01
Dimension confidence
A score close to 50 indicates that the two sides received similar support. The assigned letter may be useful shorthand, but it should not be treated as a strong distinction. Greater distance from the midpoint supports a clearer preference within this response set.
02
Response-quality signals
The system also examines position bias, unusually long runs of the same displayed side, very fast completion, time spent per item and agreement across related prompts. These signals do not diagnose dishonesty. They only identify patterns that make a precise interpretation less secure.
Random left–right placement means repeatedly choosing one screen position is distinguishable from consistently endorsing one scored tendency.
03
Published response-quality rules
The browser supplies the presentation seed, response times and answer-change count; the server validates their ranges and applies the rules below from a starting score of 100, with a floor of 0. Scores 75–100 are good, 50–74 need review and scores below 50 are low confidence. These are interpretation heuristics, not a diagnosis of attention or honesty.
- Choosing one screen position 70%–79% of the time deducts 15 points; 80% or more deducts 35 points and adds a flag.
- A run of 12–19 choices on one side deducts 15 points; 20 or more deducts 30 points and adds a flag.
- Completion from 3 to under 6 minutes deducts 10 points; below 3 minutes deducts 25 points and adds a flag.
- When at least 75% of items have timing data, a median from 0.8 to under 1.4 seconds deducts 10 points; below 0.8 seconds deducts 25 points and adds a flag.
- Across at least 8 related-item pairs, agreement from 45% to under 60% deducts 10 points; below 45% deducts 20 points and adds a flag.
04
What confidence can and cannot do
Quality concerns can lower the confidence shown for a result but never change the scored type. A user can inspect midpoint dimensions, reconsider rushed answers or retake the assessment later. The system does not manufacture a more decisive profile when the evidence is mixed.
05
Interpreting a low-confidence result
Start with the dimensions furthest from the midpoint and test them against specific examples. Treat balanced dimensions as context-sensitive questions. If the session was interrupted, rushed or answered from an idealized self-image, retaking under calmer conditions may be more informative.
REF