How the score is actually calculated.

The practice test is 69 questions: 33 algebra and functions questions and 36 reading comprehension questions. That matches the published structure of the real IBEW aptitude test, which allows about 46 minutes for the math section and 51 minutes for reading.

You choose whether the clock runs. Self-Paced has no cutoff, because a timer measures speed and skill mixed together and a first sitting should isolate skill. You still see how long you took, compared against the real section limits, so speed shows up as information rather than as pressure. Exam Timed reproduces the real sitting: each section gets its own clock, the clock runs on real time whether or not the page is open, and a section does not reopen once it closes. The real test has a brief break between the two sections, so this does too, and reading's clock starts when you start reading.

Every question is original. The real exam's items are secured, so no one outside the test maker has them, and anyone selling you “real” questions is guessing. VLTG's questions are written to match the published structure and difficulty of the real battery.

Every math answer is verified programmatically before it is ever served, and correct answers are distributed evenly across A, B, C, and D so the answer pattern itself carries no information. A wrong answer key is the most damaging thing a practice test can do, because the person walks away believing something false about themselves.

The score is a stanine, a nine-point scale that describes where you land relative to other test takers rather than the percentage you got right. By definition a 5 is average, and the nine bands are designed to hold a fixed share of test takers: about 4, 7, 12, 17, 20, 17, 12, 7, and 4 percent from 1 to 9.

The conversion works in three steps:

1. Your raw correct count becomes a proportion of 69 questions.
2. That proportion is converted to a z-score against an assumed applicant distribution: a mean of 0.60 proportion-correct and a standard deviation of 0.18.
3. The z-score is placed into the standard stanine bands, which are 0.5 standard deviations wide and centered on the mean (cut points at plus and minus 0.25, 0.75, 1.25, and 1.75). The percentile shown alongside your score comes from the normal distribution function evaluated at the same z-score.

Those two constants, the assumed mean and spread, are the entire model. Everything else is standard method. Under the current model, the raw scores that produce each stanine are:

A 4 is the minimum most locals require to move on to an interview, which is why that row is highlighted. Locals rank applicants by score, so a higher number generally means an earlier call.

Every question is tagged to exactly one of 17 skills: 12 in math and 5 in reading. Your accuracy is computed per skill, and any skill below 80 percent is treated as a gap worth addressing.

Skipped questions count as wrong. They count against the composite score, so they count against the skill too. Otherwise a mostly-skipped test would report as strong on every skill while the score at the top said otherwise.

Questions the clock took away are different. In Exam Timed, a question you never reached still counts as wrong in the composite score, because it would have on the real test. It is left out of the per-skill percentages, though, because you never saw it: counting it would report whatever those unseen questions happened to cover as a weakness and send you to study a topic when your actual problem was pace. A skill whose every question went unreached is reported as having no data rather than as a zero. Deciding to skip a question you did see is a choice, and still counts against the skill.

Each skill also records what it builds on. That prerequisite structure is what lets the plan distinguish between a gap that stands alone and a gap that is being held down by something underneath it. If you are weak on Linear Equations and also weak on the Basic Arithmetic beneath it, practicing linear equations is the wrong first move.

Gaps are sorted into four categories, and the plan is ordered by leverage:

• Fix the root first. The skill is weak and so is something it depends on. Start underneath.
• Foundation. Nothing underneath it is broken, and other skills depend on it, so improving it lifts more than itself.
• Quick win. Close to the line already, so a small amount of targeted practice moves it.
• Stretch. Harder material worth picking up once the foundations are solid.

The time estimate attached to each skill scales with how much of it you missed and how hard that skill typically is to move, on a curve with diminishing returns rather than a flat hours-per-topic number.

The 1 to 9 score is an estimate. The official conversion tables belong to the test maker and are not published, so the model above stands in for them. It is built from the standard stanine method and a documented assumption about applicant performance. It will be re-normed against real VLTG response data as enough tests accumulate, and this page will change when it is.

A practice score is not a prediction. It tells you where you stand today against a reasonable model of the applicant pool. It does not tell you what you will score on test day, and no practice test can.

A timed score and an untimed score are not the same measurement. Both are reported on the same 1 to 9 scale against the same model, because there is not yet enough data to norm them separately. Treat a Self-Paced result as the ceiling of what you know and an Exam Timed result as closer to what test day would produce. When the two differ, the gap is pace, not knowledge.

The skill breakdown is the sturdier half. Whether you missed most of the fraction questions does not depend on any norming assumption. That part is a direct measurement, and it is the part the study plan is built on.

The skill taxonomy, the scoring and prioritization logic, and the structure of the learning data collected are published as a case study at github.com/bhavjotkhurana/vltg. The questions themselves are not published, for the obvious reason.