OSIRIS v5.3.1: Peer-Audited Rescoring & Empirical Noise Floor Bounds on IBM Heron r2

OSIRIS v5.3.1: Referee-Audited Rescoring, Held-Out Nulls, and Inter-Arm Noise Floor Bounds on 156-Qubit IBM Heron r2

Release Tag: v5.3.1
Zenodo Deposition: Continuance of 10.5281/zenodo.23983003
Authority Ledger: evidence_ledger.db (Sequence Blocks #1–#26)
Backends Audited: ibm_marrakesh & ibm_fez (156-qubit Heron r2, $dt = 4.0\text{ ns}$)

---

1. Executive Summary & Erratum / Rescoring Disclosures

OSIRIS v5.3.1 incorporates a rigorous peer-review statistical audit across all prospective physical hardware experiments (E0–E4) executed on IBM Heron r2 processors. Following independent review against analytical shot-noise lower bounds ($\sigma_{\mathcal{F}} \approx 1.01\text{–}1.03\text{ pp}$ per arm per block) and empirical background drift, this release formally updates all headline claims and replaces initial heuristic assertions with verified statistical boundaries:

Rescored Estimands Matrix

Target Claim	Initial Reported	Rescored Value	Forensic Audit Explanation
E3 Pilot Primary (v2 - Fixed80)	$+0.651\text{ pp}$	$+0.038\text{ pp}$	Arm F executed continuous non-preregistered offsets ($40\text{ ns}$ and $120\text{ ns}$). Evaluating strictly the pre-registered binary policy using Arms C ($0\text{ ns}$) and E ($80\text{ ns}$) yields $+0.04\text{ pp}$.
E2 Stagger Benefit (Fixed80 - SYNC)	$+0.446\text{ pp}$ ($p < 0.03$)	$+0.446 \pm \ge 0.206\text{ pp}$	Sample cluster SEs ($0.090\text{–}0.113\text{ pp}$) with $df = 3$ swung below the physical shot-noise floor ($0.206\text{ pp}$). Bonferroni threshold ($\alpha = 0.006$) clears nothing.
Fez Cross-Chip Transfer (Adaptive - Fixed80)	$-0.803\text{ pp}$	$-0.25\text{ pp}$	The reported $-2.00\text{ pp}$ on (100,101) was an arm-to-arm drift between identical $0\text{ ns}$ schedules (accidental A/A test). Transfer is decisively null.
E4 Mechanistic Discrimination (H1 vs H2)	"Decisive Falsification of H1"	Neither Resolved	Both couplers fit a flat response ($\chi^2$ $p = 0.866$ on (14,15) and $p = 0.854$ on (61,62)). Step fraction ($9.5%$) is a ratio of noise-sized values.
Static $ZZ$ Rate $\zeta$ on (14, 15)	$3.170\text{ kHz}$	$0.91\sigma$ from zero	$\Delta f = 6.34 \pm 6.93\text{ kHz}$ ($p = 0.36$). Bounding static coupling to $\le 10\text{ kHz}$ over a $2,\mu\mathrm{s}$ window, without resolving non-zero rate.

---

2. Key Empirical Findings & Methodological Contributions

1. The Accidental A/A Inter-Arm Noise Floor ($1.03\text{ pp}$ RMS):
   Across 7 cell pairs where identical pulse timing was dispatched in separate execution arms, the empirical RMS difference was $1.03\text{ to }1.11\text{ pp}$ ($\chi^2 = 11.09, df = 6, p = 0.086$). This demonstrates that inter-arm calibration drift on commercial hardware equals or exceeds published sub-percentage DD gains.
2. Cross-Chip Scalar Transfer Falsification:
   Calibrations based purely on single-qubit dephasing metrics ($\eta_\phi$) failed to produce an advantage when transferred to ibm_fez ($\Delta \mathcal{F}_{\text{transfer}} = -0.80 \pm 0.60\text{ pp}$), falsifying naive parameter portability.
3. Pooled Fixed 80 ns Stagger Hint:
   Across all 17 evaluated coupler-duration cells on Heron r2, uniform $80\text{ ns}$ staggering shows an inverse-variance pooled advantage of $+0.22 \pm 0.18\text{ pp}$ ($1.19\sigma$) and unweighted mean of $+0.25 \pm 0.16\text{ pp}$ ($1.6\sigma$).
4. Sample Size Power Law for Quantum Error Mitigation:
   Resolving a $0.50\text{ pp}$ fidelity difference at $2\sigma$ significance requires $\ge 33,500\text{ shots per basis per arm}$, establishing the quantitative benchmark for future experimental campaigns.

---

3. Cryptographic Governance & Ledger Receipts

Every circuit execution, analysis script, raw bitstring count, and release manifest is anchored in evidence_ledger.db across 26 sequential cryptographic blocks under operator identity Devin Phillip Davis / Enki.