8.1 Shape Odd-One-Out

ADF YOU Session Job Opportunities Assessment

Track your progress as you read

Finished sections get a check mark.

1 Introduction & JOA Overview

1.1 What the JOA Is and Why the ADF Uses It 1.2 Current Exam Facts (51 Questions, 20 Minutes) 1.3 JOA vs YOU Session vs the Wider Process

2 Logistics, Access & Attempt Rules

2.1 Access, Devices & Browser Setup 2.2 Three Attempts & Six-Month Wait 2.3 After the JOA: JOR and Recruiter Contact

3 Official Tips & Test-Day Strategy

3.1 Official Top Tips & Natural Ability 3.2 Pacing, Flagging & Order of Attack 3.3 Integrity, Help Rules & Supervised Verification

4 Numerical Reasoning: Series & Matrices

4.1 Number Series Patterns 4.2 Number Matrices & Missing Values 4.3 Numerical Speed & Estimation Under Time

5 Numerical Reasoning: Word Problems

5.1 Rates, Distance & Averages 5.2 Ratios, Proportions & Percentage Reasoning 5.3 Multi-Step Numerical Reasoning

6 Verbal Reasoning

6.1 Word Meaning & Relationships 6.2 Verbal Analogies 6.3 Verbal Odd-One-Out 6.4 Verbal Logic & Multi-Statement Proofs

7 Abstract Reasoning: Sequences & Rules

7.1 Sequence & Next-Item Patterns 7.2 Rotation, Reflection & Orientation 7.3 Multi-Attribute Pattern Rules

8 Abstract Reasoning: Odd-One-Out & Complex Figures

8.1 Shape Odd-One-Out 8.2 Complex Figures & Spatial-Style Abstract Items 8.3 Abstract Speed Habits

9 Integrated Prep & Common Traps

9.1 Mixed Timed Practice 9.2 Common Traps Across All Item Types 9.3 Day-of Checklist

10 Results & Next Steps

10.1 Understanding the Job Opportunities Report 10.2 Pathway After the JOA

8.1 Shape Odd-One-Out

Key Takeaways Explain This Topic AI

Shape odd-one-out items ask which figure (or figures) do not share the rule that unifies the rest — not which drawing you subjectively like least.

Official-style stems may show several similar shapes (for example circles with different internal dots) and require spotting the count, fill, or orientation rule that fails for one option.

Always read how many answers are required: some JOA items ask for one odd figure; multi-select stems can ask for two that do not belong.

Name the shared rule in a short phrase before you click — “all have three dots except…” beats vague “this one looks different.”

If two candidate rules disagree on which figure is odd, keep testing attributes (count, shading, rotation, border) until one rule fits every remaining figure.

Last updated: August 2026

What shape odd-one-out measures on the JOA

Abstract reasoning on the Job Opportunities Assessment (ADF-RECREF132, revised 27 June 2025) includes odd-one-out figure sets. You are shown several abstract shapes or diagrams and asked which one (or which ones) do not belong . These items sit in the single 51-question / 20-minute paper with numerical and verbal reasoning. They are not drawing tests, art judgment, or school geometry proofs, and they are not part of any retired two-test mathematics curriculum.

What the ADF is sampling is practical: can you detect a shared visual rule , check it against every option, and isolate the figure that breaks it — quickly, under the shared clock? The skill is the same family as abstract sequences (classify → verify → answer), but the task is classification of a set , not “what comes next.”

Mental model: name the club, find the outsider

On every odd-one-out stem:

Scan all figures once without deciding — notice what varies (count of marks, shading, rotation, number of sides, open vs closed, internal vs external features).

Propose a short rule that could put most figures in one “club” (for example: “each figure has exactly three filled dots”).

Verify that every non-odd figure obeys that rule and that the odd figure(s) clearly break it.

Check the instruction for how many answers are required (one vs two, single-select vs multi-select).

Commit or flag — do not rebuild a baroque story after two failed rules when a simpler count rule was available.

If two rules both seem plausible but pick different outsiders, at least one rule is incomplete. Keep testing until one rule is consistent with the whole set.

Official-style example family: circles and dots

Familiarisation-style abstract items often use simple shapes with internal marks . A classic pattern family is a row of circles, each containing a different arrangement or count of dots .

Worked example A — one circle breaks the count rule

Imagine five circles labelled A–E:

• A: circle with 2 black dots inside
• B: circle with 3 black dots inside
• C: circle with 3 black dots inside
• D: circle with 3 black dots inside
• E: circle with 3 black dots inside

If the instruction is “which figure does not belong,” the unifying rule is “three internal dots.” A is the odd one out. The positions of the three dots in B–E may differ slightly; that is noise unless the stem is really about arrangement. Count is the first attribute to test because it is fast and binary.

Worked example B — count is equal; arrangement or colour differs

Now suppose every circle has exactly four dots , but:

• A–D: all four dots are filled black
• E: three filled black dots and one open (unfilled) circle mark

The odd-one-out is E on fill type , not count. Candidates who only count and then freeze will invent a weak position story. Always run a second attribute pass when counts match.

Worked example C — two attributes compete

Suppose five squares:

• A: square, one diagonal, shaded top-left triangle
• B: square, one diagonal, shaded top-left triangle
• C: square, one diagonal, shaded bottom-right triangle
• D: square, one diagonal, shaded top-left triangle
• E: square, no diagonal, fully unshaded

A shallow look might call C odd (different shaded region). A deeper look might call E odd (missing diagonal entirely). Which is correct depends on the tightest shared club . If A, B, C, D all share “square with one diagonal” and E lacks the diagonal, E is the cleaner outsider — C still has the structural feature; only its shade region differs. Prefer the rule that puts the largest consistent majority together and makes the outsider fail a clear feature, not a subtle style difference.

Teaching point: when two candidates look odd, ask which rule leaves a cleaner majority . Vague aesthetic difference loses to structural absence.

Attribute checklist (run in roughly this order)

Order Attribute Quick test

1 Count How many sides, dots, lines, segments, loops?

2 Fill / shading Solid, hatched, empty, half-shaded, which region?

3 Orientation Rotated, flipped, which way does a pointer face?

• 4 Open vs closed Gaps in the outline, broken sides
• 5 Internal structure Cross, diagonal, nested shape, grid
• 6 Symmetry Mirror left-right, rotational symmetry present or not
• 7 Relative position Mark is centre vs edge, above vs below a line

You will not need all seven on every item. Stop when one attribute yields a verified club-and-outsider story. Under JOA pacing, the first three rows of this table solve a large share of odd-one-out stems.

Worked example D — sides and openings

Five polygons:

• A–D: closed shapes (triangle, square, pentagon, hexagon — sides differ)
• E: a “C” shaped open curve that looks a bit like a broken circle

If side count varies among A–D, number of sides is not the club . The shared rule may be closed outline . E is odd. Do not force “all have even sides” when A is a triangle — that is first-pair style error applied to shapes.

Worked example E — rotation-only difference is not always odd

Four arrows all point up ; one points right . If everything else matches (same shaft, same head style, same fill), orientation is the rule and the right-pointing arrow is odd. But if the set is clearly “same shape at 0°, 90°, 180°, 270°,” then every orientation belongs and you must look for a different attribute (a missing arrowhead, a double shaft, a different colour). Sequence items reward rotation patterns; odd-one-out only cares whether orientation is the unifying property or the varying property. Decide which before you eliminate.

Multi-select and “find one or two that do not belong”

Official JOA guidance stresses reading how many answers are required . Abstract odd-one-out is a high-risk place for multi-answer mistakes because the set can contain one clear outsider or two that break different aspects of a dual rule.

• Single odd figure (most common teaching case)
• Instruction pattern: “Which figure does not belong?” or “Select the odd one out.”

Process: one rule, one outsider, verify the other four (or however many) all match.

• Two figures that do not belong
• Instruction pattern: “Which two figures do not belong?” or “Select all that do not fit the set.”

Process changes:

Find the strongest rule that unifies a majority subset.

Mark every figure that fails that rule — there may be two.

Confirm you did not invent a rule that only excludes one when the stem asks for two (or vice versa).

If the interface is multi-select, select both before submitting; partial selection is a wrong response even if one of your choices was correct.

Worked example F — dual outsiders on count and fill

Five figures, each a circle with dots:

• A: 2 filled dots
• B: 3 filled dots
• C: 3 filled dots
• D: 3 unfilled (outline only) dots
• E: 3 filled dots

If the stem says “which two do not belong,” a coherent dual-break story is: the club is “ exactly three filled dots .” Then A fails count and D fails fill — both are odd. If the stem only asked for one odd figure, the designer would usually make only one attribute break, or would clarify a primary rule. Always obey the count of answers required before you lock the story.

Worked example G — false multi-select trap

Four identical black triangles and one white triangle. Stem: “Which figure does not belong?” Only one answer is expected — the white triangle. Do not also deselect a black triangle because its rotation looks slightly different if the rotation was part of the intended variation. Over-selecting is as wrong as under-selecting.

Distinguishing odd-one-out from sequence / next-figure items

• Feature Odd-one-out Sequence / next item
• Goal Find figure(s) that break a set rule Find what continues a progression

Time order Usually unordered set Ordered frames 1 → 2 → 3 → ?

• Rotation May be noise or the club property Often the rule of change
• Answer type One or more of the shown figures Often a new option from a bank

Do not “complete the series” when the stem is odd-one-out. Completing a series invents a fifth figure; odd-one-out classifies the given set .

Common traps and defences

• Trap What it looks like Defence
• Aesthetic bias “This one is uglier / busier” Require a nameable attribute rule
• First-figure bias Rule built only from A vs B Must fit the full set
• Ignoring multi-select count Selecting one when two are required Re-read the stem’s “how many”
• Overfitting position noise Dot positions differ slightly in all Prefer count/fill before micro-geometry
• Two competing rules Different outsiders for each story Prefer cleaner majority + structural break
• Treating rotation as always odd Every frame is a turn of the same shape Check whether orientation is the club
• Rushing without naming the rule Click the “different looking” one Whisper a 5-word rule first
• Worked trap example

Five hexagons each contain a small triangle. In four of them the triangle points up ; in one it points down . The odd figure is the down-pointing triangle — orientation of the internal mark . A trap answer is the hexagon whose outer stroke is slightly thicker if line weight is rendering noise rather than a designed attribute. Stick to clear categorical features (up vs down, 2 vs 3, filled vs empty).

JOA process habits for shape odd-one-out

Budget: many count/fill odd-outs should fall in ~15–25 seconds . Multi-attribute or dual-outsider stems may need ~25–35 . Past ~40 with no named rule → flag (section 8.3).

Scratch lightly: a tiny tally of dots or a mental “3 / 3 / 3 / 2 / 3” is enough; do not redraw figures.

Response format: single-select or multi-select — confirm the interface matches the stem language.

Goal alignment: one verified odd figure is better than a perfect essay about three possible rules. The official aim is as many correct attempts as possible inside 20 minutes.

Mini drill set (self-check in prose)

Four circles with 5 dots; one circle with 4 dots → outsider is the 4-dot circle (count).

Five identical outlines; four fully shaded, one empty → outsider is the empty one (fill).

Stem asks for two that do not belong; club is “closed shape with a vertical line”; two figures lack the vertical line → select both line-less figures, not just one.

All five shapes are the same arrow rotated; one arrow is missing its head → outsider is the headless arrow (structure), not any particular rotation.

If any drill felt like “I just stared until one looked weird,” restart with the attribute table. Odd-one-out skill on the JOA is disciplined attribute testing , not gut aesthetic ranking.