A number next to someone's name is unusually persuasive. It is worth being precise about what one can actually support, because the gap between that and how people read it is where the damage happens.
What a record of confirmed outcomes can support
That something happened, and that someone other than the subject said so. This is the whole foundation. A completed shift confirmed by the manager on duty is a fact with two parties behind it. A self-reported one is a claim.
A rate, over a stated window. Turned up for 47 of 50 assigned shifts in the last six months is a defensible statement. It is also the least exciting sentence in this guide, which is roughly the right level of excitement.
A direction. Someone who missed twice a year ago and has been clean since is telling you something different from someone whose misses are last month. Recency is one of the few things a record genuinely carries.
What it cannot support, no matter how good the data is
Whether someone is good at the work. Delivery and quality are different properties. A record can show that a job was completed and accepted; it cannot tell you whether it was done well, and any score claiming otherwise is smuggling someone's opinion in through a number.
Why. A missed shift and a missed shift look identical in the data. One was a hospital admission and one was a better offer. If your decision depends on which, the record cannot make it for you, and treating the number as though it can is how you end up deactivating the wrong person.
What happens next. Past behavior is evidence about the future, not a prediction of it. This distinction sounds academic until you are the one being sorted by it.
Anything about a thin record. Three confirmed outcomes is not a track record, it is three data points. A system that hands someone a confident-looking number after three is not measuring reliability, it is measuring how recently they signed up. The honest treatment is to say the record is unproven and mean it.
The failure mode nobody plans for
Every scored system creates an incentive to produce the input rather than the outcome. If confirmations move the number, people will collect confirmations, and the cheapest way to do that is with someone who owes you a favor.
This is not hypothetical, it is the single most predictable thing about launching a score. Two questions are worth asking of any system before you rely on it:
Who confirms, and what do they have to lose? A confirmation from a party with no independent interest is barely evidence. The strongest ones come from someone who was going to be annoyed if the work had not happened.
Can the subject curate what you see? If they can hide the unflattering half, you are reading a highlight reel. That is fine as long as you know it, and misleading the moment it is presented as a complete picture.
If you are the one being scored
The useful move is unglamorous: get outcomes confirmed by the person on the other side, as they happen. A record built that way is worth something to the next employer precisely because you could not have written it yourself.
Be wary of anything that lets you raise the number without doing the work. If endorsements move it, so will the endorsements of people who barely know you, and a number that can be collected will be.
If you are the one reading a score
Treat it as one input that narrows a question, not as an answer. The reasonable use is deciding who to call first, or what to ask about. The unreasonable use is deciding who never hears back, because at that point a number that cannot explain itself is making a decision you are accountable for.
Ask what the score is computed from and whether that formula is published. If nobody can tell you, you are not evaluating evidence. You are trusting a vendor.