Use weights only after eligibility
Weights cannot repair a failed requirement. Start with a shortlist whose exact variants have passed your non-negotiable gates or have only low-consequence unknowns that you deliberately accept. Then use weights to describe trade-offs among eligible candidates. State the job at the top of the table. A commuter bag, occasional event layer and daily walking shoe need different criteria even when the product images look similar. Hacoo's public discovery and creator environment can provide many attributes, but the amount of content around an item should not determine its weight. Weight is your statement about consequence: how much would this factor change whether the product performs the job and gets used? This distinction stops a visually prominent feature from dominating simply because it is easier to see or more frequently discussed.
Choose four to six independent criteria
A short set is easier to explain and less likely to count the same idea twice. For clothing, possible criteria include fit-evidence strength, compatibility with the intended layer, care burden, material-information completeness and current item-price fit. For a bag, capacity evidence, closure, carry method, empty dimensions and price may be more relevant. Test independence by asking whether two criteria would rise and fall for the same reason. “Fits the laptop” and “adequate internal width” may be duplicates. Combine them or define different roles. Avoid universal categories such as quality, value or best. They hide multiple judgements and encourage unsupported claims. Each criterion should have a one-sentence scoring rule that identifies what a low, middle and high score means using evidence available for the selected option.
Assign weights from consequences
Imagine the top criterion failed while the others were strong. How much would that change the decision? Use that answer to rank the criteria before assigning numbers. Give the highest-consequence factor the largest share, then distribute the remainder so all weights total 100. Round numbers such as 30, 25, 20, 15 and 10 are usually enough. Decimal weights imply a precision the reasoning does not have. Write one justification beside each weight: “Fit evidence receives 30 because unusable dimensions defeat the layering brief.” If you cannot explain the difference between 20 and 25, use the same weight or simplify. Set weights before scoring candidates. Adjust them only when the use case changes, and then rescore every option. Changing a weight after seeing the total is a common way to make a favourite appear objectively superior.
Define an evidence-based scale
A five-point scale works when every level has a meaning. Zero can mean direct evidence contradicts the criterion; one weak support; three adequate support with a remaining limitation; five strong direct support for the exact option. Unknown should stay unknown rather than becoming 2.5. Add a separate confidence label—high, medium or low—to show how well the score is supported. A precise measurement linked to the variant can create higher confidence than a general size word. Several current consistent images may support visible construction better than one cropped post. The confidence label does not rate product quality. It rates your knowledge. This separation prevents a polished page from receiving an inflated performance score and allows two candidates with similar totals to be distinguished by the strength of evidence behind them.
Calculate a range when important facts are missing
For each known score, multiply the criterion weight by the score and divide by five. For a 30-point criterion, a score of four contributes 24. When an important factor is unknown, calculate a lower and upper bound instead of choosing an invented middle. The lower bound uses the lowest defensible result; the upper bound uses the highest result the available evidence could reasonably support. If Candidate A spans 62–82 and Candidate B sits at 74 with stronger confidence, the ranking is unresolved. The missing fact could change the winner. If A spans 76–82 and B reaches 68, the uncertainty may not alter the conclusion. Ranges turn missing information into a practical research priority. Resolve the unknown with the largest ability to change the result rather than collecting more facts that have no decision impact.
Run a sensitivity check
Change the two largest weights by five or ten points while keeping the total at 100. Does the winner change? If a tiny adjustment reverses the result, the score does not support a confident ranking. Treat the candidates as effectively tied and use stronger evidence, lower downside or the original brief as the tie-breaker. If the result remains stable across reasonable weights, the decision is less dependent on an arbitrary number. Also remove one low-weight preference entirely and recalculate. A candidate should not win solely because several decorative details were counted separately. Sensitivity checking is more honest than adding decimals. It shows where judgement matters and prevents the final score from looking like an external measurement. The numbers organise your reasoning; they do not discover an objective best product.
Read totals beside evidence and downside
Never publish or save a total without the gate result, confidence labels and key unknowns. A score of 78 supported by exact measurements and clear option contents is different from 82 built on indirect images and a general description. Compare the downside if the most important assumption is wrong. The candidate with a slightly lower total may be the safer decision when its evidence is clearer and the consequences of error are smaller. Use the total to focus discussion, not end it. Write a short explanation: which two criteria created the lead, which unknown could still matter and why the weighting fits the intended use. This narrative protects the table from being copied into a context with different needs. Another buyer may rationally choose different weights because the job is different.
Finish with a rounded decision rule
Choose a simple rule before accepting the result: all gates pass, no high-consequence factor remains low confidence, and the leading candidate stays ahead under a reasonable sensitivity check. If the rule fails, recheck the named fact or choose none. Do not convert a score such as 76.4 into a claim that one product is 2.7 percent better. Round to whole numbers and describe the difference qualitatively—clear lead, narrow lead or unresolved tie. Reopen the live variant before acting and update any changed input. The method cannot guarantee performance, quality, authenticity, availability or delivery. It can show that the selected candidate better fits your declared criteria under the evidence available on the checking date. That is the appropriate level of confidence for a practical Hacoo comparison.