Bioprocess
Yes for screening, and yes for the shape. The magnitude may be published provided it is stated that the model is anchored to a single 2014 laboratory column and has never been compared to a field installation. Not yet a basis for design.
instrument · digital twin
Methodology
What the model computes, what it refuses to compute, and what four reviewers changed about it.
How it is built
Six modules: an uncertainty engine, a parameter registry, a biofilm reactor, a mine-water model, an abatement controller, and the coupling that runs them as one state through time. Gas, water and abatement share a single state variable — the water level — and the sign of its effect differs for each. Rising water seals workings and cuts emission; it also displaces gas ahead of it and raises emission; it eventually surfaces and has to be treated. Modelling those separately is why the numbers never reconcile.
twin-uncertainty.jsPRNG, distributions, Latin hypercube, Iman–Conover, quantiles, sensitivity, AACE/TRL grading, Monte Carlo drivertwin-priors.jsthe parameter registry, sources, correlations, self-audittwin-bioreactor.jsHenry's law, biofilm diffusion–reaction, gas film, non-ideality, field envelopetwin-water.jsrebound, first flush, Fe(II) kinetics, treatment sizing, thermal recoverytwin-abatement.jssource term, safety gate, routes, climate accounting, controllertwin-scenario.jsthe coupling and the Monte Carlo entry point
What it refuses to do
The estimate class is ceilinged by the worst evidence tier in the dependency chain, and callers cannot opt out. Where no AACE class fits the spread, none is named. Where a wetland cannot be sized without an acidity load, it returns a refusal rather than a zero. Where a run's tail is not sampled well enough to quote, it says so. A model that will not say what it does not know is the only kind worth putting in front of a regulator.
- Parameters sampled
- 87
- of which verified against a cited source
- 18
- Tier 1
- 15
- Tier 2
- 29
- Tier 3
- 33
- Tier 4
- 10
Known limits
Model-form uncertainty is not propagated and cannot be within this architecture: the intervals reflect parameter spread only. The managed water-level regime that governs the Ruhr is not represented, only warned about. Chloride is absent. The reactor is anchored to a single 2014 laboratory column and has never been compared to a field installation.
Four independent expert reviews
The model was reviewed adversarially across four domains, with a second opinion in each. Verdicts as the reviewers wrote them.
Hydrogeology
Yes for screening. No for a Ruhr or Upper Silesia release. Reviewed as a screening model for a general north-west European coalfield; not suitable for the Ruhr or Upper Silesia until a managed-level regime and chloride are added.
Uncertainty quantification
The mathematics was already sound. What was wrong was the layer that turns a distribution into a claim. I regard the resulting figures as fit for screening and design-basis framing, and not for design — which is what the document already claims for them.
Utility MRV and verification
Sign the numbers for verification: yes, unqualified. Fund a measurement campaign: yes. The accounting is better than most of what gets third-party verified in this sector, and the reason is that it reports the things that count against it.
The single most useful thing the review produced: a model can be numerically flawless and still tell an asset owner something false. Every defect found across four streams sat not in the mathematics but in the layer between a correct computation and what a reader is told — the part that looks like presentation, and is almost never reviewed.
The full specification, including the complete peer-review record and every retraction, is maintained alongside the source.