Improvement is a comparison.

That sounds obvious.

Yet families, teachers and tutors often begin training before they have a sufficiently clear picture of the starting state.

A child begins tuition.

Three weeks later, homework feels easier.

Did capability improve?

Possibly.

But perhaps the school topic changed.

Perhaps the learner slept better.

Perhaps the latest worksheet was easier.

Perhaps the tutor supplied more prompts.

Perhaps the learner genuinely improved.

Without a baseline, all of these explanations compete.

A training baseline is a defined starting sample of the learner’s current capability under known conditions, collected before or at the beginning of an intervention so later change can be interpreted against something real.

The baseline is not a label.

It is not “weak at Mathematics.”

It is evidence.

---

Quick Read: Baseline Before Intervention

Define the Target → Sample Current Performance → Record Conditions → Classify the Weak Link → Intervene → Retest Against the Baseline

A useful baseline answers:

• What can the learner already do?
• What fails?
• Under what conditions does it fail?
• How much help is required?
• How quickly can the learner respond?
• Does the skill survive a changed example?
• What error patterns recur?

This makes the later phrase “it is getting better” more precise.

A Baseline Is Different From a Grade

A grade compresses many performances into one number.

A baseline should preserve enough structure to guide training.

Mira scores 58% in Mathematics.

That number matters.

But it does not tell us whether the limiting problem is:

• algebraic fluency;
• method selection;
• representation;
• time pressure;
• careless execution;
• one missing prerequisite;
• or weak checking.

A useful baseline decomposes the score into trainable signals.

This connects to Training Signals.

Signals tell us what is changing.

The baseline tells us where those signals started.

Define the Conditions

A baseline is only meaningful if we know the conditions under which it was collected.

Was the task:

• timed or untimed?
• familiar or unfamiliar?
• with notes or without?
• with tutor prompts or independently?
• blocked by topic or mixed?
• completed immediately after teaching or after delay?

If Mira completes an untimed familiar worksheet with prompts, do not compare that directly with a timed unfamiliar test and announce deterioration.

The performance conditions changed.

Good baselines make conditions visible so later comparisons are fair.

Baseline the Capability, Not the Mood

One difficult day can distort a baseline.

The learner slept badly.

School ended late.

An argument happened at home.

Or the learner is unusually fresh and performs better than normal.

This is why a baseline should usually be treated as a small evidence window rather than one sacred score.

A 2025 article from the New Zealand Council for Educational Research on interpreting test scores stresses that day-to-day factors such as fatigue, anxiety, interruptions and guessing can create measurement error, making apparent rises and falls look more meaningful than they are.

The training response is simple:

Use enough baseline evidence to see the learner state, not merely the learner’s Tuesday.

Baseline Accuracy

Record whether the learner succeeds.

But do not stop there.

Ask:

• Which item types fail?
• Does failure cluster around one representation?
• Do errors recur in the same step?
• Does accuracy collapse when topics are mixed?

Accuracy provides the visible result.

Error structure tells us what to train.

Baseline Latency

Two learners can both be correct.

One answers in twenty seconds.

One needs three minutes and several restarts.

These are different baseline states.

Latency matters when the future performance will be timed or when slow lower-level operations consume attention needed for higher reasoning.

Record time when time is part of the capability.

Do not turn every lesson into a stopwatch contest.

Baseline Support Dependence

A learner can appear highly capable because the environment is supplying invisible support.

The tutor reminds.

The parent checks.

The worksheet groups every question by method.

The chapter title names the technique.

So baseline the amount of help required.

This connects to Training Cue Hierarchy.

If Mira needs a Level 4 partial step now and only a Level 1 orientation cue later, that is measurable improvement even before the grade moves.

Baseline Transfer

A learner may succeed on the exact trained surface and fail as soon as presentation changes.

Therefore a useful baseline should include at least one fresh or varied case when transfer matters.

Mira can solve ratio questions about recipes.

Can she recognise the same relationship in map scale?

Jonas can answer inference questions in narrative passages.

Can he carry the same evidence rule into an article?

Nadia can identify variables in familiar plant experiments.

Can she do so in an unfamiliar electrical setup?

The transfer sample protects against an inflated baseline built only from familiar rehearsal.

Mathematics Baseline: Mira

Mira begins a new training cycle with weak recent examination results.

The tutor does not begin with ten chapters of revision.

A baseline set samples:

• algebraic manipulation;
• equation solving;
• representation switching;
• method selection;
• one timed item;
• one unfamiliar transfer item.

Mira is accurate in algebra when the method is named.

She becomes slow and error-prone when topics are mixed.

The baseline changes the plan.

The main target is not “more algebra.”

It is recognition and method selection under mixed conditions.

English Baseline: Jonas

Jonas receives one reading sample, one vocabulary sample and one short writing task.

His literal comprehension is strong.

His vocabulary is adequate.

His inference answers overclaim.

His writing has good grammar but weak idea development.

That baseline is much more useful than “English: 67%.”

Two different training jobs are now visible.

Science Baseline: Nadia

Nadia knows factual Science well.

The tutor samples:

• concept retrieval;
• diagram interpretation;
• variable reasoning;
• evidence-based explanation;
• one unfamiliar application.

Concept retrieval is strong.

Evidence integration is inconsistent.

The baseline prevents the tutor from wasting time reteaching content Nadia already knows.

Baselines and Training Priorities

A baseline reveals many weaknesses.

Not all should be trained at once.

Use Training Priorities.

Which weakness:

• appears frequently?
• costs many marks or much time?
• blocks several later skills?
• can be repaired efficiently?
• is likely to matter in the next school stage?

The baseline maps the terrain.

Prioritisation chooses the route.

Baselines and Training Dependencies

A baseline can expose a lower-level dependency that is limiting higher performance.

This links directly to Training Dependencies.

If Mira’s geometry failures disappear when algebra is supplied, algebra may be the dependency.

If Jonas’s inference improves when one difficult word is defined, vocabulary may be the dependency.

The baseline becomes more powerful when it is designed to separate such hypotheses.

Baselines and Training Review

Training Review asks whether the plan still fits the learner.

The baseline gives review a reference point.

What changed?

What stayed the same?

Which weakness disappeared?

Which new constraint emerged?

Without baseline evidence, review can become storytelling after the fact.

Do Not Baseline Everything

A baseline is useful only if it changes a decision.

Do not spend three sessions measuring every possible skill before teaching begins.

Sample enough to identify the likely first weak link.

Then train.

Collect more baseline detail only where uncertainty remains decision-relevant.

Do Not Turn a Baseline Into an Identity

“You are weak at Science.”

That is not a baseline.

“On three unfamiliar experiment questions, you identified the variables correctly but your conclusions exceeded the evidence.”

That is actionable.

Baselines should describe performance states that can change.

They should not become permanent descriptions of the child.

Do Not Let Practice Contaminate the Baseline

If the tutor teaches the method during the baseline task and then records the final answer as the starting state, the measurement has changed the thing being measured.

Sometimes intervention during diagnosis is necessary and humane.

Just record it honestly.

“Independent attempt failed; Level 3 cue restored the route.”

That is better evidence than pretending the final supported success was independent.

Measurement Quality Matters

A 2025 systematic review in School Effectiveness and School Improvement examined the measurement quality of student outcomes in teaching-effectiveness research and found that theoretical grounding, reliability and validity evidence were often inadequately reported. The research context is broader than individual tutoring, but the lesson travels well: conclusions about improvement are only as interpretable as the measures used to support them.

For a tutor, that does not require psychometric software.

It requires discipline.

• sample the right capability;
• record the conditions;
• avoid over-reading one item;
• use comparable later tasks;
• separate supported from independent performance;
• look for patterns across evidence.

A Parent Baseline Audit

• What exactly are we trying to improve?
• What evidence shows the starting state?
• Was the baseline independent or supported?
• Were the tasks representative?
• Were the conditions similar enough to compare later?
• Are we relying on one score or a small pattern?
• Can we describe the baseline without labelling the child?

A Tutor Baseline Audit

• What capability am I measuring?
• Which tasks sample it?
• What conditions should be recorded?
• How much cueing occurred?
• What error classes appeared?
• Do I have enough evidence to identify a first priority?
• What future task will be comparable enough to judge change?

The Deeper Idea: Progress Needs a Before

Families want to know whether training is working.

That is a reasonable question.

But improvement should not be inferred from hope, workload or one recent score.

Start with a sufficiently clear before.

Then change the system.

Then look again.

A baseline does not predict what the learner will become. It gives us a fair place from which to observe whether the training is changing what matters.

Research Foundations

Useful anchors include the 2025 systematic review of student-outcome measurement quality, the 2025 New Zealand Council for Educational Research article on measurement error and student test-score interpretation, and the wider educational-measurement literature on reliability and validity. The practical claim here is deliberately modest: a baseline should sample the intended capability under recorded conditions well enough that later change can be interpreted without pretending any one score is a perfect measurement of the learner.

Continue Through How Training Works

Read this alongside Training Signals, Training Priorities, Training Review, Training Branching, Training Dependencies and Training Cue Hierarchy.