A fund research house must rate three funds on people, process, parent, performance and price: one cheap with an average record, one with the best 5-year return and a new manager, one with a long-serving manager and high fees. Score them with stated weights and explain the ranking.

Fund R ranks first (3.75 out of 5), Fund P second (3.50) and Fund Q last (3.05), despite Q's best five-year return. With people, process and price weighted at 25% each, R's long-serving manager and process outweigh its fees, P's low cost lifts an ordinary team, and Q's record belongs to a manager who left. Weight past performance at 60% and Q would come first, which is why the weights must be stated.

When a restaurant's famous chef leaves, last year's reviews tell you about a kitchen that no longer exists. A forward-looking rating asks how likely a fund is to do well from here, so it weighs who runs it and what it costs more heavily than what it did in the past. Past performance enters, but as evidence that the people and process work, not as the score itself. Price earns a heavy weight because it is the one thing known in advance: every year, the fee is subtracted for certain.

Set the weights before scoring, and say them out loud: people 25%, process 25%, parent 10%, performance 15% and price 25%. Then score each pillar 1 to 5 from the facts. Fund Q gets 2 on people, because its record belongs to someone else, but 5 on performance; Fund R gets 5 on people and 2 on price; Fund P gets 5 on price and 3 on everything else.

Fund R first: fourteen years of the same manager and a documented process make it likely that the 1.1 points of past outperformance after fees came from skill that is still in the building. Its 2.1% fee is the real drag; it needs to beat a cheaper rival by more than a point a year before fees just to keep pace. Fund P second: nothing distinctive, but at 0.9% an investor keeps more of whatever the market gives. Fund Q last for now: the five-year record cannot be credited to the new manager, so the rating should be held at a provisional level and revisited after he has a record of at least two to three years.

The weights decide a lot. Weighted 60% on performance and 10% on everything else, the order flips to Q 4.10, R 3.90, P 3.20. That does not make one set of weights right; it shows the rating is a judgement, and the published method should say which judgement it makes. The scores for people and process are themselves opinions formed in meetings, so record the evidence behind each one, and check the ratings later against what happened, which is the only test of whether the method works.