How the Red Flag Test (“Is My Partner a Red Flag?”) is scored

This test scores behaviours you describe, not a person. It cannot know your partner; it only organises what you observe into eight dimensions, with rules you can read in full below. It never produces a diagnosis or a label.

What the test measures

The partner test uses the same eight dimensions and the same dimension weights as the self test, so both sides of a relationship are read with one ruler:

Each dimension has three scored questions (24 in total). Every question describes an observable behaviour — "gets upset when you make plans without them", "respects a limit the first time you say it" — rather than a trait or an intention. That wording is deliberate: you can report what happens far more reliably than why.

Severity weights inside a dimension

Not every behaviour in a dimension carries the same weight. Since version 2, six behaviours count 1.5 times inside their dimension, because they undermine trust, safety or respect even at low frequency; the other eighteen count once:

The score is therefore not a plain average of the answers: "often" on a high-severity behaviour moves a dimension more than "often" on a standard one, and jealousy or boundaries move the overall score more than empathy. The weights are public so the number can be understood, not just received.

From answers to points

Each question offers never (0), rarely (1), sometimes (2), often (3). Protective items ("does your partner apologise without a 'but'?") are reversed, so that a higher number always means "more of the pattern to watch".

Dimension scores and the signal index

For each dimension: score = 100 × (sum of weight × points) ÷ (sum of weight × 3), rounded to one decimal. The red flag score (the signal index) is the weighted average of the eight dimension scores, rounded to the nearest whole number:

index = 0.20 × boundaries + 0.15 × jealousy + 0.15 × accountability + 0.15 × communication + 0.10 × honesty + 0.10 × availability + 0.10 × conflict + 0.05 × empathy

Result levels

The thresholds are the same as the self test; the wording is more careful, because you are describing someone who is not here to answer:

• 0–24 — Mostly healthy signals
• 25–49 — Some zones to observe
• 50–74 — Several concerning behaviours
• 75–100 — Strong signals — consider support

Main and secondary signal

The main signal is the group of behaviours attached to your highest-scoring dimension; the secondary signal is the second-highest. If the index is 24 or lower and no dimension reaches 40, the result is "Mostly green signals" and the secondary signal is simply the highest dimension. Exact ties resolve in the order of the table above, so the same answers always give the same result.

The strengths, the things to watch, the action for today and the sentence to say are attached to the main signal. They are written for you — what you can observe, ask or say — never as instructions about the other person.

Safety screening

Five additional questions (financial control, isolation from friends or family, fear of reactions, pressure after a "no", threats or physical roughness) are not scored. Answering "sometimes" or "often" — or "rarely" for the last one — shows a calm notice with support resources, whatever the index says. A result in the "strong signals" level shows the same notice. Screening never changes the score.

When one of these answers is flagged, the result switches to its safety state: support resources and a quick-exit button come first, and the page no longer shows sharing prompts, a "say this to your partner" suggestion or the paid report — which the server also refuses to sell for that result. Fear, threats, control or violence are not communication problems to solve better.

How the result is read

On top of the scores, a fixed set of rules turns the eight dimensions into a reading. They never change a score, and no AI model takes part:

• A dimension at 25 or more (where the "watch" level starts) is an area that stands out; at 15 or less, a protective factor.
• Two dimensions that both reach 45 form a named pattern — for example jealousy & control with boundaries & autonomy ("checking, and a shrinking space"), or conflict management with accountability & repair ("fights that escalate and are not repaired"). Six such pairs are defined, plus two protective ones.
• The one-sentence reading depends on the shape of the result: nothing above 25, one area leading the next by 15 points or more, four or more areas at once, or a mix.
• The premium report adds, from the same rules: which answers drove the result (read from your own device, since we never receive your answers), what may ease with attention (watch level) versus what needs an explicit conversation (attention level and above), a 30-day plan built on your two highest areas, and questions to discuss.

What this result is not

It is not an assessment of your partner, and it is not evidence. It reflects your observations on one day, with all the bias that implies in either direction. Two people describing the same relationship can get different results. We deliberately never use the words "toxic", "narcissistic" or "abusive", and the test never tells you to stay or to leave. If what you describe worries you, the useful next step is a conversation with someone outside the relationship — a trusted person or a professional.

Determinism, versioning and testing

The scoring engine is a pure function: the same answers give the same result on any device. It is covered by automated regression fixtures (extremes, band boundaries, reversed items, ties, each main signal, safety flags). The test content is versioned; every result records the test version, the scoring version and the copy version it was computed with, and a premium report is always regenerated from that exact version.

• Version 1 (September 2026): eight weighted dimensions, every question counting once.
• Version 2 (19 September 2026): six high-severity behaviours weighted 1.5 inside their dimension. A version 1 result keeps its version 1 reading.

Intended use

A structured mirror for what you notice — to see which zones deserve a conversation, and to compare notes with the self test or answer the two-person compatibility test together. Not a clinically validated instrument, and not a substitute for talking to a professional.