Compare Assembly

Two build-ups, or three, held to the same criteria with the same rigour. The usual failure is not picking wrong; it is comparing an assembly you have costed against one you have not, and calling the first one cheaper.

Inputs the command needs

1. The assemblies, named and listed layer by layer, in order.
2. The criteria. Ask if not given. Common ones: thermal performance, condensation risk, drying capability, fire, acoustic, weight, thickness, buildability, sequencing, maintenance access, cost, embodied carbon, warranty availability, availability of the trade locally.
3. A value per assembly per criterion, each with its source, or the word unknown.

Never substitute a typical value for an unknown. Never carry a value from one assembly across to another because they "use the same insulation".

Workflow

Step 1 — Read the assemblies back

Number the layers of each, in order, and confirm the order is inside face to outside face for all of them. An assembly entered in a different direction from its neighbour produces a matrix that looks fine and compares nothing.

Step 2 — Fix the criteria before looking at any value

Agree the criteria list first and write it down. Criteria chosen after the values are on the table are criteria chosen to reach an answer.

Ask which criteria are thresholds (must pass) and which are trade-offs (more is better). Say back which is which.

Step 3 — Build the matrix

Criterion	Threshold or trade-off	Assembly A	Assembly B	Source per cell	Comparable?

The last column is the point. Mark a row not comparable when one assembly has a figure and another has unknown, when the two figures came from different kinds of source, or when they were measured or modelled differently. Do not fill a gap to make a row look complete.

Step 4 — Report

• Which assemblies fail a threshold, and on which criterion.
• Which leads on each trade-off, among the rows that are comparable.
• The count of rows that are not comparable, listed by name. This number is reported before any conclusion, because it is what decides whether there is a conclusion.
• What to go and find to make the incomparable rows comparable.

If more than a couple of rows are not comparable, say plainly that the comparison is not yet decidable and name what is missing, rather than ranking on the rows that happen to be filled.

Step 5 — Save, if asked

Ask whether to write the matrix to a file and where. Markdown or CSV.

Rules

• Never supply a value. Not a property, not a cost, not a carbon figure, not as a placeholder that looks like a number.
• Never recall a code minimum as the threshold. Thresholds are quoted by the user.
• Fix the criteria before the values are visible.
• An unknown stays unknown for the whole run.
• Report the incomparable count before the conclusion, every time.
• Never rank on partial data without saying the data is partial.

Anti-patterns

• Filling a gap with a typical figure so the table looks finished.
• Comparing a modelled value for one assembly against a tested value for another without flagging it.
• Adding a criterion after seeing the values.
• Ranking assemblies when half the matrix is unknown.
• Quoting a manufacturer's figure found by web search rather than from the data sheet the user names.
• Recommending an assembly. This command reports; you decide.