MindOS Learning Manual: Rule-Induction State | Can You Discover the Rule Without Being Told the Rule?

Wait, What?

Sometimes being told the rule first can hide whether you can see the rule at all.

A student may apply a formula accurately after the teacher names it, yet fail to recognise the same relationship when the formula is not announced.

Rule induction asks for a different operation: inspect several cases, notice what changes, notice what stays invariant, propose the underlying relationship, and then test whether the proposed rule survives a new case.

Quick Answer

Rule-Induction State is useful when the learner needs to construct or recognise an underlying regularity from examples rather than merely retrieve a rule that has already been taught.

It can support abstraction and generalisation, especially when the examples vary in ways that expose what is structurally important. But unguided induction can also waste effort or reinforce a false pattern. The operation therefore needs carefully chosen examples, feedback and a clear point at which discovery gives way to explicit consolidation.

Owned Learning Operation

RULE-INDUCTION STATE = inspect examples → compare variation → identify invariant relation → propose rule → test against counterexample/new case → refine → state rule explicitly.

This is different from Example-Generation State, where the learner already knows a concept and invents a new instance. It is different from Comparison State, which may stop at similarities and differences. Rule induction requires the learner to infer a general relationship that explains the cases.

The Invariant Is the Target

Consider these Mathematics examples:

2(x + 3) = 2x + 6
4(x + 5) = 4x + 20
3(x + 1) = 3x + 3

The numbers change. The brackets change. The surface is variable.

The invariant is that multiplication outside the bracket applies to every term inside it.

A learner who sees only “multiply the first number” has induced the wrong rule. A learner who says “the outside factor multiplies every term inside” has extracted the structure.

Why Variation Matters

If every example looks nearly identical, the learner may succeed by memorising a surface pattern. More varied examples can make initial learning harder while making the shared structure easier to generalise later.

But more variability is not automatically better. Variability should change features that are not essential while preserving the target relationship. If too many dimensions vary before the learner has enough prior knowledge, the task can become noise.

The MindOS Rule-Induction Protocol

Step 1 — Present at Least Two Cases That Share the Target Structure

The learner needs enough information to compare. One example can be remembered; several examples make an invariant possible to infer.

Step 2 — Ask What Changes

Names, numbers, contexts, shapes, wording or representations may change. Making the changing dimensions explicit helps stop the learner from treating them as the rule.

Step 3 — Ask What Stays Relationally the Same

Do not accept “they are all about fractions” or “they all have plants.” Ask for the relationship: what operation, condition, causal pattern or logical structure is preserved?

Step 4 — State a Candidate Rule

The rule must be clear enough to be wrong. Vague guesses cannot be tested.

Step 5 — Test a New Case

Give a case with different surface features. If the rule predicts it correctly, confidence rises. If it fails, revise the rule.

Step 6 — Test a Near Counterexample

Choose something that resembles the examples but violates one load-bearing condition. This prevents the learner from turning a rough pattern into an overgeneralised law.

Step 7 — Consolidate Explicitly

Once the learner has induced the rule correctly, name it, explain it, retrieve it later and practise applying it. Discovery is not the end of learning.

Worked Examples Across Subjects

English: show several passages where a character’s motive must be inferred from evidence. Ask what makes these questions inference rather than retrieval questions. The learner should induce that the answer requires combining textual evidence with reasoning beyond the literal sentence.

Mathematics: compare several percentage-change problems with different quantities and contexts. Ask what the denominator represents in each. The learner should induce that change is measured relative to the original quantity, not simply “divide by whichever number is smaller”.

Science: show several fair-test investigations. Change the apparatus and topic while preserving the structure: one variable deliberately changed, one outcome measured, relevant others controlled. Ask the learner to infer the experimental rule.

Competing Explanations When Induction Fails

The learner may not have enough prior knowledge to interpret the examples.
The examples may not vary in informative ways.
The learner may focus on the most salient surface feature.
The target rule may be too complex for unguided discovery.
The learner may infer a plausible but incomplete rule.
Feedback may arrive too late to correct the emerging pattern.
The task may require direct explanation rather than induction.

MindOS does not romanticise discovery. The question is whether induction is doing useful cognitive work for this learner at this point.

How Do We Know?

A 2022 review in Trends in Cognitive Sciences synthesised evidence across categorisation, language, motor learning and other domains showing that more variable learning input often slows initial acquisition but can support broader generalisation. The authors also emphasised that different kinds of variability matter differently.

A 2026 study in Educational Psychology Review examined transfer learning with varied worked examples and retrieval practice. Its instructional discussion explicitly distinguishes two goals: inducing an underlying rule and strengthening a rule already understood. In the varied conditions, learners could compare instances and identify structural regularities; once a rule was established, retrieval became a stronger consolidation operation.

Evidence Boundary

Evidence about variability and induction spans many domains and experimental tasks. It does not justify replacing explicit teaching with unguided discovery. Novices often benefit from direct guidance and worked examples, especially when the underlying structure is difficult to infer. Rule induction is strongest when examples have been deliberately chosen to expose an invariant and the learner receives feedback before a false rule stabilises.

When Rule Induction Is the Wrong Tool

When prerequisite knowledge is missing.
When the domain contains a convention that cannot reasonably be discovered from examples alone.
When the learner is already overloaded by the examples.
When a misconception is becoming more entrenched with each guess.
When the rule is already understood and the real job is retrieval, automaticity or transfer.

The Independence Test

Can the learner state the rule without the examples?
Can the learner explain which features were irrelevant?
Can the learner identify a counterexample?
Can the learner apply the rule to a new surface form?
Can the learner distinguish this rule from a neighbouring one?

If the learner needs the original examples to remember the rule, cue dependence may remain. If the rule is remembered but applied to everything, concept boundaries need repair.

Teaching Guide for Parents, Tutors and Teachers

Do not ask only “What is the rule?” Ask the learner to show where the rule came from.

“What changes across these examples?”
“What stays the same?”
“What rule would explain all three?”
“What example would prove your rule is too broad?”
“Can you apply the rule to a case that looks different?”

Then stop discovering. Once the learner owns the rule, move into retrieval, practice and transfer rather than repeatedly making them reinvent known knowledge.

MindOS Direction

If examples look similar but the learner cannot identify the invariant, use Comparison State. If the learner has induced a rule but cannot decide what belongs, use Concept-Boundary State. If the rule is understood but not retained, use Retrieval State and Spacing. If the learner can state the rule but fails on changed tasks, use Transfer State.

MindOS rule: induction succeeds when the learner can discover what stays true across changing examples, state it explicitly, and survive a new case that was never part of the original pattern.