Catalyst-Q Benchmark Evidence

Generated: May 2026

This repository is evidence infrastructure for Catalyst-Q. It contains benchmark manifests, validators, raw result formats, and generated scorecards for named public workloads.

The benchmark claim boundary is intentionally narrow:

• Catalyst-Q results are publishable only for named instances with committed inputs, raw records, validators, and reproducible artifacts.
• High-qubit circuit evidence is query-native targeted exactness for supported structured and bounded-contraction families. It is not a claim that arbitrary dense output distributions can be materialized at those widths.
• Solver comparisons require external baseline runs before broad SOTA language is allowed.

Generated Evidence Package

scripts/build_full_evidence_package.py generates the current scorecard from raw records and high-qubit exactness artifacts.

Checked-in files under results/ are generated snapshots. Regenerate the package before publishing fresh claims or assuming it reflects uncommitted source changes.

Publishable evidence includes:

• SDK route coverage for SAT, TSP, knapsack/MKP, portfolio, QUBO, Max-Cut, and DAG optimization request surfaces.
• Exact reference objectives for bundled smoke instances.
• Live Catalyst-Q QUBO and Max-Cut API checks when --execute-api is enabled.
• Compact exact high-qubit targeted answers with zero dense state materialization across the generated high-qubit campaign.

The generated package lives in results/full_evidence_package.md.

Benchmark Surfaces

Surface	What is measured	Claim status
Public solver routes	Request validity, objective extraction, exact smoke references	Publishable per named record
QUBO and Max-Cut smoke runs	Live API objective vs exact reference and simple heuristic baseline	Publishable when raw live records are present
High-qubit exactness	Targeted probabilities, amplitudes, marginals, and observables	Publishable for named supported query families
External SOTA campaigns	SAT, MaxSAT, TSPLIB, OR-Library, Biq Mac, QPLIB, MIPLIB, QAPLIB	Not yet complete

Reproduce Locally

python3 -m pytest tests -q
PYTHONPATH=src python3 scripts/run_high_qubit_exactness.py
PYTHONPATH=src python3 scripts/build_full_evidence_package.py

Use live API execution only when intentionally collecting remote evidence:

PYTHONPATH=src python3 scripts/build_full_evidence_package.py --execute-api

Claim Policy

See docs/claims_policy.md. In short: publish measured, named outcomes; do not publish generalized complexity, universal quantum simulation, or category-leadership claims without the corresponding raw campaigns and external baselines.