The motivation behind BenchEWS
What happens when our world changes faster than people and societies can adapt?
This page explains why BenchEWS exists, what the research program can investigate today—and what remains an unproven vision.
01 · Motivation
Systems can overlook or misread warning signals, or fail to respond effectively even after problems have been recognized. Under what conditions do they lose their capacity for self-correction?
Read more ↓02 · Vision
BenchEWS investigates whether mathematical models and reproducible testing procedures can help detect critical changes earlier and more reliably under defined conditions. No universal detector for societal systems exists to date.
Read more ↓03 · Falsifiability
Our hypotheses must hold up against appropriate baselines and independent data. If they show no robust added value, they are revised or rejected.
Read more ↓Scope: motivation and orientation. This page reports no new research results. Every statement is labeled—for example as an established finding, an analogy, a hypothesis, an open question, a research goal, a vision, or a value decision.
01 / Adaptation
In biology, evolutionary theory explains how populations adapt to changing environments through variation, inheritance, and selection. This is very well supported empirically.
In cybernetics, Ashby's law of requisite variety applies to formally described regulatory systems: a regulator can counter the effects of disturbances only to the extent that it has a variety of responses available. It is a statement about formally defined systems.
References: Ashby (1956) ↓
Applying these ideas to companies, institutions, or societies is an analogy, not an empirical finding. Institutions are not inherited and selected like organisms, and their "variety of responses" cannot simply be determined as in a formal control loop. The analogy helps us ask questions; empirical research has to answer them.
In this sense, we treat self-correction not as something a system has once and for all, but as an ongoing process of sensing, feeding back, evaluating, and acting—one that can succeed or fail.
Full background: the program's intention →02 / Pace
Research on technology governance discusses the so-called pacing problem: in some domains, technologies develop faster than legal, ethical, and institutional frameworks catch up. At the same time, studies of technology diffusion show that adoption speeds vary widely across technologies and countries.
References: Marchant, Allenby & Herkert (2011); Comin & Hobijn (2010) ↓
On this page, the "adaptation gap" is a conceptual metaphor, not a measured societal quantity. There is as yet no accepted measure of a society's "pace of change" or "pace of adaptation." Institutions learn, too, and can become faster. BenchEWS assumes neither universal acceleration, nor necessarily exponential growth, nor inevitable overload.
The three figures stand side by side with equal status. None is presented as more likely; whether any of them applies to a real system remains open.
Hypothetical scenario A
The pace of change accelerates more than the pace of adaptation; the distance grows.
Hypothetical scenario B
Adaptation lags at first, then learns and closes the distance again; a gap can be temporary.
Hypothetical scenario C
Periods of rapid change alternate with periods of successful adaptation; the gap opens and closes repeatedly.
03 / Limits
People decide with limited information, time, and attention. Rules of thumb (heuristics) can work very well in suitable environments; under other conditions, they lead to systematic errors of judgment.
References: Simon (1955); Tversky & Kahneman (1974); Gigerenzer & Gaissmaier (2011) ↓
In laboratory experiments with delays and feedback, participants often control simulated systems poorly, for example with costly oscillations. This is a laboratory finding. How far it transfers to real organizations and societies is an open question.
References: Brehmer (1992); Sterman (1989) ↓
In system dynamics models, delayed feedback can lead to overreaction, oscillation, or late correction. Whether and how strongly this applies to a specific real institution has to be tested empirically in each case.
References: Sterman (2000) ↓
It is well documented that people often seek and evaluate information in ways that support existing beliefs (confirmation bias), and that wishes and group identity shape judgment (motivated reasoning). Whether such biases are equally or differently strong across political camps is contested: one meta-analysis found similar levels, critics argue the selected studies cannot answer that question, and the authors disagree. That question is not decisive for the motivation of this page. Hypothesis: if such biases reinforce each other in groups and institutions, they may screen out contradicting feedback and delay self-correction. This has not been demonstrated.
References: Nickerson (1998); Kunda (1990); Ditto et al. (2019); Baron & Jost (2019); Clark et al. (2019) ↓
Laboratory findings and model results are not direct evidence about societal dynamics. And this is not about any individual's intelligence; it is about structural limits that apply to everyone—including science and this research program.
04 / The mathematical compass
Mathematical models can make assumptions explicit, make chains of cause and effect computable, and make claims falsifiable. In some systems, critical transitions are preceded by measurable changes, such as slower recovery after disturbances ("critical slowing down"). At the same time, the literature shows that such early-warning signals can fail to appear, raise false alarms, or be detectable only under favorable conditions.
References: Scheffer et al. (2009); Ditlevsen & Johnsen (2010); Boettiger & Hastings (2012) ↓
Published work (including a peer-reviewed result on Φ(Σ) for specified model classes), the BenchEWS Studio reference software, and a validation framework in which early-warning indicators are compared under the same data and model conditions.
Testable methods for early warning and recoverability in clearly bounded systems. Work is in progress; no superior indicator has been demonstrated so far.
An aid to orientation that might one day support decisions in complex systems. Whether it is achievable remains open.
No validated universal detector of societal self-correction currently exists. Wherever this website speaks of a "mathematical compass," it means a research goal or a vision—not an existing tool.
Benchmarking tests how well a quantity signals known transitions in models or selected data sets. It does not directly measure a society's capacity for self-correction—accepted definitions and measurement procedures for that are still lacking.
A model is free of contradictions, its derivations are correct, and its results follow from its assumptions.
The model describes a real system well enough—tested against data, against alternatives, and with stated uncertainty.
The first is necessary but not sufficient for the second. BenchEWS keeps both levels visibly separate.
The figure organizes questions; it is not an empirically confirmed causal model. No step is inevitable. Each step has other outcomes—possible, but not necessarily equally probable.
Technology, climate, the economy, or communication change a system's conditions.
Also possible: The change stays within the range of existing routines.
Established in the literatureExisting routines fit less well; adaptation becomes necessary.
Also possible: Adaptation succeeds; the system learns and stabilizes.
Established in the literatureIf change persistently outpaces adaptation, reserves may be depleted.
Also possible: Relief, reserves, or learning restore the capacity to act.
Hypothesis · context-dependentFeedback no longer reaches the places able to act; alternatives disappear.
Also possible: The loss remains partly reversible, or the system reorganizes.
Hypothesis · context-dependentWhether this loss can be measured early and reliably is what BenchEWS investigates.
Also possible: Signals fail to appear or raise false alarms; detection does not guarantee action.
Research goal · not solvedWhat follows is decided by people and institutions on the basis of their own values.
Also possible: Every action itself changes the environment—and the cycle begins again.
Value decision · not mathematicsFeedback: every stage can act back on earlier ones. Successful adaptation, recovery, and reorganization are possible outcomes at every stage—possible, but not necessarily equally probable.
05 / Historical illustration
Germany in 1933 shows how quickly a democracy's corrective mechanisms can be suspended. Within a few months—often through seemingly legal means—central feedback channels were removed:
The Reichstag Fire Decree suspends basic rights such as freedom of expression, of the press, of assembly, and of association.
The Enabling Act allows the government to enact laws without parliament. The Communist Party's (KPD) seats had already been annulled; the Social Democrats (SPD) were the only parliamentary group to vote against it.
The Law Against the Formation of Parties makes the Nazi Party (NSDAP) the only legal party.
This removed a free press, parliamentary oversight, and legal opposition—channels through which mistakes can be recognized, named, and corrected. The Nazi regime led Germany into the Second World War and committed the genocide of the Jews of Europe. The example illustrates what the process model means by "loss of self-correction."
Sources: bpb: Reichstagsbrand – auf dem Weg in die Diktatur ↗ · Deutscher Bundestag: 23. März 1933 – Ermächtigungsgesetz ↗ · DHM LeMO: Ermächtigungsgesetz ↗ · DHM LeMO: Einparteienstaat ↗
06 / BenchEWS today
BenchEWS is the overarching research program. It asks not only whether a system becomes unstable, but whether it loses its capacity for self-correction—and whether it can still recover (recoverability). Today, this concretely means comparing early-warning indicators under the same data, model, and observation conditions, disclosing assumptions, and documenting limitations.
07 / Method
What is observed are time series and clearly defined events or states in models and selected data sets. "Capacity for self-correction" and the "adaptation gap," by contrast, are theoretical constructs: they are not measured directly and can at most be inferred through stated indicators and assumptions. The hypothesis tested in each case is that an indicator provides a robust information gain over established comparison methods.
A specific hypothesis counts as testable only once it has been fixed, before the analysis, how it could fail. For each study, the following are therefore specified in advance—preregistered where possible:
Which system, which system boundary, and which clearly observable event or state counts as a hit?
Which indicators are tested—against which established methods and null models?
How early must a signal appear, and over what period is it evaluated?
How many false alarms and missed events are acceptable, and are the data sufficient to detect an effect of the expected size at all?
Testing on data not used during development—with disclosed code and, where possible, by third parties.
Which pre-specified results lead to rejecting the hypothesis or ending the project?
There are no universal thresholds. Specific numbers depend on the domain, the available data, and the question, and are justified study by study. The list describes methodological requirements, not results: which tests have actually been carried out and what they found is documented in the evidence matrix and the publications.
CRTI development was discontinued because its added value could not be sufficiently demonstrated. This is a documented example of correcting a research decision—not a success of a detector. It shows that this rejection procedure was applied in one specific case. It is not evidence that the research program as a whole, or any other indicator, is validated or superior.
Falsifiability criteria →08 / Limitations & open questions
Scientific models cannot determine democratic or ethical goals. Under stated assumptions, they can show which consequences, delays, or trade-offs are possible. Which goals a society pursues is decided by people through democratic processes. This page is not a campaign and does not represent any party.
More: policy & society →09 / Invitation
BenchEWS is meant as an open research program. Criticism, independent replication, and well-founded disagreement are explicitly welcome—especially where the results are inconvenient.
Scientific collaboration: BenchEWS@vonmallinckrodt.de · Dipl.-Ing. Bernd von Mallinckrodt · ORCID 0009-0005-5279-6607