What does Scorecard: Turning 'Better' into a Judgable Fact do?
The entire judgment of the 37% rule depends on one sentence: Is this candidate better than the baseline? Without a scorecard, that sentence can only rely on gut feeling—and gut feeling drifts with the weather, fatigue, and the other person's attitude. The scorecard must be set before seeing the first candidate, and it must not be changed during the review process.
1. Design Rules (Five)
-
3-5 dimensions, max 7. More than 7 dilutes the weight of each dimension, making every impact negligible, and the total scores won't separate meaningfully.
-
Minimize overlap between dimensions. "Salary" and "Income stability" are highly correlated—that's weighting the same thing twice.
-
Hard constraints should not be scoring dimensions; they should be veto thresholds. For example, "must have a household registration", "budget cap of 5 million", "must support SSO"—these are one-veto items. If they aren't met, the candidate is eliminated immediately, and they should not be averaged with "growth potential." (In the
score.pyscript, thevetofield records veto items; instop.py, the candidate is simply skipped.) -
Define anchors for each dimension before scoring. Write a specific description for scores of 1, 5, and 10. Then, when scoring, refer to the anchors, not your gut. Example: "Commute" 10 = within 20 minutes one-way; 5 = 40 minutes; 1 = over 90 minutes.
-
You must be able to say 'why this one is the most important' for every weight. A weight you can't explain is probably someone else's value system borrowed. Weights are the only thing the user must personally decide in this method—don't gloss over with defaults.
2. How to Assign Weights (Users Often Get Stuck Here)
Don't ask "What percentage does each dimension take?"—people have no intuition for absolute weights. Use these two approaches:
Method A: 100-Point Distribution "I'll give you 100 points to divide among the dimensions you care about most. The most important one gets the most first." Then let the user adjust.
Method B: Ranking (Faster) First rank them, then apply a default decreasing value and let the user fine-tune:
| Rank | Default Weight |
|---|---|
| 1st most important | 35 |
| 2nd most important | 25 |
| 3rd most important | 20 |
| 4th most important | 12 |
| 5th most important | 8 |
The defaults are just a starting point. Always ask: "Do you agree with this distribution?" If not, let them change it.
3. Common Scenario Dimension Lexicon (For Reference, Not to Copy)
Choosing an Offer / Changing Jobs Total compensation · Growth potential · Direct supervisor · Work-life balance · Company stability · Commute · Resume value
Renting / Buying a Home Commute time · Price · Natural light and ventilation · Surrounding amenities · Property condition · Landlord/property management · Noise
Hiring Skill match · Learning speed · Communication and collaboration · Stability · Salary expectations · Cultural fit
Selecting a Vendor / Procurement Price · Delivery capability · After-sales support · Compliance and qualifications · Long-term stability · Migration cost
Technical Solution Selection Team familiarity · Performance · Ecosystem maturity · Operational cost · Migration cost · Long-term maintainability
Investment / Finance Expected return · Maximum drawdown · Liquidity · Understanding (don't invest in what you don't understand) · Correlation (risk diversification)
The lexicon's purpose is to remind, not to serve as a template. Pick only 3-5 each time. After picking, ask the user: "Is there anything you truly care about that isn't listed here?"—the one you miss is often the most important.
4. Three Scoring Pitfalls
| Pitfall | Symptom | Solution |
|---|---|---|
| Grade Inflation | All candidates score 7-9, no differentiation | Force anchor descriptions; allow scores of 2 or 3 |
| Halo Effect | A candidate "feels good," so every dimension gets a high score | Score by dimension, not by candidate—score "Salary" for all candidates first, then move to the next dimension |
| Recency Bias | The most recently reviewed candidate scores systematically higher | Record scores immediately during the review; don't delay until the next day; if delayed by days, mark confidence as low |
5. Confidence Level (Important but Often Overlooked)
For each scored candidate, mark a confidence level on the spot: High / Medium / Low.
- High: Sufficient information (already visited in person, already spoken with, complete data)
- Medium: Key information is available, details pending confirmation
- Low: Mostly speculation or based on one-sided claims
Confidence does not participate in the calculation, but it must be presented in the conclusion. When two scores are close but confidence levels differ significantly, choose the one with higher confidence—that is far more meaningful than a two-decimal-point difference in scores.