Reproducibility benchmark · Case study
Cornstover TEA uncertainty reproducibility
Seeded economic uncertainty on the BioSTEAM cornstover reference model, qualified for reproducibility and restoration of deterministic process and TEA state.
Claims on this page apply only to the stated system boundary, methods, source revision, and evidence status.
- Source revision
- BioSTEAM source cca9243522dcd3062da9ca76893a0fba8f84c126 · qualification run 20260915T192856Z
- Last verified
- 2026-09-15
Decision question
Can economic uncertainty be evaluated repeatedly without leaving sampled price or IRR state behind or perturbing the solved cornstover process model?
System boundary
BioSTEAM cornstover reference model with economic-only uncertainty applied to feedstock price, electricity price, and IRR. The benchmark targets adapter reproducibility and state isolation, not real-world validation of the underlying biorefinery economics.
Functional or reference basis
Two independent Monte Carlo runs with N=200 and seed=42, followed by deterministic state and named-model rebuild checks.
Methodology
The workflow snapshots the economic state, evaluates economic samples without rerunning unit operations for every draw, refreshes operating-cost accounting from the already-solved process state, and restores all sampled economic attributes in a finally block. A separate causal diagnostic established that repeated full process simulation could itself move recycle-dependent state, so the uncertainty path was intentionally isolated from unnecessary process reruns.
Assumptions
The BioSTEAM cornstover reference model and its existing solved process state are taken as the computational basis. The reported uncertainty distribution reflects the frozen parameterization and is not an empirical confidence interval for a commercial plant.
Results
The final qualification restored economic state within 1e-12 and deterministic TEA within 1e-9. A post-uncertainty named rebuild reproduced the original process summary exactly. Two independent N=200, seed=42 runs returned identical API-level distributions and Spearman payloads. The runs produced mean MSP 0.6699 USD/kg and P5/P50/P95 values of 0.5919/0.6707/0.7570 USD/kg, with Spearman rho +0.940 for feedstock price and -0.322 for electricity price.
Interpretation
The benchmark shows that the qualified economic-only uncertainty path is repeatable and does not contaminate later deterministic results. It also demonstrates why uncertainty workflows need explicit state-management tests in addition to statistical outputs.
Limitations
The qualification does not establish real-world TEA accuracy, validate the underlying cost correlations, prove bit identity of unexposed raw sample arrays, or guarantee idempotence of arbitrary repeated system.simulate() calls.
Reproduction
Load the qualified cornstover model, capture the deterministic process and TEA state, run the frozen N=200 seed=42 uncertainty workflow twice, compare API-level distributions and sensitivity payloads, then rebuild the named model and verify restoration tolerances.
Review the underlying methods.
Use the documentation to inspect the tools and workflows behind this study.