## 7. Activation and commands

Do not paste an API key into the configuration or command history. Save IBM Quantum credentials using the provider's supported account mechanism and expose only the instance CRN through `IBM_QUANTUM_INSTANCE_CRN` if required.

The shipped file is intentionally blocked by placeholders. A real run requires an explicit accessible backend and fresh UUID. The current IBM interface also requires compatible Qiskit and Qiskit Runtime packages.

```bash
# 1. Inspect the unactivated package without hardware.
python p3_hardware_runner.py validate-config --config p3_run_config.json

# 2. Exercise the complete ledger and scorer on seeded synthetic data.
python p3_hardware_runner.py dry-run --config p3_run_config.json --run-dir p3-dry-null --world null --seed 73021

# Optional discovery-scale synthetic positive control (never physical evidence).
python p3_hardware_runner.py dry-run --config p3_run_config.json --run-dir p3-dry-planted --world planted --seed 73022 --trials 20000

# 3. After replacing run_id and backend, collect independent calibration.
python p3_hardware_runner.py ibm-calibrate --config activated_config.json --output calibration.json

# 4. Freeze a conservative envelope into a new activated config.
python p3_hardware_runner.py build-envelope --config activated_config.json --calibration calibration.json --output calibrated_config.json

# 5. Commit predictions, then generate delayed settings.
python p3_hardware_runner.py freeze --config calibrated_config.json --calibration calibration.json --run-dir p3-live-run

# 6. Submit one-shot PUBs and append outcomes.
python p3_hardware_runner.py ibm-submit --run-dir p3-live-run

# If the client stops after submission, retrieve the recorded jobs without resubmitting.
python p3_hardware_runner.py ibm-retrieve --run-dir p3-live-run

# 7. Verify and score without importing production functions.
python verify_p3_hardware_package.py --run-dir p3-live-run
```

## 8. Claim ladder for this run

| Result | Maximum statement |
|---|---|
| Dry-run passes | “The software recognizes its declared synthetic worlds.” |
| Hardware pipeline completes, no crossing | “No declared lift was detected in this pilot.” |
| Outcome crossing only | “The declared single-device composite null was rejected.” |
| Crossing plus chronology, leakage, drift, memory, reset, and reversal controls | “A held-out residual survived the tested ordinary single-device explanations.” |
| Independent devices/laboratories | “The frozen residual externally replicated.” |
| Spacelike Bell implementation | Only then may W6–W8 be adjudicated. |

A 256-trial pilot is not powered for a small new-physics effect. A discovery-sized maximum trial count must be computed from calibration-derived information-rate lower bounds and frozen in a later run; it cannot be inferred by peeking at these pilot outcomes.

## 9. Current verdict

The experiment definition, commitment order, candidate grammar, conservative score, and independent replay can be executed now. The supplied configuration is not activated, Qiskit is not bundled into the audit, and no IBM job has been submitted. Therefore the package creates a real route to evidence but contains no physical evidence yet.

## Official implementation references

- IBM Quantum, [Run your first circuit on hardware](https://quantum.cloud.ibm.com/docs/en/guides/hello-world).
- IBM Quantum, [Set up to use IBM Cloud](https://quantum.cloud.ibm.com/docs/en/guides/cloud-setup).
- IBM Quantum, [Manage your instances](https://quantum.cloud.ibm.com/docs/en/guides/instances).