How Standards Proficiency Is Calculated

Every colored square, ring and bar in Forefront comes from the same question: given everything we know about this student, how proficient are they on this standard? Forefront answers it in one of two ways, and which one your district uses is a setting rather than something you choose per report.

This article walks through both, using one student’s standards wheel as the running example, and is deliberately honest about the places where each answer can look wrong at first glance.

If you just want to know which one you’re in, skip to Which one is my district using?.

Neither algorithm looks at raw points. Both start by turning every score into a proficiency band, and both then do the same two things with dates.

Questions carry the alignment, not assessments. A question is aligned to one or more standards — often several frameworks at once, so the same interview task feeds a Number Sense progression, a state standard and a Common Core standard. Only aligned questions say anything about a standard.

Each question is one equal vote. A question’s score is converted to a band using that question’s own rubric ranges, and from then on only the band matters. A 12-point task and a 1-point task count the same, and a student who tops the band counts the same as one who barely reached it.

Bands are numbers 0–4 while the math happens.

Averages are rounded up at the halfway point when they’re turned back into a color: 2.5 and above is Proficient, 1.5 up to 2.5 is Basic, and so on.

More recent evidence counts for more. When a standard has been assessed more than once, the values are combined with a weighted average rather than a plain one. The weight falls off with the gap between an assessment and the most recent evidence in view:

Two things about that table surprise people. The gaps are measured against the newest assessment in view, not against today, so a body of evidence doesn’t quietly decay over the summer — it re-weights when new evidence arrives. And the fall-off is steep: with a fall screener and a midyear screener just over five weeks apart, midyear carries about 91% of the answer and fall about 9%.

You never have to take that on faith. Click any node of a standards wheel and the body of evidence lists each assessment with the Weight it was given.

For one assessment, a standard’s value is the average of the bands of every question aligned to that standard or to anything beneath it. That value is rounded to a band and stored. Across assessments, those per-assessment bands are combined with the recency weighting above.

So a domain is, in effect, a straight average of all its questions, and the deeper structure of the standards tree doesn’t change the arithmetic — only which questions are counted.

Place Value: Below Basic

The Number Sense progression for one second grader. The inner ring holds domains, the middle ring clusters, the outer ring individual standards — and grey means nothing has assessed that standard yet.

A broad, recent assessment can move a whole domain by itself. Because a domain counts every question beneath it, a screener that touches twelve standards writes a fresh value for every ancestor of every one of them. That new value is the most recent evidence, so it carries almost all of the weight, and a domain that a class has been working on all year can be repainted by a single morning’s data. This is the single most common complaint about the default algorithm, and it’s the reason the roll-up mode exists.

A domain reports on whatever it has the most questions about. Question counts are rarely even. If a domain has six questions about one sub-skill and one question each about two others, three-quarters of the domain is a report on that first sub-skill. Nothing is broken — but a domain reading Basic can mean “fine at the two things we barely asked about, weak at the thing we asked about six times.”

Values are rounded twice. Each assessment’s value for a standard is rounded to a band before the recency average runs, so a student at 1.49 and a student at 0.51 both enter the time average as Below Basic. Two students with visibly different papers can come out identical.

A standard with no aligned question stays grey, and stays grey no matter how much evidence its parent or its siblings have. “Not Assessed” always means literally that: no question in view is aligned to that standard.

Roll-up mode inverts the order of operations. Each standard is scored only from the questions aligned directly to it, using the same banding and the same recency weighting. Then those scores are averaged up the tree: a standard’s value is the average of its direct sub-standards’ values, together with its own score if it has questions of its own — which sit alongside the sub-standards rather than above them, for the reason in Standards aligned above the leaves.

Three consequences follow from that, and they’re the whole point of the mode:

Place Value: Basic

The same student, the same scores, the same lens — read by the roll-up algorithm. Place Value and its Computation, Estimation and Problem Solving cluster both come out a band higher.

A domain can read higher — or lower — than the questions under it “feel”. Giving each branch equal say is exactly what makes a domain robust against an uneven question count, and it also means the domain is no longer the average of the question list you’re looking at. The worked example below is one of these.

One question can outweigh five. A sub-standard assessed by a single question counts as much as a sibling branch assessed by five — that happens in the worked example below. It is the intent, since the two sub-skills matter equally, but it does mean one unlucky task can hold a domain down.

Filling a gap can move a parent a lot. Because empty branches are skipped, a sub-standard’s first ever score doesn’t just fill in one grey wedge; it becomes a full equal voice in its parent. Expect a domain to shift the first time a previously unassessed sub-skill gets assessed.

The Weight column can look like it contradicts the answer. Recency is applied per standard, before anything is averaged up, so an older assessment is only competing with newer evidence about the same standard. If a sub-standard was assessed in the fall and never since, the fall screener is 100% of that sub-standard’s score — and that score is then a full equal voice in its parent, no matter how far back it was. So a parent can read Basic while the panel above it shows a Below Basic assessment at 91% and a Basic one at 9%. Those percentages describe how the two assessments weigh against each other within one standard; they are not the recipe for a parent’s number.

Not a bug: Place Value is Basic even though the midyear screener (91% of the weight) reads Below Basic. Ones, Tens, and Hundreds was only ever assessed in the fall, so within that sub-standard the fall screener carries everything — and the sub-standard carries a third of the domain.

A cluster’s own questions are a sibling of its children, not a summary of them — and they lend nothing to those children. This one is worth its own section — see Standards aligned above the leaves.

Assessments that compute their own standards don’t roll up. A few assessments (DIBELS-style ones, for example) come with a bespoke calculation that authors a value at every level it cares about, including cluster and domain levels. Those values are read exactly as the assessment authored them; rolling them up would average an authored domain against the children it was derived from. So in a roll-up district a DIBELS composite still behaves like a DIBELS composite.

Here is the Place Value domain for the student in the screenshots above. Eight questions across the fall and midyear screeners; three sub-standards, one of which has three sub-standards of its own.

(The counts add to nine because one midyear task is aligned to two sub-standards at once — Mental Math and Solve Written Problems — so it is evidence for both.)

Question Average counts the questions. The midyear screener’s six Place Value questions banded as Basic, Below Basic, Below Basic, Basic, Well Below Basic and Below Basic, which averages to 1.17 — Below Basic. The fall screener’s two came out Basic and Basic, so fall’s value is Basic. Midyear is just over five weeks newer, which earns it 91% of the weight against fall’s 9%, and the domain comes out at 1.09 — Below Basic.

Roll-up works up the tree instead. Computation, Estimation and Problem Solving averages its three sub-standards: (2.0 + 2.0 + 1.0) ÷ 3 = 1.67. Place Value then averages its three sub-standards: (1.67 + 1.0 + 2.0) ÷ 3 = 1.56 — Basic.

Neither number is a bug. The default answer says most of what we asked this student about place value, they got wrong. The roll-up answer says of the three things place value is made of, they’re shaky on one and basic on two. They disagree because the questions aren’t spread evenly over the sub-skills — and that is the situation the two algorithms are built to read differently.

In a roll-up district the body of evidence shows you this directly: expanding an assessment groups its questions by the sub-standard they belong to, with each group’s score beside it, so a domain’s number can be traced to the groups it averaged rather than to a flat list of questions.

The default algorithm: one flat list of every question that speaks to Place Value, each one an equal vote. Note the Weight column — the midyear screener carries most of the answer.

The same evidence in a roll-up district. The questions are grouped by the sub-standard they belong to, and each group header carries the score that went into the average — the arithmetic the domain actually did.

Assessments don’t only align questions to the bottom of the tree. A question can be aligned to a cluster or a domain — the standards wheel above has one: a midyear task aligned to Solve, which is itself the parent of One Step Word Problems and Comparisons.

Read that alignment as a statement about coverage: this question speaks to Solve, and specifically to the part of Solve that the standards beneath it don’t already cover. A cluster stands for a set of ideas; its children carve named pieces out of that set; a question aligned to the cluster itself is evidence about the remainder — the complement of the children within the parent’s meaning.

It is not a shortcut for “this question covers everything under Solve.” If that is what you mean, align the question to each child instead — that alignment does reach them, because you said so explicitly.

Everything else follows from that reading, and it holds in both algorithms:

One thing to watch: the complement is only as real as the tree. If a cluster’s children were meant to name everything in it, a question aligned to the cluster has no leftover ideas of its own to speak to — but it will still be averaged as though it did. That’s usually a sign the question wants a more specific alignment, or the cluster wants another child.

Roll-up is enabled per district by Forefront, and there is no district-level setting for it, so the quickest answer is to ask your Forefront contact. Two things in the app will also tell you:

Switching a district over rebuilds its stored proficiency, so numbers change on reports that nobody edited — worth knowing before a data meeting, and worth scheduling deliberately rather than mid-conference-week.