When comparing two job offers on a weighted scorecard, why fix the criteria weights before entering any numbers?
answer
- Order of operations, not arithmetic
- Rows, columns, and one locked column of weights
- What existed before the anchor did?
- Roughly six rows, each with named evidence
- Re-weighting after the totals is the tell
basics
~20 sWeights chosen after the scores get bent until the offer you already prefer wins. Fixing them first — criteria in rows, offers in columns, weights locked — makes the comparison a test of your preference rather than a receipt for it.
solid answer
~50 sA weighted scorecard puts your criteria in rows and the offers in columns, with a weight on each row that you set before a single figure goes in. Roughly six criteria works: adjusted pay, level and scope, manager and team, growth, stability, and life fit. Score each cell on one small scale, multiply by the weight, add up the columns. The point of locking the weights first is that they encode what you wanted *before* an anchor existed. Once you have seen that one package is larger, every weight becomes negotiable in your own head, and the exercise degrades into justification. If, after the totals, you badly want to re-weight a row, treat that as evidence — say which criterion you under-weighted and why, change it consciously, and re-run both columns rather than nudging one.
go deeper
Know the shape of the tool: criteria in rows, offers in columns, a weight per row, and the weights written down before any offer figure goes in.
Explain the mechanism — the weights record what you wanted before an anchor existed, which is the only thing separating a real comparison from a justification.
Demonstrate handling the messy cases: thin evidence on a row, near-tied columns, and the honest way to change a weight once rather than nudging it repeatedly.
Own the limits: the weights are personal, the scale is coarse, and the table informs a judgement rather than replacing it. Say where you would override it and why.
## The artifact The weighted scorecard is a small table: **criteria in rows, offers in columns, weights fixed before any number is entered**. It is the most useful object in offer evaluation precisely because it is boring — its value comes from the order in which you fill it in, not from its arithmetic. ## Building it **Step 1 — choose about six criteria.** Fewer than four and you are back to comparing headline pay; more than eight and every row's weight shrinks until nothing moves the result. A serviceable six for an engineering offer: adjusted total pay, level and scope, manager and team, growth and learning, stability and risk, life fit (location, on-call, hours). **Step 2 — weight them, now, before you have any offer in hand.** Weights that sum to one hundred are easiest to argue about. Deliberately uneven, non-round weights force real decisions — for example 24 on level and scope, 19 on adjusted pay, 17 on manager and team, 15 on growth, 13 on life fit, 12 on stability. Those exact figures are illustrative only, not a recommendation and not market data; yours should come from your own situation. **Step 3 — write down what each row's evidence will be.** "Manager and team" is not a mood. Decide in advance that you will score it on things like whether the manager could describe what the role owns, how they talked about a project that went badly, and whether the people you met asked you anything specific. Rows without a named evidence source are rows you will fill in from whichever way you are already leaning. **Step 4 — only now enter the offers.** Score each cell on a small scale — one to five is plenty. Apply any cost-of-living correction to the pay row *before* scoring it, so the row you compare is what the package buys, not the number printed on it. **Step 5 — total each column and read the result honestly.** An illustrative pair of totals might come out 3.86 against 4.12 on a five-point scale. That gap is real but not enormous, which is the usual outcome and is itself information. ## What to do with a near-tie Most genuine choices land close. When two columns finish within a few hundredths, the scorecard has told you the truth: on the criteria you named, these offers are equivalent, and no further arithmetic will separate them. Do not invent a seventh criterion to break the tie — the criterion you invent after the totals is chosen to produce the answer you want. Better moves: go back and gather one more piece of evidence on the row you scored most weakly; or accept that a tie licenses you to decide on something you deliberately kept out of the table, and say so. ## Where scorecards go wrong - **Motivated re-weighting.** The totals come out, one column loses, and suddenly growth "obviously" deserves more weight. Sometimes it genuinely does — the fix is to state the change, apply it to both columns, and note that you made it. - **Adding a row late.** Same defect, wearing a new hat. - **Scoring rows you gathered nothing on.** An unevidenced score is noise multiplied by a weight. - **A pay row so heavy that the rest is decorative.** If pay carries most of the weight, the other rows exist to make the decision look considered. Weight it honestly and let it win openly if it wins. - **False precision.** A five-point scale and six rows do not resolve a two-hundredths difference. Treat close totals as ties. ## Rehearsing it The scorecard gets much better when someone argues with it. A mock-interview partner is ideal for this because their instinct is already to probe: hand them the table and have them ask, row by row, *why is that weight there, what did you score it on, what evidence would change it?* Two things usually happen. A weight you cannot defend collapses within a sentence, which is the table working as intended. And you find yourself saying the decision out loud in one line — for example "The higher number is at a lower level, with a manager I did not click with" — which is the sentence you will still be able to defend to yourself a year from now. ## Why this survives contact with reality The scorecard does not compute the right answer; nothing does. It does one narrow, valuable thing: it separates *what you wanted* from *what you were offered*, by recording the first before the second exists. That is the entire mechanism, and it is why the order of operations is not a formality.
- The two columns finish within a few hundredths of each other — what does that tell you?That on the criteria you named, the offers are equivalent, and more arithmetic will not separate them. Do not add a seventh row after the fact; the row you invent late is chosen to produce the answer you already want. Either collect one more piece of evidence on your weakest-scored row, or accept that a tie frees you to decide on something you consciously left out.
- How do you score manager and team when you only spent a couple of hours with them?Score the observations you actually have, and say the confidence is low. Useful signals: whether the manager could describe what the role owns without hedging, how they talked about something that went badly, whether the engineers you met asked you anything specific. If the row is thin, the honest response is to gather more evidence before deciding, not to fill it in from a mood.
- You want to increase one row's weight after seeing the totals. Is that always cheating?No, but it has to be conscious. Say which criterion you under-weighted and why you now think so, change it explicitly, and re-run both columns with the new weight. What is not acceptable is nudging weights silently until the column you already preferred moves ahead — that is the failure the fixed-weights rule exists to catch.
- Why not use twenty criteria to be thorough?Because weight per row falls until nothing changes the outcome, and twenty rows guarantee that most are scored on no evidence at all. Around six rows keeps each weight large enough to matter and each row backed by something you actually observed. Thoroughness lives in the evidence behind a row, not in the number of rows.
saying these in an interview costs you the question
- Setting the weights only after seeing which package is larger
- Adding a criterion late to break a tie the preferred way
- Scoring rows for which no evidence was ever collected
- Weighting pay so heavily that every other row is decorative
- Treating a two-hundredths gap between columns as a decisive result
- Presenting the table as objective when the weights are personal by design