Commented abstract
The paper holds together two ideas to be kept in balance: models are necessary but always partial. Complex systems react through feedback, delays, stocks, and flows; many interventions fail because the decider treats a variable as isolated and ignores the accumulations that feed it.
Structured commentary
Introduction
The paper articulates a tension MARTRO must maintain: models are indispensable and, at the same time, always partial. Sterman recalls the structure of complex systems — feedback, delays, stocks, and flows — to show that numerous interventions fail not through erroneousness of intention but because the decider isolates a variable while ignoring the accumulations that feed it. The failure, in other terms, is one of representation before it is one of will.
In small firms the organisational stocks are often invisible. Backlog, exceptions, rework, documentary debt, distrust between roles, postponed decisions, and personal dependencies accumulate until they manifest as sudden crises. A commercial campaign increases the load on the acquisition process; an indicator modifies local behaviours; a procedure added without removing prior work increases saturation. System dynamics permits reading these effects as circuits, and not as separate events.
The MARTRO reading is markedly anti-linear. Before recommending an intervention, one must ask which feedback circuit will be modified. Signal, decision, action, delay, effect, feedback: even an elementary map may prevent grave errors. In this sense the source sustains the principle of sequence — it is not enough to introduce a solution; one must establish which stocks it will modify and with what delay.
On the operational plane it suffices to trace a circuit on a concrete friction: new leads generate cases, the cases require clarifications, the clarifications saturate the owner, the response is delayed, the customer presses, the pressure heightens urgency and interrupts other work. Such a representation is not yet simulation but renders the mechanism visible — and often it suffices to understand why an intuitive solution would worsen the system.
The boundary consists in not promising complete dynamic models. In the absence of data and explicit boundaries, simulation risks proving more seductive than useful. The entry must present system dynamics as a grammar of structural caution — stocks, flows, delays, feedback — and resist the use of "system" as a totalising word, always naming which accumulation, which delay, or which feedback is being read.
Why it matters for MARTRO
it reduces intervention failures to the partiality of representation — isolated variables, ignored accumulations.
Limits and boundaries of use
complete simulation requires data and boundaries; without them, it is more seductive than useful.
"system" is no totalising word; the accumulation, delay, or feedback under examination must be named.
Practical application for SMEs
trace a feedback circuit on a concrete friction before intervening.