QuartumSE R&D‑Centric Roadmap (Updated, 2025–2026)¶

Last updated: 2025-10-24

Phase snapshot (Oct 2025)¶

✅ Phase 1 scaffolding, provenance pipeline, and CI harness are live.
✅ S‑T01 GHZ baseline + S‑T02 noise-aware runs validated on IBM ibm_torino (smoke test Oct 22, 2025).
✅ IBM Runtime CLI (quartumse runtime-status) operational with webhook notifications.
⚠️ Extended IBM hardware validation (target SSR ≥ 1.1 across repeated runs) scheduled for Nov 2025.
⚠️ Cross-workstream starter experiments (C/O/B/M) need first data drops before Phase 1 closes.
📝 Patent theme shortlist drafting in progress ahead of the Phase 2 gate review.
📋 See phase1_task_checklist.md for the consolidated execution checklist that enumerates every outstanding task before the Phase 1 gate review.

Principle: Front‑load research & hardware iteration. Build on IBM Quantum free‑tier devices until we have an attractive, validated, and patentable measurement stack. Only then open Early Access for design partners.

This roadmap folds in: (i) a sophisticated classical shadows program, (ii) concrete experiments & tests mapped to each phase, and (iii) clear publication/patent gates before external onboarding.

Glossary (metrics & terms)¶

SSR (Shot‑Savings Ratio): shot‑count (baseline) ÷ shot‑count (QuartumSE) at equal error tolerance.
RMSE@$: cost‑for‑accuracy — dollars (or credits/time) to reach a target RMSE on an observable/metric.
CI coverage: frequency a 95% CI contains ground truth (simulation) or gold standard (hardware cross‑checks).
Provenance Manifest: JSON artifact capturing circuits, calibrations, mitigations, backend, seeds, versions.
MEM / M3: measurement error mitigation (confusion matrices); ZNE: zero‑noise extrapolation.
PEC: probabilistic error cancellation; RC: randomized compiling.

Program Structure at a Glance¶

Workstream S (Shadows): Classical Shadows Engine v0→v4 (baseline → noise‑aware → fermionic → adaptive/derandomized → robust Bayesian/bootstrapped).
Workstream C (Chemistry/VQE): Shadow‑VQE for small molecules (H₂, LiH, BeH₂).
Workstream O (Optimization/QAOA): Shot‑frugal QAOA on MAX‑CUT & MIS toy instances.
Workstream B (Benchmarking): RB/XEB/Quantum‑Volume + Shadow‑Benchmarking (fidelity/entropy/purity via shadows).
Workstream M (Metrology): Variational entangled probes (GHZ/W states) for phase‑sensing toy tasks.
Workstream P (Provenance & Reporting): Manifest schema, CI pipelines, PDF/HTML reports, reproducibility notebooks.

Each phase below enumerates Experiments & Tests with IDs that recur across phases for iteration & scaling.

Operational cadence checkpoints¶

Monthly (first business day): Run quartumse runtime-status --json --backend ibm:ibmq_brisbane --instance ibm-q/open/main and log runtime minutes, queue caps, and fallback readiness in OPS_RUNTIME_RUNBOOK.md. Schedule a recurring calendar reminder for the ops lead.
Weekly (Mondays): Trigger the runtime status CLI with Slack webhook enabled to post queue depth/quota snapshots into the project notifications channel. Use the summary to reprioritise hardware jobs if the queue is saturated.

Phase 1 — Foundation & R&D Sprints (Now → Nov 2025)¶

Focus: Ship scaffolding and start real algorithmic experiments immediately (sim + small IBM jobs).

Objectives¶

Solidify repository, CI/CD, SDK skeleton, provenance/reporting.
Implement Shadows v0 (random local Clifford) + v1 (noise‑aware inverse‑channel + MEM).
Stand up baseline C, O, B, M toy pipelines against Aer simulator and at least one IBM free‑tier backend.
Deliver a tractable test suite and benchmarking harness (pytest + notebooks).

Deliverables¶

SDK modules: Estimator, Shadows, Report; Provenance Manifest v1; quickstart notebook.
Mitigation core: MEM (M3) (production) and ZNE scaffolding; PEC/RC hooks planned post-Phase 1.
Shadows v0–v1 reference implementation with CI.
Test harness: datasets, seeds, fixtures; storage: Parquet/DuckDB; PDF/HTML report.
Internal whiteboard spec for patent themes (see Phase 2 gate).

Experiments & Tests (P1)¶

S‑T01 (Shadows‑Core): Random local Clifford shadows on GHZ(3–5), Bell pairs; estimate ⟨Zᵢ⟩, ⟨ZᵢZⱼ⟩, purity. Targets: CI coverage ≥ 0.9; SSR ≥ 1.2 on sim, ≥ 1.1 on IBM.
S‑T02 (Noise‑Aware): Calibrate per‑qubit inverse channel; compare with/without MEM; compute variance reduction.
C‑T01 (H₂@STO‑3G): Hardware‑efficient VQE (depth ≤ 2) + Shadows readout of Hamiltonian terms; energy error ≤ 50 mHa (sim), ≤ 80 mHa (IBM).
O‑T01 (MAX‑CUT‑5): QAOA p∈{1,2} on 5‑node ring; shot‑frugal optimizer; compare cost estimate variance with/without Shadows proxy.
B‑T01 (RB/XEB): 1–3 qubit RB; XEB on depth‑limited random circuits; log into Manifest; compare to IBM backend calibration metadata.
M‑T01 (GHZ‑Phase): Prepare GHZ(3–4), encode small Z‑phase, estimate via optimal readout; CI coverage ≥ 0.8 on sim; explore ZNE for readout bias.

Exit / Success Criteria¶

End‑to‑end run from notebook → manifest → report on Aer + at least one IBM free‑tier backend.
SSR ≥ 1.2× on Shadows‑Core (sim) and ≥ 1.1× (IBM).
CI coverage ≥ 80%, zero critical issues, reproducible seeds & manifests.
Patent themes shortlist (top‑3) + experiment data to support novelty.

Phase 2 — Hardware‑First Iteration & Patent Drafts (Nov → Dec 2025)¶

Focus: Iterate on hardware. Elevate shadows & domain demos; lock initial patent filings; prep first papers.

Objectives¶

Implement Shadows v2 (Fermionic) for 2‑RDM estimation; integrate with VQE readout.
Prototype Shadows v3 (Adaptive/Derandomized): choose measurement ensembles to minimize estimator variance given target observable set.
Harden error mitigation combinations (MEM + RC + ZNE) with ablation studies.
Run structured hardware campaigns (blocked time windows) to control drift.

Deliverables¶

IBM hardware campaign #1 dataset + full manifests + PDF/HTML reports.
Provisional patent draft(s) for: Variance‑Aware Adaptive Classical Shadows (VACS); Shadow‑VQE readout integration; Shadow‑Benchmarking workflow.
Two arXiv preprints: (i) Shadows engine on IBM, (ii) Shadow‑VQE for H₂/LiH small‑basis.
Updated SDK APIs (stabilize experimental flags), plus “replay from manifest” tooling.

Experiments & Tests (P2)¶

S‑T03 (Fermionic‑Shadows): Direct 2‑RDM from shadows; H₂/LiH energies within 40–60 mHa on IBM at ≤ baseline shots; SSR ≥ 1.3× (IBM).
S‑T04 (Adaptive/Derand): Greedy/importance‑sampled basis selection vs plain random; measure variance ↓ ≥ 25% for fixed shots (IBM).
C‑T02 (LiH@Minimal): VQE with Shadow‑readout vs grouped‑Pauli readout; RMSE@$ ↓ by ≥ 30% at matched error bars.
O‑T02 (MAX‑CUT‑6/7): Depth‑aware layout + RC; shot‑allocation per‑iteration; track optimizer steps saved vs fixed‑shot budget.
B‑T02 (Shadow‑Benchmarking): Estimate linear entropy, multi‑qubit purities, and fidelity to GHZ using the same shadows dataset; compare to direct methods; sample‑efficiency ≥ 2×.
M‑T02 (Variational‑Metrology): Variational state+measurement co‑optimization for phase sensing; demonstrate > classical shot‑noise scaling on sim, and robust advantage trend on IBM within CI.

Exit / Success Criteria (Gate to P3)¶

SSR ≥ 1.3× on IBM for at least one domain test (Shadows‑Core or Fermionic‑Shadows).
Draft provisional patent(s) filed; arXiv preprints ready.
CI artifacts: reproducible notebooks, manifests, reports for all P2 tests.

Phase 3 — Internal Validation & Publication/Patent Gate (Jan → Mar 2026)¶

Focus: Consolidate results; conduct controlled comparisons; submit publications; finalize patents. No external users yet.

Objectives¶

Build automated benchmark suite: GHZ, VQE(H₂, LiH, BeH₂), QAOA(MAX‑CUT‑k), Shadow‑Benchmarking panels.
Statistical validation: SSR, RMSE@$, CI coverage, reproducibility (< 2% drift under re‑runs).
Implement Shadows v4 (Robust/Bayesian): bootstrap CI, variance debiasing, heteroscedastic weighting by device cal data.
Prepare journal submissions (PRX Quantum/npjQI/Quantum) and non‑provisional patent filings.

Deliverables¶

Benchmark suite (pytest + CLI) with per‑test Manifest templates and reporting.
Internal whitepaper and slide deck with full ablation matrices.
Code‑frozen R&D branch tagged for archival reproducibility (DOI/Zenodo).

Experiments & Tests (P3)¶

S‑T05 (Robust‑Shadows): Bootstrap CI coverage ≥ 0.9 on sim and ≥ 0.85 on IBM across GHZ and small‑chemistry states.
C‑T03 (BeH₂@Minimal): Shadow‑VQE energy within 80–100 mHa on IBM; RMSE@$ ↓ ≥ 35% vs grouped‑Pauli baseline.
O‑T03 (MAX‑CUT‑7) and O‑T04 (MIS‑6): Evaluate solution quality vs shots; show optimizer steps ↓ ≥ 20% using shot‑frugal and variance‑aware estimates.
B‑T03 (Cross‑Provider Sim): Aer vs IBM reproducibility; manifest “replay” round‑trip equality.
M‑T03 (Sensor‑Tuning): Variational probe robustness to readout noise; CI width ↓ 15–25% after robust shadows weighting.

Exit / Success Criteria (Gate to Early Access)¶

SSR ≥ 1.5× achieved on internal benchmarks; RMSE@$ consistently better than baselines.
At least one paper accepted (or under strong revise‑&‑resubmit) and patents filed.
Provenance & replay validated; CI green across full suite.
Repository made public: Audit Git history for secrets, make repo public to enable external contributions.
CI matrix expanded: Restore full cross-platform testing (12 jobs: 3 OSes × 4 Python versions). See docs/ops/ci_expansion_guide.md.

Only once the above gates are cleared do we begin external onboarding.

Phase 4 — Early Access (Design Partners) & Multi‑Provider Expansion (Apr → Jun 2026)¶

Focus: Limited Early Access after patents/papers. Add AWS Braket connector. Gather external evidence on real workloads.

Objectives¶

Onboard 2–3 design partners (academia/industry) with NDAs referencing filed IP.
Implement AWS Braket connector; cross‑provider parity and consistency tests.
Partner‑coauthored case studies; feedback loop into APIs & docs.

Deliverables¶

Partner playbooks; onboarding notebooks; Slack/Discord channels.
Cross‑provider tests: same circuits on IBM vs AWS; delta analysis reported.
Case study draft(s) + testimonial(s).

Experiments & Tests (P4)¶

B‑T04 (Cross‑Provider Parity): Within 10% agreement on observables post‑mitigation across IBM/AWS for GHZ and VQE(H₂) tasks.
C‑T04 (Partner‑Chemistry): Run partner‑provided small chemistry model; maintain SSR ≥ 1.5×.
O‑T05 (Partner‑Optimization): QAOA on a partner toy instance; capture wall‑clock + cost deltas in RMSE@$.
S‑T06 (Partner‑Shadows): Validate adaptive shadows on partner circuits; document any domain‑specific gains.

Exit / Success Criteria¶

≥3 partners actively running; parity across providers; partner satisfaction survey ≥ 8/10.
External replication of SSR ≥ 1.5×; stable APIs for public beta drafting.

Phase 5 — Public Beta & Pilot Conversion (Jul → Sep 2026)¶

Focus: Stabilize and open up. Convert Early Access into pilots. Prepare commercial posture.

Objectives¶

Public Beta (v1.0) on PyPI + GitHub; full docs; examples; tutorials.
Secure 2–3 pilot customers/LOIs; webinar/demo using published results.
Verify SSR ≥ 2.0× in multi‑provider benchmarks; publish follow‑ups.

Deliverables¶

v1.0 release; docs portal; community channels; issue triage.
Pilot SOW templates; pricing experiments around RMSE@$ value metric.
Public benchmark report with manifests for community reproduction.

Experiments & Tests (P5)¶

B‑T05 (Public Benchmarks): Community‑reproducible GHZ/VQE/QAOA panels with manifests and reference CI.
C‑T05 (Chemistry‑Scale‑Up): Largest feasible molecule instance on accessible hardware; publish shot & cost curves.
S‑T07 (Shadows‑Ablation Public): Public ablation notebook isolating contributions from v0→v4 components.

KPIs¶

≥5 orgs using (3 design partners + ≥2 pilots); ≥2 paying/committed customers.
Verified SSR ≥ 2.0× on multi‑provider suite; community replications reported.

Algorithm & Test Matrix (IDs referenced above)¶

ID	Category	Technique	Circuits / Instances	Backends	Primary Metrics	Evidence Artifacts
S‑T01	Shadows	v0 (random local Clifford)	Bell, GHZ(3–5)	Aer, IBM free‑tier	SSR, CI coverage	Manifest, notebook, PDF
S‑T02	Shadows	v1 (noise‑aware + MEM)	As above	Aer, IBM	Var. reduction, bias	Manifest, ablation table
S‑T03	Shadows	v2 (fermionic)	H₂/LiH 2‑RDM	Aer, IBM	Energy error, SSR	Manifest, data parquet
S‑T04	Shadows	v3 (adaptive/derand)	Target Pauli sets	Aer, IBM	Variance ↓	Manifest, policy snapshot
S‑T05	Shadows	v4 (robust/Bayesian)	GHZ + chemistry	Aer, IBM	CI coverage, width	Manifest, bootstrap logs
C‑T01	Chemistry	VQE + Shadow readout	H₂@STO‑3G	Aer, IBM	Energy error, RMSE@$	Manifest, report
C‑T02	Chemistry	Shadow‑VQE vs grouped	LiH@minimal	Aer, IBM	RMSE@$ ↓	Notebook, plot
C‑T03	Chemistry	Scale‑up	BeH₂@minimal	Aer, IBM	Energy error, SSR	Manifest, report
C‑T04	Chemistry	Partner task	Partner circuit	IBM/AWS	SSR ≥1.5×	Case study
C‑T05	Chemistry	Public benchmark	Largest feasible	IBM/AWS	Shot & cost curves	Public repo
O‑T01	Optimization	QAOA p≤2	MAX‑CUT‑5	Aer, IBM	Cost var., steps	Manifest, runtime logs
O‑T02	Optimization	Shot‑frugal + RC	MAX‑CUT‑6/7	IBM	Steps ↓, RMSE@$	Report
O‑T03	Optimization	MIS‑6	MIS‑6	IBM	Quality vs shots	Manifest
O‑T04	Optimization	MAX‑CUT‑7	MAX‑CUT‑7	IBM	Steps ↓	Logs
O‑T05	Optimization	Partner task	Partner graph	IBM/AWS	RMSE@$	Case study
B‑T01	Benchmark	RB/XEB	1–3q RB; XEB	IBM	Gate error trends	Manifest, plots
B‑T02	Benchmark	Shadow‑Benchmarking	Purity, entropy, GHZ‑Fid	IBM	Sample‑efficiency	Report
B‑T03	Benchmark	Reproducibility	Aer vs IBM	Aer/IBM	Replay fidelity	Replay manifests
B‑T04	Benchmark	Cross‑provider	IBM vs AWS	IBM/AWS	Parity	Report
B‑T05	Benchmark	Public suite	Community panels	All	Replications	Public repo
M‑T01	Metrology	GHZ phase toy	GHZ(3–4)	Aer/IBM	CI cov.	Manifest
M‑T02	Metrology	Variational	GHZ/W	Aer/IBM	Advantage trend	Logs
M‑T03	Metrology	Robust probes	GHZ k‑qubit	Aer/IBM	CI width ↓	Report

Governance, Gating & Quality¶

Gates: P2→P3 requires IBM SSR ≥ 1.3×; P3→P4 requires SSR ≥ 1.5× + patent(s) filed + ≥1 paper accepted/near‑accept; P4→P5 requires cross‑provider parity and partner replication of SSR.
Reproducibility: Every result ships with a Manifest, raw parquet shot data, and a replay notebook.
Risk controls: Time‑boxed hardware campaigns; ablations to isolate mitigation effects; fail‑closed CI (no release if tests or coverage gates fail).

Timeline (target)¶

Phase	Focus	Key Milestone	Target Date
P1	Foundation & R&D sprints	First IBM runs + Shadows v1	Nov 2025
P2	Hardware‑first iteration	IBM campaign #1 + provisional patents + preprints	Dec 2025
P3	Internal validation	SSR ≥1.5× + journal submits + non‑provisionals	Mar 2026
P4	Early Access & expansion	2–3 partners + AWS parity	Jun 2026
P5	Public Beta	v1.0 + pilots	Sep 2026

Notes for Engineering¶

Keep device‑agnostic connectors; IBM first, AWS next.
Prefer layout‑aware, shallow circuits; re‑seed experiments; record cal snapshots.
Implement manifest replay (offline) as a first‑class command; integrate with CI.
Publish small, frequent preprints; convert to journals as results mature.