Trade-offs, uncertainty and transparent decisions

Several goals can remain in tension

A design may seek low mass, low resource burden, manufacturability and durability simultaneously. One candidate can improve one objective while worsening another. Show the consequences rather than hide them in an unexplained weighted score.

Suppose five finished mounting modules could locate the same small instrument. Each alternative is a complete material, forming and joining choice within the same interface envelope. The comparison concerns an imposed 100 N static force at a specified loading point, at 23 °C after the same conditioning. The required displacement at the instrument datum is no more than 0.500 mm. All other requirements in this deliberately restricted problem are stipulated satisfied; these few numbers do not qualify an actual instrument mount.

For each manufactured module, m is installed mass in kilograms, including its retained joints and fittings. W is the electricity entering the specified preparation, forming and joining operations per accepted finished module, in megajoules (MJ). The supplied account includes electricity used on rejected pieces allocated to accepted output. It excludes raw-material production, transport, use, maintenance and recovery. W is therefore a particular manufacturing input, not a complete environmental impact, carbon footprint or life-cycle assessment.

The table is original invented data, not measured material ratings. In this first comparison every displacement is an exact stipulated value for the stated test. Small displacement is a mandatory constraint; lower m and lower W are two separate objectives. Treating displacement as merely another score would let low mass compensate for an unusable mount. Functions, constraints and selection criteria develops that distinction.

Screen first, then remove avoidable sacrifices

A–D meet the 0.500 mm screen; E does not. E would look attractive on both objectives, but its low mass and low electricity cannot repair its excessive displacement. Keep its failed requirement in the record rather than allowing it to win an unrelated score.

Among the feasible alternatives, B uses less mass and less manufacturing electricity than D: 0.300 < 0.350 kg and 6.50 < 7.00 MJ. B dominates D for these two minimised objectives: it is no worse in either and strictly better in at least one. If every consequential requirement is already included, choosing D would sacrifice something without gaining either declared objective. An omitted benefit such as access or repairability would require expanding the comparison, not pretending that benefit has been measured by W.

A, B and C are nondominated in this finite set. Moving from A to B increases installed mass by 0.100 kg but saves 1.50 MJ per module. Moving from B to C adds another 0.100 kg but saves 2.50 MJ. None is at least as good as another in both objectives. The set of such alternatives is the Pareto set; its plotted objective points form the finite Pareto front. It identifies unavoidable trade-offs among the supplied options, not a unique answer.

The first figure keeps actual alternatives as separate points. A line between A and C would not establish a manufacturable intermediate module. Dominance here is conditional on the stated feasibility and accounting boundary. It does not say A, B or C is intrinsically the best constituent material, or that unlisted alternatives cannot improve both objectives.

A score encodes a preference and a scale

One possible decision rule uses fixed reference scales m₀ = 0.200 kg and W₀ = 4.00 MJ/module. These are declared comparison anchors, not the smallest observed values of a changing candidate list. Define the dimensionless penalty S = w(m/m₀) + (1 − w)(W/W₀), with 0 ≤ w ≤ 1, and prefer a smaller S among feasible alternatives. Both ratios are dimensionless. w expresses a preference on these normalised scales; it is not a measured material property or a probability that mass matters.

With the nominal table, S_A = 2 − w, S_B = 1.625 − 0.125w and S_C = 1 + w. At w = 0.25 the penalties are 1.750, 1.59375 and 1.250, so C is preferred. At w = 0.75 they are 1.250, 1.53125 and 1.750, so A is preferred. Solving 2 − w = 1 + w gives w = 0.500. At that weight A and C tie at 1.500; B is 1.5625.

This linear score never selects B for any permitted weight, even though B is nondominated. At B’s mass, the arithmetic midpoint of the A and C objective points has W = 6.00 MJ, below B’s 6.50 MJ. Consequently S_B = (S_A + S_C)/2 + (1 − w)/8. For w < 1 it is greater than the average and therefore greater than at least one of A and C; at w = 1 A is lighter. This is an algebraic comparison, not permission to build the nonexistent midpoint module.

B can still be a reasoned choice under a different explicit rule. If the brief additionally requires m ≤ 0.320 kg and W ≤ 7.00 MJ/module, B is the only remaining option: A misses W, C and D miss mass, and E still misses displacement. This is a changed constraint problem, not a way of adjusting weights until a favoured candidate wins.

Never add raw kilograms and megajoules and call the result a physical performance. Changing kg to g while retaining the same raw coefficients can reverse a ranking. A consistent unit conversion leaves m/m₀ and W/W₀ unchanged; changing an anchor while retaining the same w changes the stated preference. Linear compensation is itself an assumption. Mandatory limits, unacceptable modes and unquantified attributes must remain visible outside a convenient score.

Test whether the decision is robust

Vary consequential uncertain inputs and assumptions. If small changes reverse the choice, identify the evidence worth improving. A robust feasible option can be preferable to an apparent winner resting on a narrow optimistic assumption.

Preference uncertainty and evidence uncertainty differ. A disagreement about w calls for a clear decision about priorities. An uncertain displacement or manufacturing input calls for better evidence about the finished alternative. Another test cannot decide how much mass a stakeholder should exchange for a given electricity saving; another opinion cannot establish a module’s mechanical response.

Now replace the exact displacement assumptions with a separate supplied bounded account: δ_A ∈ [0.380, 0.560] mm, δ_B ∈ [0.400, 0.460] mm, δ_C ∈ [0.420, 0.480] mm, δ_D ∈ [0.460, 0.500] mm and δ_E ∈ [0.760, 0.840] mm. These are invented complete admissible sets for the same defined task, not confidence intervals or a distribution estimated from the preceding five values. Every value in each interval is permitted; no probability is supplied.

B, C and D meet the displacement limit throughout their intervals. E fails throughout. A has both passing and failing permitted values, so its feasibility is unresolved. Its midpoint below 0.500 mm does not establish compliance throughout the set. If the decision requires every permitted state to pass, admit B–D and retain A as an unresolved alternative needing evidence. This is a robust screen with respect to this stated set only, not a guarantee against omitted mechanisms or conditions.

Measurement variation can arise within a sequence and across days, operators or environmental conditions. Repeating a measurement many times under unchanged conditions need not resolve a missing service-state assumption. Explain how a supplied bound was justified before using it for a real decision. Here the bounds are simply teaching inputs; no instrument uncertainty, fatigue life or long-term reliability is inferred.

Find the evidence that can change the choice

For a combined uncertainty exercise, retain those displacement sets and set w = 0.500 with the same fixed anchors. B, C and D are robustly admitted; B still dominates D because their objective values remain fixed. Now replace C’s nominal W with W_C ∈ [3.50, 6.00] MJ/module. All values in that range are allowed, while B’s W stays 6.50 and all installed masses stay fixed. The complete admissible model permits every supplied combination; it supplies no correlations or probabilities.

B has S_B = 0.5(0.300/0.200) + 0.5(6.50/4.00) = 1.5625. C has S_C = 0.5(0.400/0.200) + 0.5(W_C/4.00) = 1 + W_C/8 when W_C is expressed in MJ/module. Their equality is at W_C = 4.50 MJ/module. Below it C is preferred; above it B is preferred; at it they tie. C’s score spans 1.4375–1.750, so the original range supports both rankings. A cannot be silently restored from a favourable nominal value because its displacement screen remains unresolved.

The second figure plots this exact threshold, keeping the fixed preference separate from the uncertain input. It is a sensitivity diagram, not a frequency distribution or proof of how electricity varies in a factory. No population likelihood follows from the length of a coloured part of the interval.

A newly justified C electricity range [5.00, 6.00] would make B preferable throughout the supplied set under this fixed score and robust screen. A range [3.50, 4.40] would instead make C preferable throughout. A very precise result for D’s mass would not resolve this B–C ambiguity; it is already dominated. Better A displacement evidence might reintroduce A, so the report must state whether resolving that excluded uncertainty is within the current decision.

An actual next investigation must compare the same input boundary per accepted output, including yield allocation, and establish the relevant finished displacement under the required conditions. Its cost, delay and ability to resolve the threshold matter. There is no universal rule to measure the parameter with the largest numerical range: a smaller uncertainty crossing a feasibility or preference boundary can be the consequential one.

Make the recommendation conditional and reproducible

A useful decision record names the function, candidate material–process–joining states, mandatory limits, objective definitions and accounting boundaries. It then records input provenance, unresolved data, the feasible and nondominated sets, any weights and fixed scales, sensitive thresholds and what further evidence would change the conclusion. State the selected rule before reporting its result.

For the combined bounded example, an accurate conclusion is: B and C remain candidates under the 0.500 mm robust screen; at w = 0.500 the preference changes at C’s manufacturing electricity of 4.50 MJ/module, and the present range crosses that threshold. No unique B–C recommendation follows from that rule and evidence. D is dominated for the declared objectives; E fails; A needs narrower displacement evidence before re-entry. This retains alternatives for a specific reason rather than declaring every choice equally good.

Changing service duration, loading, inspection access or the definition of environmental performance can change which data and objectives matter. A static datum-displacement comparison does not establish durability, and manufacturing electricity alone does not establish whole-life resource burden. Carry the same transparency into service and material-cycle lessons rather than using this illustrative score as a universal selection method.

Worked examples

Try the reasoning

1. Must every objective be collapsed into one number?

   Show the answer and reasoning

   No.

   A transparent trade-off comparison can preserve separate objectives.

2. What makes an uncertainty consequential?

   Show the answer and reasoning

   It can change feasibility or the decision.

   Its role matters, not only its numerical size.

3. Under the nominal exact data, use w = 0.60. Find all three nondominated scores and the preferred option.

   Show the answer and reasoning

   A: 1.400; B: 1.550; C: 1.600. A is preferred.

   Substitute into 2−w, 1.625−0.125w and 1+w. This is a result of the specified scales and preference, not a universal material ranking.

4. A colleague enters m in grams in S = w(m/m₀)+(1−w)(W/W₀) but keeps m₀ = 0.200. What correction preserves the original decision?

   Show the answer and reasoning

   Use m₀ = 200 g when m is in grams, or convert the entered mass back to kilograms.

   Both numerator and reference must use the same units. 0.300 kg/0.200 kg equals 300 g/200 g = 1.5; 300/0.200 changes the mass term by 1000 and changes the rule.

5. Keep the nominal displacement screen and add mandatory m≤0.320 kg and W≤7.00 MJ/module. Which alternative survives, and may the caps be replaced by a favourable weighted average?

   Show the answer and reasoning

   Only B survives. The mandatory caps cannot be replaced by compensation in a score.

   A misses electricity; C and D miss mass; E still misses displacement. B meets all three. B’s absence from the linear score winners never implied it was an unreasonable compromise under another rule.

6. Under the bounded displacement screen and w = 0.500, newly established W_C = 4.50 MJ/module is exact. Does it favour C over B? What if the justified range is [3.50, 4.40]?

   Show the answer and reasoning

   At 4.50 they tie. Over [3.50, 4.40], C is preferred throughout.

   C’s score is 1+4.50/8 = 1.5625 at equality. At 4.40 its maximum is 1.550, below B’s 1.5625; its lower values reduce the score further.

7. New relevant evidence narrows A displacement to [0.380, 0.490] mm while its nominal mass/electricity stay fixed. At w = 0.75, may it still be excluded using the old interval?

   Show the answer and reasoning

   No. It now meets the robust displacement screen, and its 1.250 penalty beats nominal B 1.53125 and C 1.750.

   Update the admissible set using the new justified evidence and re-screen before ranking. The old unresolved classification must remain recoverable as history, not be used as current contrary evidence.

8. A decision maker accepts A because its displacement midpoint passes and calls that the same robust rule. Identify the missing change.

   Show the answer and reasoning

   The acceptance rule changed; the complete permitted set still includes displacements above 0.500 mm.

   A midpoint rule requires its own justification. No probability is supplied to turn the amount of passing interval into a reliability level.

9. The B–C ranking is unresolved because W_C crosses 4.50 MJ/module. A proposed experiment refines D mass from 0.350±0.001 kg to 0.350±0.0001 kg. Would it resolve that ranking? Would manufacturing electricity alone answer a new whole-life environmental question?

   Show the answer and reasoning

   Neither conclusion follows. The proposed mass refinement does not resolve W_C, and the electricity boundary omits other life-cycle inputs and impacts.

   Seek relevant matched accepted-output electricity evidence or explain why another decision-sensitive uncertainty matters. New whole-life objectives need a stated functional service and wider system boundary; they cannot be relabelled from this partial account.

What to carry forward

• Expose trade-offs and assumptions.
• Prioritise evidence that can change the decision.
• Screen requirements before allowing objectives to compensate.
• Nondominated alternatives need an explicit preference or another stated decision rule.
• Fixed normalisation anchors and consistent units are part of a weighted comparison.
• A linear weighted score can miss a nondominated discrete compromise.
• Separate uncertain performance evidence from uncertain priorities.
• Record sensitive thresholds and unresolved cases; a bounded screen only covers its supplied set.