Interactions make project savings non-additive

A package is not generally the sum of standalone measure promises. Measures can affect the same load, change the base on which a later measure operates or share an enabling system. A supply change can serve demand that an efficiency measure later removes. An added process can create load to which a subsequent efficiency assumption applies. These interactions change the meaning of incremental benefits. The analyst should identify the affected states and compare the complete package before treating individual reductions as independent opportunities.

Overlap is a physical relationship before it is a ranking problem. If an efficiency measure removes part of a load, a supply measure can serve only the remaining load unless another use is represented. If supply already covers most annual demand, a later efficiency measure may reduce renewable energy used rather than the grid energy that would have been removed in a standalone case. The package can still have a strong final result. The incremental attribution differs because the preceding state differs. A measure does not possess an invariant annual abatement amount outside its package and reference state.

The paired overlap studies make this distinction explicit. They contain the same efficiency and supply measures, the same baseline, factors, prices, capital, operating costs and financial convention. Only the supplied sequence is reversed. In this particular pair the final demand, energy mix, emissions and financial results are identical. The emissions reduction assigned to each step is different. The result is package invariance with changing marginal attribution. It does not establish that order never matters; it establishes exactly what order changes in this defined package.

Incremental attribution can be useful for explaining the path from the baseline to the package. It shows which change occurs at each supplied stage and helps reveal overlap. It becomes misleading when those increments are separated from their sequence and presented as standalone opportunities. A later measure's displayed reduction can be much smaller after an earlier measure has already changed the state. That difference is not necessarily inferior engineering performance. The reviewer should preserve the preceding state and identify whether the increment answers an implementation question, an attribution question or a selection question.

The sequence in the annual engine is an order of state transformations. It is not a construction schedule, a commissioning plan or an optimisation result. A real project may require some measures to precede others because of connection, control or process dependencies. The model does not enforce those requirements. It can show a conditional supplied order and compare its reversal, but the operating team must determine whether either order is feasible and whether the percentage scope represents the actual affected loads. The numerical permutation is a challenge, not an implementation instruction.

Added demand provides a counterexample to universal order invariance. In the sequence-interaction study, demand is added before a percentage reduction is applied. Reversing the order applies efficiency to the original demand and then adds the new load. The final demand therefore differs under the engine's supplied interpretation. The study exposes a physical assumption: whether the efficiency measure can affect the added load. The appropriate engineering response is to examine that scope, not assume that whichever numerical order yields the larger benefit is the correct implementation plan.

A load-specific model could represent those relationships differently. If the new load is unaffected by the efficiency measure, applying the facility-wide percentage after addition would overstate the saving. If the efficiency intervention genuinely controls the enlarged system, the later percentage can be a reasonable abstraction. If the measures affect separate loads, a different decomposition may be preferable. The annual engine uses the supplied remaining-demand convention because it is transparent. The engineering record must establish whether that convention is adequate for the actual package under consideration.

Shared infrastructure introduces another kind of interaction. Connection work, controls, supporting equipment or operating capability may enable several measures. Counting their full cost against each standalone measure can overstate package expenditure; omitting them because no individual measure owns the interface can understate it. The screen sums listed upfront capital and constant operating costs without a dependency architecture. The analyst should allocate or represent shared terms coherently in an expanded case. The committee needs to understand both the cost of the complete operating state and the consequence of excluding a measure.

The same issue applies to avoided costs. A package may eliminate a system only when all required measures are implemented. A standalone measure cannot claim the entire removal saving if the old system remains necessary. Conversely, a package that changes supporting demand should retain the resulting expenditure even if it sits outside the principal equipment contract. The review should trace every claimed shared benefit to the condition that makes it available. A package calculation with coherent scope is stronger than a collection of attractive component business cases that cannot coexist.

Selection should therefore follow dependency and package analysis. An abatement-cost curve that sorts supplied increments by price can be a useful communication device, but it can suggest independent selection when increments overlap. The Transition lab retains the supplied sequence for its incremental display. It does not solve a constrained measure portfolio or search all feasible packages. The committee should treat the increments as an explanation of a defined package and develop the relevant alternatives separately. A sorted visual should never be allowed to create feasibility or independence that the underlying model lacks.

The interaction record can prevent a negotiation error. A buyer may ask each supplier to guarantee a standalone saving, then discover that the combined installation removes the same waste more than once on paper. A stronger procurement record identifies the shared state and the basis for package acceptance. Suppliers can still be held accountable for their defined mechanism, but the total claim is reconciled at the relevant perimeter. This avoids asking separate contracts to prove additive benefits that the physical system cannot deliver.

Staging can be diagnosed by separating the state needed for a later measure from the benefit claimed for the first stage. An interface may make conversion possible while delivering little immediate reduction itself. Removing that interface from the pathway should prevent the later state; if another credible route still permits conversion, the particular design is an option rather than a necessary enabler. The annual screen cannot value that strategic choice or enforce its dependency. Keep the enabling condition explicit instead of reallocating later reductions to make every stage look attractive. This preserves the difference between a weak standalone ratio and a coherent implementation sequence.

The ultimate test is whether the package description and its numerical state tell the same story. If the narrative says the efficiency measure affects only existing equipment while the arithmetic applies it to a newly added load, the result requires repair. If the narrative says surplus energy will be sold while the model curtails it without revenue, the commercial case requires expansion. The interaction review uses such discrepancies to direct development work. It improves the recommendation by identifying which relation must be supported, rather than by making the package appear uniformly favourable.