How the Enneavine test works
We're building the most accurate Enneagram test we can. This page shows how it works, what we know so far, how we'll measure it, and what we don't know yet. When we have real accuracy numbers, they'll be published here, including the ones that fall short.
Updated October 21, 2026 · Statement bank v0.2 · Scoring engine web-0.2
Core type, wing, instincts, subtype, then two short checks where you decide
Typical length; about 12–15 minutes. The test stops early when it's confident
270 for core type (30 per type), 80 for instincts, 162 for the 27 subtypes
Real-world accuracy. Our goal is at least 90% agreement with expert interviews
Why most Enneagram tests struggle
The Enneagram describes why people do things, and that's hard to measure. Common problems:
- Behavior instead of motivation. Two people can keep a tidy desk for opposite reasons: one because mistakes feel wrong, another because order keeps them safe. A question about the desk can't tell them apart.
- Rating scales reward looking good. When you rate each statement from agree to disagree, it's easy to agree with whatever sounds most admirable.
- Lookalike types. Some types look alike on the surface (for example, Twos and Nines can both put others first, and Sixes and Nines can both hesitate) but for different reasons.
The published research shows the cost. In studies reviewed by Sutton (2012), two established Enneagram tests gave the same type only 42% of the time (Dameyer, 2001), and a paragraph-choice test matched people's final type 37% to 68% of the time, depending on the type (Daniels & Price, 2000; chance is 11%).
How a statement is written
Every statement in the test follows the same rules:
- About motivation. What you want, what you fear, where your attention goes, what you avoid, and how you see yourself.
- One idea, in plain words. 20 words or fewer, at about an 8th-grade reading level, with no double negatives and no idioms. In October 2026 we reworded 164 statements and portrait sentences that people could read more than one way.
- Lifelong. You're asked how you've been for most of your life, since stress or a hard season can make anyone look like another type.
- Original. We never copy or adapt other tests or books. Every statement is checked against our reference library for borrowed wording.
- No faith language. The test must work the same for anyone, and it mustn't mistake devotion for type. Faith belongs in your results and growth path, not in the questions.
- Matched for appeal. Each statement is rated for how appealing it sounds (1 to 7), and the three statements in a set are matched as closely as the bank allows (the aim is within half a point), so you choose by fit rather than by what sounds nicest. These ratings are provisional for now: they come from a panel of AI raters, and a study with human raters will replace them.
The four parts
You see three statements at a time. You pick the one most like you, then the one of the remaining two that's more like you, which ranks all three.
| Part | What it finds | Sets |
|---|---|---|
| 1. What drives you | Your core type. Each set puts statements from three different types side by side, often lookalikes. | Until the test is 90% sure, or 40 sets at most (about 24 in simulations) |
| 2. Your wing | Which neighboring type flavors yours more. Sets compare the two neighbors of your type. | 3 to 8 |
| 3. Where your attention goes | Your instinct order. Each set has one self-preservation, one social and one one-to-one statement. | 6 to 12 |
| 4. Your version of your type | Your subtype, from statements written for your type's three subtypes. | 3 to 6 |
| Last step: you decide | You read three short descriptions of your most likely types, with the names hidden, and pick the one that fits. Then you choose between two short descriptions of your possible wings. | 2 choices |
If you already know your type, the short version skips part 1 (usually 15–25 sets).
How scoring works
- A running probability for each type. Every answer updates how likely each of the nine types is, given everything you've answered.
- The next set is chosen for you. The test picks the set that best separates your leading possibilities, so it spends its questions where you're hardest to tell apart.
- It stops when it's confident. Part 1 ends when one type reaches 90%, or after 40 sets. The later parts stop once they reach 85% (after a minimum number of sets), or at their maximum.
- Your own judgment counts. If the description you pick at the end isn't your top type, the result says so and softens its wording. Your wing choice counts as extra evidence too.
- Instincts and subtype agree. Your subtype includes your dominant instinct, so if the two parts disagree, the more confident one wins and the result notes the change.
For the technically curious: answers are modeled with a Thurstonian (probit) model of ranked choices. Type probabilities are estimated with a particle method and checked with the GHK simulator, so the stated confidence matches accuracy in simulation.
How results are worded
| The test is… | Your result says |
|---|---|
| 80% or more sure of one type | “You're likely a Type X” |
| 55% to 80% sure | “Most likely Type X, and also consider Type Y,” with help deciding between them |
| Less than 55% sure | “Somewhere between Type X and Type Y” |
The bars on your result compare the types with each other. They aren't percentages yet, because the scoring rests on our best starting estimates until real results let us fine-tune it.
What we know so far
From simulations, not real people
Before anyone took the test, we ran it on simulated people who answer according to the scoring model, using the real statement bank and assumed settings. Simulations show whether the design works. They aren't accuracy claims: real people are less consistent, and the settings are guesses until real results come in.
| Measure (simulated people) | Result |
|---|---|
| Core type right (top pick) | 78% (180 people, 20 per type) |
| Right type in the top two | 89% |
| “Likely” results that were right | 89% (109 of 123) |
| Stated confidence vs. actual accuracy | 80% vs. 78%: the test is about as sure as it should be |
| Sets used in part 1 | About 24 on average |
| Dominant instinct right | 69% (300 people); full instinct order 44% (chance is 17%) |
| Subtype right, when the core type was right | 75% to 91%, depending on how strong the subtype statements turn out to be |
Two things stand out. Two in three simulated people got a “likely” result, and those were right about 9 times in 10; the rest got an honest “also consider” or “between”. And instincts are the hardest part to measure, so we show your instinct order as tentative.
From published research on the Enneagram
- The evidence is mixed. The largest review (Hook and colleagues, 2021; 104 samples) found some support (types relate sensibly to the Big Five personality traits, and several studies found it helped personal and spiritual growth) and some weaknesses (studies usually find fewer than nine distinct factors).
- Wings and the stress and growth lines have little research support, and subtypes and instincts are largely untested. We teach them as traditional teaching, gently.
- Benchmarks we aim to beat (from Sutton, 2012): a leading test gave 82% of people the same type when they retook it (Dameyer, 2001), and a general trait questionnaire placed 75% of people in their correct type (Brown & Bartram, 2005).
How we'll measure real accuracy
A test is only as accurate as what it's measured against. Ours will be measured against careful expert interviews, which identify a person's type independently of the test:
- A recorded 60–75 minute interview about motivation across life: childhood, work, relationships, conflict, stress, and what you avoid.
- Two people experienced at identifying Enneagram types each type the person from the recording, on their own, without seeing the test result.
- Disagreements go to a third reviewer. The participant then says whether the agreed type fits.
- We publish how often the two raters agree, alongside the test's accuracy, every time.
| Study | Purpose | People | Status |
|---|---|---|---|
| Appeal ratings | Rate how appealing each statement sounds, for matching | 150–200 | Forms ready |
| Think-aloud interviews | Check that people read statements as intended | 20–30 | Planned |
| Pilot | Measure how well each statement works; drop weak ones | 1,000+ | Planned |
| Retest | Do people get the same type 2–4 weeks later? | 200+ | Planned |
| Expert-interview validation | Accuracy, overall and type by type | About 640 (at least 71 per type) | Planned |
| Fairness | Statements work the same across gender and age | 200+ per group | Planned |
| Ongoing | “Does this fit?” ratings, opt-in research answers and anonymous result counts | Everyone who takes part | Running |
| Measure | Our goal | Minimum before we call it accurate |
|---|---|---|
| Core type matches the expert interview | 90% or more | 80%, and above the best published benchmark |
| Right type in the top three | 98% | 95% |
| Lowest single type | 80% | 65% (and reported) |
| Same type on retest | 85% | 80% |
| Stated confidence vs. actual accuracy | Within 5 points | Within 10 points |
| Typical time to finish | 20 minutes or less | 25 minutes |
Safeguards, because we built the test ourselves:
- We'll publish the validation plan before collecting the data, so the goalposts can't move.
- People in the validation study are never used to tune the test.
- An outside expert reviews the analysis before we make any public accuracy claim.
- The interview raters don't write test statements and never see test results.
- We'll report every measure above, including any we miss, with sample sizes.
What we don't claim
- No accuracy figure yet. We won't quote one until the expert-interview study measures it.
- Not a diagnosis. The Enneagram describes personality patterns. It says nothing about mental health.
- Not the last word on who you are. A result is a starting point for reflection. Your identity is in Christ, not in a number.
Changes
- October 21, 2026 · bank v0.2: 164 statements and portrait sentences reworded for clarity, measuring the same things. Anonymous result counts began (see below).
- October 18, 2026 · bank v0.1: the test launched.
Your privacy
Your answers stay on your device unless you choose to share them for research. When you finish, we keep an anonymous record of the result (type, wing and instincts) to see how results are spread across the types, and you can opt out. Details are in our privacy policy.