Skip to content

OSIRIS v5.2.0: Research-to-Evidence Roadmap, Reproducibility Foundation v1.0, and Frontier QPU Validation

Choose a tag to compare

View all tags
@devinphillipdavis devinphillipdavis released this 05 Nov 10:29
· 98 commits to main since this release
v5.2.0
c5ad5fc

OSIRIS v5.2.0: Research-to-Evidence Roadmap, Reproducibility Foundation v1.0 & Frontier QPU Validation

Permanent DOI: 10.5281/zenodo.23976303
Cryptographic Authority Ledger: Sequence Blocks #1–#19 committed to evidence_ledger.db (Block #19: 6e90d79c92387310...)
Clean-Room Verification: python osiris-benchmarks/run_benchmark.py --verify-all

1. What's New in v5.2.0

A. The 180-Day Research-to-Evidence Roadmap (OSIRIS_RESEARCH_TO_EVIDENCE_ROADMAP.md)

A formal six-month research program to convert OSIRIS into a benchmarked, independently reproducible research platform across four core tracks:

  1. Track A (Governance): OSIRIS-GOV-BENCH-1 — Can cryptographic payload receipts and AST gates reject unauthorized mutations without impeding legitimate engineering?
  2. Track D (Scientific Falsification): OSIRIS-SCI-BENCH-1 — Can evidence-governed execution autonomously identify flawed experimental models, mislabeled states, and statistical multiplicity errors?
  3. Track B (Edge AI): OSIRIS-EDGE-BENCH-1 — Memory footprint, tok/s throughput, and perplexity on constrained edge hardware.
  4. Track C (Quantum Control): OSIRIS-QCTRL-BENCH-1 — Hardware-aware predictive dynamical decoupling schedules on physical superconducting processors.

B. Phase 1 Reproducibility Foundation (osiris-benchmarks/)

  • benchmark_contract.json: Pinned machine-readable contracts defining performance claims, baseline comparators, primary metrics, and pass/fail thresholds.
  • environment.lock: Locked runtime hashes across host AMD64 Windows and clean-room Linux containers.
  • verify_results.py: Independent deterministic verifier that recomputes every reported mean, standard error, and confidence interval directly from raw event records (raw_results.jsonl).
  • REPRODUCE.md: Step-by-step instructions for third-party replication on fresh machines.

C. Completed Benchmark Milestones

  1. OSIRIS-GOV-BENCH-1 (1,000 Tasks):
    • Critical Adversarial Escapes: Baseline prompt/regex arm permitted 75.0% ($375 / 500$); OSIRIS Governed Arm achieved 0.0% ($0 / 500$, 95% Wilson CI: [0.00%, 0.76%]).
    • False Rejection Rate: 0.0% ($0 / 500$) with domain-profiled whitelisting.
    • Mean Latency: $0.0214\text{ ms}$ overhead.
  2. OSIRIS-SCI-BENCH-1 (100 Tasks):
    • Defect Recall: Baseline heuristic agent caught $0.0%$ ($0 / 50$); OSIRIS Evidence-Governed Verifier caught 100.0% ($50 / 50$, 95% Wilson CI: [92.9%, 100.0%]).
    • False Discovery Rate: 0.0% ($0 / 50$).

D. Frontier QPU Campaign: Experiments E0, E1, E2, E3 on IBM Heron r2

  • Experiment E0 (dblup49kbkrc73cof20g): Certified genuine two-qubit entanglement ($\mathcal{F}{\text{Bell}} \in [93.55%, 97.30%]$ at $0,\mu\mathrm{s}$; $[80.57%, 87.81%]$ at $4,\mu\mathrm{s}$) using the corrected preparation qc.h(u); qc.h(v); qc.cz(u,v); qc.h(v). All 12 product-state controls satisfied $\mathcal{F}{\text{Bell}} \le 0.50$.
  • Experiment E1 (dblutq9kbkrc73cofcg0): Mapped the multi-duration stagger response landscape across 3 edges and 3 durations, identifying the single-qubit dephasing saturation ratio $\eta_\phi(e, T) = T \cdot \max(\Gamma_\phi(u), \Gamma_\phi(v))$ as the governing physical parameter.
  • Experiment E2 (dblv7sm74qds73dr8ho0): Evaluated frozen OSIRIS Adaptive Scheduler v1.0 on 4 unseen heavy-hex couplers ((15,19), (32,31), (34,33), (35,34)) across $K=4$ blocks ($288\text{ circuits}$, $294{,}912\text{ shots}$). Demonstrated pre-registered replication of positive stagger gain on (15,19)@8us ($+0.56\text{ pp}$, $p = 0.0344$, 4/4 blocks) and (34,33)@4us ($+1.33\text{ pp}$).
  • Experiment E3 Transfer (dblvsnl8ft8s738tsi6g): Pre-registered and dispatched 81-circuit cross-chip transfer evaluation to ibm_kingston testing whether the adaptive policy transfers across physical hardware without retuning.

2. Release Artifacts & SHA-256 Checksums

File Artifact SHA-256 Digest Description
osiris_research_evidence_v5.2.0.zip 42f589cc2e5a8e9c938cb608a8c8ab84103fc0e760f5cfad1d2febcbd9c2bb16 Complete v5.2.0 evidence bundle ZIP
osiris_research_evidence_v5.2.0.tar.gz d3822fbcb88a10b65645cdac8c7364056910c57b094903511c153640d530c320 Complete v5.2.0 evidence bundle TAR.GZ
SHA256SUMS.txt 9d919c57738b2ae49205da391ebbee378bc0924b23d77470e46dbedc417fa68f Checksum manifest of all 80 indexed evidence files
OSIRIS_RESEARCH_TO_EVIDENCE_ROADMAP.md 5ec6cc81b06678d02c4d036ca7fdca04a87d210540731b997462e36c14e444f1 Formal 180-day research program
benchmark_contract.json cbaafab45a4f578b2ad8114e01459f6b0e7d569a11ba8d941968cd6ccc3b5cb2 Machine-readable benchmark contract
OSIRIS_GOV_BENCH_1_REPORT.md 670bc8281f7164d81ea708de29d225d1e4c9eec71064d18ea97db8e8e7266a5f OSIRIS-GOV-BENCH-1 empirical evaluation report
OSIRIS_SCI_BENCH_1_REPORT.md edc1ecfd603f83768e6627e8686284aa07722f9b968f86cea0944cdb67e00db9 OSIRIS-SCI-BENCH-1 empirical evaluation report
zenodo_v520_publication_receipt.json Published (Deposition #23976303) Zenodo publication receipt with permanent DOI
evidence_ledger.db Sequence Blocks #1–#19 Immutable SQLite cryptographic authority chain
Assets 8
Loading