Interview odds, not job matches: what our score measures and what it doesn’t

Every job product eventually puts a number next to a listing. Most will not tell you what went into it, which is a reasonable thing to be suspicious of. So here is ours, in full, including the parts that make it look worse.

What it estimates

The score estimates how likely you are to get an interview for a specific role. Not whether you would get the offer, and not whether you would enjoy the job. The interview step, only. That is the narrowest useful question and the only one the available data can speak to.

The five signals and what they weigh

• Skill fit — 50%. Overlap between your skills and the role’s, with required skills weighted more heavily than the broader nice-to-have list.
• Role fit — 15%. Whether the role’s level and function match what you actually do.
• Remote fit — 15%. Whether the role’s eligibility and time zone genuinely include you.
• Company fit — 15%. Company-side signals: size, stage, hiring behaviour.
• Language fit — 5%. Stated language requirements against yours.

Skill fit dominating at half the total is a deliberate bet: it is the signal most predictive of clearing a first screen, and the one where the underlying data is strongest. The weighted total is then adjusted — modestly — for how contested a role appears to be and how fresh the listing is. A role that has been open a long time and collected a lot of attention is a harder interview to win than the same role posted yesterday.

Four things it cannot see

• Referrals. A warm introduction beats every signal in the list and is invisible to us. If you know someone there, the score is understating your odds and you should ignore it.
• Whether the role is real. Some listings are open for pipeline-building rather than to fill a seat. We flag some of these separately, but the score itself does not know.
• How you interview. It reads the match between a profile and a posting, not how you perform once you are in the room.
• Your actual outcomes. This is the important one, below.

It has no outcome validation yet

The score has not been checked against whether people who scored highly actually got interviews, because that requires a base of users reporting outcomes back and we do not have one yet. Its weights come from reasoning about what predicts a screen, not from fitting to results.

That means it should be read as a considered ranking, not a probability. "This role is a better use of your evening than that one" is supported. "You have a 74% chance" is not, which is why we do not phrase it that way.

We would rather say this now than be asked later. When there is enough outcome data to validate the thing, we will publish what it says — including if it says the weights are wrong.