Life-cycle assessment software

An LCA tool takes a bill of materials and process flows, matches them against an inventory database of unit processes, and returns impact-category numbers — global-warming potential, water use, eutrophication. The computation itself is ordinary linear algebra, trivial for a computer. What decides the result is the modelling frame built before it runs, and the software’s job — often unfulfilled — is to keep that frame visible.

The model underneath

The functional unit quantifies the service delivered, not the object compared: not a kilogram of packaging but a litre of product protected for a shelf life. Comparisons between systems stand or fall on this definition, because systems of different durability or functionality are otherwise incommensurable. The system boundary then decides which processes count as inside. The inventory itself is a matrix of unit processes exchanging material, energy and emission flows; scaling that web so it delivers exactly one functional unit is a matrix inversion — the same input-output structure economists use for whole economies. Characterisation factors finally convert each emitted substance into impact units, with CO2 as the reference for warming over a horizon that is itself a convention.

Allocation: where shared burdens are divided

Most real processes have several outputs at once: milk and meat from one herd, naphtha and diesel from one barrel, recyclate and virgin resin from one plant. Their shared burdens must be divided, and division is not measurement. Dividing by mass, by energy content, by economic value, or refusing division through system expansion each yields different numbers. The economic basis is the most unstable: prices move with markets while emissions do not, so an allocation that respects the price structure can reverse which of two packages wins a comparison. Standards prescribe a hierarchy — first try to avoid allocation, otherwise partition — but the basis of partition remains a modelling choice. This single arbitrariness explains a large share of published LCA disagreements, and it lives in the data the software loads, not in its arithmetic. Attributed bio-content enters the same machinery: a certified tonnage from mass-balance traceability becomes an input whose assumptions the LCA inherits without comment.

Uncertainty and visible choices

Inventory entries are averages: one dataset row stands for processes in other countries and other years, and averaging collapses the real variance between them. Uncertainty can be propagated — Monte Carlo over the data distributions puts intervals around results, and many comparative conclusions survive those intervals — but published results are usually point values. Hence the honest criterion for software is not speed, interface or database size; it is whether the data choices are exposed: which dataset version, which allocation rule, which characterisation set, which cut-offs. Two practitioners with the same tool and different choices will legitimately get different answers, and a tool that hides the choices is hiding exactly that fact.