Skip to content
STIMSMITH

Real-world benchmarks

Technique

Real-world benchmarks are evaluation methodologies that run actual application workloads or curated datasets drawn from production environments to assess performance and correctness, as opposed to synthetic or directed tests. They are widely used in hardware verification (e.g., RISC-V SoC evaluation) and have become central to recent machine-learning research, where new benchmarks such as ComBench and NTP4VC aim to replace decontextualized or synthetic datasets with execution-verified, repository-level data mined from real open-source projects.

First seen 6/25/2026
Last seen 6/25/2026
Evidence 1 chunks
Wiki v1

WIKI

Real-world benchmarks

Definition

Real-world benchmarks are evaluation methodologies that exercise systems — whether hardware, software, or machine-learning models — using workloads, datasets, or test cases drawn from actual production or open-source environments rather than from purely synthetic or hand-crafted stimuli. They are intended to capture the complexity, diversity, and constraints that systems encounter in deployment, providing a more faithful measure of performance and correctness than idealized test suites.

READ FULL ARTICLE →

NEIGHBORHOOD

No graph connections found for this entity yet. It may appear in future ingestion runs.

explore full graph →

RELATIONSHIPS

2 connections
CoreMark ← implements 88% 1e
Performance assessment includes running CoreMark as part of real-world benchmarking.
Dhrystone ← implements 88% 1e
Dhrystone is executed to evaluate performance in the real-world benchmarking phase.

CITATIONS

4 sources
4 citations — click to collapse
[1] Real-world benchmarks in SoC verification involve running applications and performance metrics such as CoreMark and Dhrystone, together with operating-system boot sequences, to assess software–hardware integration. Comprehensive lockstep verification for NaxRiscv SoC integrating ...
[2] ComBench is a repository-level, reproducible real-world benchmark for C/C++ compilation error repair, built by mining failures from GitHub CI histories of large-scale open-source projects and execution-verifying each sample. ComBench: A Repo-level Real-world Benchmark for Compilation Error Repair
[3] NTP4VC is a real-world multi-language benchmark for automated verification-condition proving, constructed from real-world projects such as the Linux kernel and Contiki-OS using Why3 and Frama-C pipelines to produce test cases in Isabelle, Lean, and Rocq. Neural Theorem Proving for Verification Conditions: A Real-World Benchmark
[4] ComBench evaluation reports a 73% syntactic-correctness success rate for GPT-5 but only 41% semantically valid patches, revealing a gap between surface-level and semantic repair quality. ComBench: A Repo-level Real-world Benchmark for Compilation Error Repair