RetailOpt-190: A Retail Supply Chain Benchmark for Text-to-Optimization

RetailOpt-190 is a solver-validated benchmark for evaluating semantic reliability in text-to-optimization. It tests whether LLM-based agents can reconstruct the intended optimization structure—not just produce runnable code.

Dataset Summary

RetailOpt-190 contains 190 retail supply chain optimization instances designed to test compositional consistency in LLM-generated optimization code. Each instance includes a natural-language problem description, structured JSON data, and ground truth solutions from a validated MILP solver.

The benchmark spans 8 scenario families and 38 archetypes covering core retail planning mechanisms:

Family	Name	Archetypes	Key Mechanisms
F1	Core Operations	4	Multi-period inventory, seasonal demand, perishability
F2	Assortment & Substitution	6	Product substitution, promotions, ultra-short shelf life
F3	Resource Constraints	4	Storage bottleneck, supply bottleneck, volumetric limits
F4	Demand Dynamics	6	Demand surge, supply risk, peak failure
F5	Feasibility Stress	4	Impossible demand, storage overflow, strict service traps
F6	Discrete Logistics	4	Lead time, MOQ, pack size, fixed order cost
F7	Network & Multi-Echelon	6	Transshipment, hub-spoke, multi-sourcing
F8	Omni-channel	4	Reverse logistics, labor constraints, sustainability

Languages

English

Two Prompt Formats

RetailOpt-190 provides two prompt formats for different evaluation scenarios:

Format	Field	Data Location	Use Case
Schema-based	prompt_schema	External (runtime)	Large datasets, tests data access patterns
Data-embedded	prompt_full	In prompt	Direct comparison with other benchmarks

Why Two Formats?

Most existing benchmarks (NL4Opt, MAMO, IndustryOR) embed data directly in prompts. RetailOpt-190 supports both approaches to enable:

1. Fair comparison: Use prompt_full when comparing with other benchmarks in unified evaluation frameworks
2. Scalability: Use prompt_schema for production scenarios with large datasets

Both formats provide the same semantic information—only the data delivery method differs.

Dataset Structure

Data Fields

Field	Type	Description
scenario_id	string	Unique scenario identifier (e.g., retail_f1_base_v0)
prompt_schema	string	Schema-based prompt (data loaded at runtime via data variable)
prompt_full	string	Data-embedded prompt (full JSON data in prompt)
data	string	JSON-formatted instance data (parse with json.loads())
reference_status	string	Ground truth solver status (OPTIMAL, INFEASIBLE, etc.)
reference_objective	float	Ground truth objective value (null if infeasible)

Data Splits

Split	Examples
test	190

Usage

Loading the Dataset

from datasets import load_dataset
import json

dataset = load_dataset("Jacoblian/RetailOpt-190", split="test")
sample = dataset[0]

print(sample['scenario_id'])        # e.g., "retail_f1_base_v0"
print(sample['prompt_schema'][:200])  # Schema-based prompt
print(sample['prompt_full'][:200])    # Data-embedded prompt

Option A: Schema-based Evaluation

Use prompt_schema when you need external data loading (matches production scenarios):

from datasets import load_dataset
import json

dataset = load_dataset("Jacoblian/RetailOpt-190", split="test")

for sample in dataset:
    prompt = sample['prompt_schema']
    data = json.loads(sample['data'])

    generated_code = your_llm(prompt)
    exec(generated_code, {'data': data})  # Data pre-loaded

    print(f"Reference: {sample['reference_status']}, {sample['reference_objective']}")

Option B: Data-embedded Evaluation

Use prompt_full for direct text-to-solution evaluation (compatible with other benchmarks):

from datasets import load_dataset

dataset = load_dataset("Jacoblian/RetailOpt-190", split="test")

for sample in dataset:
    prompt = sample['prompt_full']  # Data is already in prompt

    generated_code = your_llm(prompt)
    exec(generated_code)  # Code parses JSON from prompt itself

    print(f"Reference: {sample['reference_status']}, {sample['reference_objective']}")

Evaluation Metrics

• Execution Rate: Percentage of instances that run without error
• Accuracy: Percentage matching ground truth (status + objective within tolerance)
• Silent Failure Rate: Executable code with incorrect answer

Accuracy Tolerances

Family	Problem Type	Tolerance
F1-F5, F7-F8	LP / easy MIP	0.01%
F6	Hard MIP (MOQ, pack-size)	10%

Dataset Creation

Source Data

All instances are synthetically generated from 38 archetype specifications. Each archetype is instantiated with 5 numerical variants (v0-v4) via controlled parameter perturbations.

Annotations

Ground truth solutions are computed using a validated MILP solver (Gurobi) with the following settings:

• TimeLimit: 60 seconds
• MIPGap: 1%
• Threads: 1

Additional Information

Citation

@article{lian2026reloop,
  author    = {Junbo Jacob Lian and Yujun Sun and Huiling Chen and Chaoyu Zhang and Chung-Piaw Teo},
  title     = {ReLoop: Detecting Silent Failures in LLM-Generated Optimization Code via Behavioral Verification},
  journal   = {arXiv preprint},
  year      = {2026}
}

License

• Code: MIT
• Data: CC BY 4.0

Related Resources

• ReLoop Framework: https://github.com/junbolian/ReLoop - Complete implementation of the ReLoop verification pipeline