Quantum Computing Project — Baseline Snapshot

Quantum Computing Project — Baseline Snapshot

Baseline date: 2026-06-30
Location: learning/ (moved here on 2026-06-30 from ~/Documents/Learning/)
Purpose: Frozen inventory of what currently exists, with measured (not just doc-claimed) state, so future review and external research have a fixed reference point.

How to read this: anything under Measured (2026-06-30) was run/verified today. Anything under Documented is a prior claim from the repo that has not been re-verified at baseline time.

0. Snapshot at a glance

1. Environment baseline

Recreated and verified today (quantum_env, both quantum_computing/ and quantum_rag_demo/):

Not recreated (intentional): qi312_env — the Python 3.12 environment for the Quantum Inspire SDK (qiskit-quantuminspire is incompatible with Python 3.13). Live QI-hardware execution requires rebuilding this separately on a 3.12 interpreter.

Third-party import surface (quantum_rag_demo): numpy, qiskit, qiskit-aer, qiskit-algorithms, scipy, scikit-learn, matplotlib, requests. Optional/credential-gated: qiskit-ibm-runtime, qiskit-quantuminspire, anthropic, openai, faiss, sentence-transformers, torch (PyTorch), spaCy.

2. Inventory by folder

2a. quantum_computing/ — learning sandbox (3 py, 1,448 LOC)

Entry-point "learn the basics → apply to Maestro" material.

2b. quantum_rag_demo/ — main system (71 py, ~46,118 LOC, 39 docs)

Status VALIDATED v1.5.0 (per repo). All core algorithms are real implementations (no NotImplementedError/TODO in core paths).

Core retrieval engines

Graph & symbolic reasoning

RAG infrastructure

Hardware runners (credential-gated, simulator fallback)

Benchmarking & evaluation
benchmark_comparison.py, progressive_benchmark.py (10→5k), enhanced_progressive_benchmark.py (→10k, sklearn embeddings), ablation_study.py, virtuous_cycle_demo.py.

Subsystem: research_agent/ (49 files) — autonomous arXiv→validate→JIRA pipeline.
3 agents (Researcher / Peer-Reviewer / Auditor) via daemon.py; pulls arXiv across 35 topics, validates with Claude, files JIRA tickets via Atlassian MCP. Components: memory manager, multi-channel graph fusion, deterministic extractor, edge denoiser, collaborative recommender, metacognitive orchestrator, token optimizer.
Runtime state: daemon NOT running (stale daemon.pid 8868; last cycle ~Jan 21). daemon.log 4.1 MB, research_state.json 1.0 MB. 4 papers validated, 8 tickets scoped (see SESSION_RESTART.md).

Subsystem: ai_extensions/ (8 files) — AI→QC direction (ML improves quantum runs).
error_predictor.py (LSTM, numpy-EMA fallback), drift_compensation.py (per-qubit Kalman), adaptive_selector.py (rule+neural method picker), data_collection.py, qi_integration.py.
State: code complete but never run end-to-end — no trained models or collected data on disk (.pt/.json absent). One real stub: dynamical-decoupling insertion.

2c. ai_quantum_symbiosis/ — research writeups (0 py, 9 docs)

Literature/analysis only: quantum state of knowledge, AI error correction, AI quantum control, AI circuit optimization, AI materials discovery, symbiosis analysis, integration notes, references.md. Conceptual basis for ai_extensions/.

3. Capability matrix (retrieval methods)

4. Empirical baseline — measured 2026-06-30

Test suite (fresh quantum_env)

• 2 failures (real): research_agent/test_token_optimizer.py — test_paper_abstract_optimization and test_consistent_reduction_across_texts (token-reduction consistency assertion, e.g. reductions [32.4, 8.4, 25.9] fail a >=10% uniformity check).
• 4 skipped: FAISS/credential-gated tests (FAISS segfault on Apple Silicon is a known guarded skip).
• Test-runner caveat (not a defect): research_agent/ tests use bare imports (import hypergraph) and only collect when run from inside research_agent/. Running the whole tree from the repo root reports them as 7 collection ERRORs — these are cwd/path artifacts, not broken code.
• Many deprecation warnings from qiskit 2.x (e.g. GroverOperator, MCXGate.get_num_ancilla_qubits, mcx(mode=...)), since this code targeted an earlier qiskit. Non-fatal today; will break on qiskit 3.0.

Documented results (from result JSONs — NOT re-measured at baseline)

• qi_hardware_results.json: Grover on aer_simulator, 3 qubits, target 101, measured prob 0.953 vs theoretical 0.945, success.
• ablation_results.json (corpus 12, 10 queries): Full hybrid recall@1 0.55, recall@3 1.0, MRR 0.767; "Without Quantum" recall@1 0.50.
• progressive_benchmark_results.json: explicitly tags speedup_metric = operation_count, wall_clock_note = "Simulator time only, not representative of quantum advantage". Reports recall@5, MRR, p50/p95/p99.
• Prior validation claims (STATUS.json / PROJECT_TRACKER): local simulator 95.31% Grover accuracy across 551 tests; Quantum Inspire hardware passed (Bell/GHZ/Grover/QFT; Grover 90.6% fidelity on Tuna-9, 1024 shots). Hardware claims not reproducible without qi312_env + credentials.

5. Honest framing of "speedup" claims

The repo's own tracker states: all "speedup" numbers refer to algorithmic scaling / operation counts (e.g. Grover ~O(√N) iterations), NOT wall-clock runtime. On a classical simulator the quantum paths are slower in real time. Any external review or paper must keep this distinction explicit.

6. Known issues / gaps / cleanup backlog (baseline)

1. ai_extensions/ unproven — complete code, zero runtime evidence (no trained model, no collected data). The AI→QC "virtuous cycle" is unvalidated end-to-end.
2. research_agent daemon stale — not running; large stale logs (daemon.log 4.1 MB) and state files. 2 failing token-optimizer tests.
3. qiskit 2.x deprecations — code will break on qiskit 3.0; needs migration.
4. Doc path drift — several docs (e.g. PROJECT_TRACKER "How to Run", STATUS) still reference the old ~/Documents/Learning/... paths.
5. requirements.txt incomplete — missing qiskit-algorithms, pytest, pytest-asyncio, sklearn, matplotlib (needed but not pinned).
6. multimodal_schema.py is schema-only — no ingestion/execution implemented yet.
7. Secrets present — quantum_rag_demo/.env holds tokens; ensure it stays out of any external sharing.

7. External-research targets (for the next phase)

STATUS — DONE (2026-06-30): This external-research phase has been executed. A

~1,213-paper / ~7,195-row literature review now lives at

quantum_rag_demo/docs/research/expansion_2026-06-30/ (reports + corpus + provenance).

It stratifies the 7 methods into Tier 1/2/3 by evidence; see that folder's README.md

and research_expansion_report.md. Derived learning path: QUANTUM_LEARNING_PATHWAY.md.

Caveat: the synthesis is agent-generated — spot-check citations before publishing.

Claims to verify / benchmark against literature:

• Grover O(√N) advantage for retrieval under realistic NISQ noise & data-loading (QRAM) costs — is there a genuine end-to-end win, or only asymptotic op-count?
• Holographic Associative Memory "O(1)" retrieval — capacity/accuracy tradeoffs vs classical ANN (FAISS/HNSW).
• Complex-valued / phase-interference embeddings for semantics — current SOTA.
• AI-for-quantum-control: ML error prediction & drift compensation — does the ai_extensions/ approach match published results?

Referenced papers already in the repo (arXiv, Jan 2026) — re-check & expand:

Suggested external scope: validate the 7 retrieval-method complexity claims against current quantum-RAG / quantum-NLP literature; survey 2026 work on quantum search for vector DBs; sanity-check the AI→QC error-correction approach.

8. Reproduce this baseline

cd learning/quantum_rag_demo
# core quantum tests (from repo root)
./quantum_env/bin/python -m pytest tests/ -q
# research_agent tests (MUST run from its own dir)
cd research_agent && ../quantum_env/bin/python -m pytest . -q

Note: qi312_env and live IBM/QI hardware runs are excluded from this baseline.