Comparing Options & Trade-offs

The most tempting way to use Claude for a decision is the one to avoid. You have three massing options for a corner plot, you are tired, and you type: "Which of these is the best option?" Back comes a crisp, reasoned-sounding verdict - Option B, because it balances daylight and efficiency - and it feels like an answer. It is not. It is the most plausible paragraph a language model can write in response to your question, and Claude will defend whichever option it names just as persuasively as it would defend either of the others. It has not visited the site, does not know that your client quietly cares more about the garden than the floor area, and will never be accountable for the choice. A confident verdict from a system that cannot be responsible for it is worse than no verdict at all, because its fluency tempts you to stop thinking.

There is a deeper problem too: a single recommendation hides its reasoning. When Claude says "B is best," you cannot see which criteria it weighed, how heavily, or what it ignored - so you cannot check it, argue with it, or defend it to a client or a jury. The verdict is a black box wearing a confident face.

The fix is a reframing you will use for the rest of your career with these tools. Do not ask Claude to choose; ask it to structure the choice. Claude's real strength in decision-making is not judgement - it is organisation: taking the fuzzy either-or swimming around in your head and turning it into an explicit, comparable, inspectable structure. Options become rows. The things you care about become named criteria. The vague sense that "B feels efficient but A has better light" becomes numbers you set and can interrogate. Once the decision is laid out in the open, the judgement is yours to make - and, crucially, yours to explain.

The workhorse tool is the weighted decision matrix, and Claude builds one beautifully. Options run down the side; the criteria that matter run across the top; each criterion gets a weight reflecting how much it counts; each option gets a score against each criterion; and the weighted total ranks them. The discipline the matrix imposes is the whole point - it forces the vague into the explicit.

What matters is who does what. Claude proposes the structure and does the arithmetic. You supply the two things that actually determine the outcome: the weights and the scores. Those are pure judgement - they encode what this client, this site and this brief truly value - and they are exactly what Claude cannot know. Consider a house on a tight urban plot with three massing options. A useful prompt:

I am comparing three massing options for a house on a
tight urban plot. Build a weighted decision matrix.

Options: A Courtyard, B Linear bar, C Split block.
Suggest 5-6 criteria an architect would weigh here, but
leave the WEIGHTS and SCORES blank for me to fill.
Then once I give you weights and scores, compute the
weighted totals and show your working.

Notice the deliberate hand-off: Claude offers criteria (it is genuinely good at remembering ones you might forget - privacy, cross-ventilation, buildability, future flexibility, cost, contextual fit), but you set the weights and do the scoring against your real knowledge of the project. The matrix in the figure shows the result: Claude has done the multiply-and-add, and Option B edges ahead at 7.4. Build it as an Artifact so it becomes a living table you can adjust in place as you refine your thinking, rather than a paragraph you re-generate each time. And always check the arithmetic - Claude can and does make calculation slips; a weighted total is trivial to verify by hand, so verify it.

The winning number is not the answer - it is the beginning of the real conversation. A matrix that spits out "7.4 beats 7.2" and ends there has taught you nothing; its value is in what the comparison surfaces. Two options separated by a whisker are effectively tied, and the honest read is "these are equivalent on the things I measured, so decide on the things I did not." That is a far more useful conclusion than a false winner.

So push past the total. The single most valuable follow-up is a sensitivity check: "If daylight mattered twice as much and cost half as much, does the ranking flip?" Ask Claude to re-run the weighted totals under two or three different weightings and tell you which option wins under each. If one option wins across every reasonable weighting, you have a robust choice. If the winner changes every time you nudge a weight, the decision really turns on your priorities, not on the options - and that is worth knowing before you commit.

Then use Claude as a foil. Ask it to steelman each option in turn - "make the strongest possible case for Option A, as if you had to defend it to the client" - and to name explicitly what each option trades away. This is where Claude earns its keep: it will articulate the courtyard's daylight-and-privacy win and its floor-area cost more even-handedly than you, quietly rooting for your favourite, tend to. Guard against one bias, though: Claude is agreeable by default and will drift toward whatever you signal you prefer. Counter it directly - "I am leaning towards B; argue as hard as you can for why B is the wrong choice." A tool that only confirms your instinct is worse than useless; a tool that stress-tests it is a genuine partner.

The other quiet service the comparison performs is surfacing criteria you never wrote down. Ask Claude, once the matrix is built, "what would a demanding client, a planner, or a contractor add to this list that I have left off?" - and it will often name something real: maintenance burden, phasing, resale, acoustic separation, the ease of a future extension. You may decide those do not belong in the matrix, but you decide it knowingly rather than by omission. That is the deeper value of structuring a decision with Claude: not the tidy total at the bottom, but the discipline of having looked at the choice from more angles than a tired mind holds at once.

The matrix is a thinking aid, never an oracle - and it is only as good as the judgement you pour into it. Garbage weights and careless scores produce a confident, precise, entirely wrong total. So the number never decides; you do. Often you will, and should, override it: some things that matter most resist scoring - delight, contextual fit, the way light moves through a room, the client's unspoken dream - and a mature designer will pick the option that scored 7.2 because the criterion that mattered most was never in the matrix. That is not a failure of the method; it is the method working, having made the trade-off visible enough for you to overrule it knowingly.

How hard you should lean on the structure scales with the stakes, as always. Choosing between two door handles does not need a weighted matrix; choosing a structural system, a facade strategy or a site-wide servicing approach absolutely does - there the discipline of naming criteria and weights protects you from a plausible-but-poor snap decision, and gives you a record.

And record it you should. Once you have decided, Claude turns the matrix straight into a decision memo - the options considered, the criteria, why you weighted as you did, and the reasoning behind the final call (drawing on the writing skills of Module 5). Six months on, when a client asks why the courtyard scheme was set aside, or when you defend the choice to a planning committee, that memo is worth its weight in fees. This is the full loop the figure shows: you direct, Claude structures and computes, you judge, you decide, and you own - Claude never closes it for you.