How the sub reliability score works

What the score is

Every sub on your bench carries a reliability score from 0 to 100, built entirely from the engagements you log against them: how good the work was, whether they finished on time, and whether you had to call them back. It is the number behind the ring on the sub's card, the ranked bars in the Reliability view, the keep, watch, or replace verdict in the analysis deck, and the review-or-replace decisions IQ raises. This article explains exactly how it is computed, so when a sub reads "At risk" you know precisely what that means and what would change it.

The three ingredients and their weights

The score blends three signals from your logged engagements, each with a fixed weight:

• Quality, weighted 40. Your 1 to 5 quality ratings, averaged across engagements and scaled to 100. A sub averaging 4 out of 5 contributes 80 on this dimension.
• On-time, weighted 30. The share of rated engagements the sub finished on time, as a percentage.
• Callbacks, weighted 30, inverted. The share of rated engagements that needed a callback or redo, subtracted from 100, so a sub called back on 20 percent of their work contributes 80 here. Callbacks count against the score because a callback is your crew's time and your client's patience spent fixing finished work.

Quality carries the most weight because it is the most direct judgment you make; the two schedule-and-rework signals split the rest evenly.

The fairness property: weights renormalize

You will rarely have logged all three signals for every sub, and the score does not punish that. The weights renormalize over the signals you have actually logged. If you have only ever rated a sub's quality, the score is built from quality alone, at full strength, not from quality plus two zeros. A sub with a single 5-out-of-5 rating scores 100, not 40. A sub with on-time and callback answers but no quality ratings is scored on those two, reweighted to cover the whole scale.

This matters because the alternative quietly penalizes the subs you know least about, which is backwards: a thin record should read as a thin record, not as a bad one. The number of engagements behind the score is always shown beside it, so you can judge how much weight to put on it yourself. One glowing engagement and twelve consistent ones can both read 90; the count tells you which is which.

The labels

The score maps to a label everywhere it appears:

• Reliable, at 75 and up, in green.
• Watch, from 50 to 74, in the caution tone.
• At risk, below 50, in red.
• Not rated yet, when there are no scored signals at all. No engagements means no score, not a zero.

These bands are what the rest of the section keys off. An at-risk sub becomes a replace candidate outright; a watch sub becomes one when there are real backcharge dollars behind them; and the section-level "Review or replace" tile names your at-risk sub with the most backcharges.

Backcharges are dollars, not score points

Backcharges are the fourth thing you can log, and they are deliberately not folded into the 0 to 100 score. They are totaled separately, in dollars, and shown as dollars: on the Performance tab, in the issues line, in the reliability view ("$2k backcharges" beside the sub's bar), and in the stake on a replace decision.

The reasoning: a backcharge already has a natural unit, money, and converting it into score points would bury the one number about a sub that is directly comparable to their invoices. A sub who scores 60 with zero backcharges and a sub who scores 60 having cost you $4,000 in backcharges are different problems, and the section keeps them legible as different problems. Backcharge dollars do, however, escalate the response to a given score: they are what turns a watch-level sub into a replace candidate, and they are counted into the dollars at stake on that decision.

Your score, not the network's

This score is yours: computed from your logged engagements, about work on your jobs, visible only to you. It moves the moment you log something and reflects nobody's judgment but your own.

That makes it deliberately different from the Verinode Score that some of these same companies carry in Benchmarks, which is built from the network's pooled experience and answers "how does this company perform across operators like you?" The two can disagree, and the disagreement is information: a sub the network rates well but who keeps missing your schedules is a fit problem on your jobs, not a bad company. When you are deciding who to put on tomorrow's loss, your own reliability read is the one that knows your jobs; see the Verinode Score for how the network-side number works.

Data sources