How BDSM Prism scoring works

What this method is designed to do

The method organizes current responses to structured prompts across themes such as direction, restraint, intensity, care, visibility, novelty, and privacy. It does not determine identity, experience, morality, compatibility, consent, or safety. There is no single official BDSM test or universally accepted BDSM personality scale that certifies this project.

Read the result as a snapshot. Ask which themes appeared, which feel accurate, and what conditions would change an answer. A result can support conversation, but every activity still requires separate, informed, freely given, ongoing, and withdrawable consent.

The nine dimensions

The nine dimensions are separate signals rather than one ladder from “less BDSM” to “more BDSM.” High values on apparently different dimensions can coexist because context, partner, activity, and role may change what a person prefers. The table describes the intended reading boundary of each published dimension; it does not add claims beyond the current scoring model.

Answer normalization, skipped items, and attention checks

Each scored prompt uses an integer from 0 to 4, with 2 as midpoint. The implementation applies (answer − 2) ÷ 2, producing −1, −0.5, 0, 0.5, or 1; reverse-keyed items change the sign. The signal is multiplied by item mappings and accumulated in dimension, candidate-archetype, and meta channels. This page does not reproduce a question-by-question weight table; the mappings are bundled with the client-side question bank and are not secret.

A skipped, missing, or invalid answer contributes zero to raw signals. The system does not impute a preference or replace the answer with an average. Skipping is preferable to forcing a false response, but extensive skipping reduces coverage. The result includes an answered-ratio quality field. Percentiles still compare those raw signals with a baseline simulated from complete answers to all scored prompts, so extensive skipping also limits how those percentiles can be interpreted.

The quick and full banks each include two attention checks. They are excluded from every score. Missing an expected response sets a low-attention note; it does not erase the result or secretly change a dimension. The quality layer also detects an all-midpoint pattern and limited response variation. Straight-lining is flagged only when at least eight scored responses are available and either response variance is below 0.15 or one repeated-value run covers at least 90 percent of answered scored items. These are caution flags, not proof of bad faith.

How responses become candidate archetypes

1. Normalize each valid answer and reverse its sign when the item is reverse-keyed.
2. Accumulate weighted signals in parallel. The nine displayed dimensions and the 18 candidate-archetype channels receive their own mappings from item responses.
3. Compare each raw channel with the mean and standard deviation in the baseline for the selected test version. The implementation converts the resulting standardized value through the normal cumulative distribution and rounds it to a percentile from 0 to 100.
4. Rank candidate archetypes by percentile, with stable published ID order used to break ties. The directory also groups them into role, practice, and style layers so unlike concepts are not presented as one undifferentiated identity list.
5. Apply the result-level interpretation rule. If every raw candidate-archetype signal is below 2, the result reports no clear tendency. If Dominant and Submissive are both at least the 60th percentile and differ by no more than 15 points, the primary role result is Switch. Otherwise, the highest-ranked role is primary and the next role is secondary.

Candidate archetypes are not calculated by simply recombining the nine displayed percentiles. Item responses feed dimension and archetype channels in parallel. The dimensions help explain the shape of a result; they are not a public substitute formula for reconstructing every archetype score.

The 20,000-response fixed-seed theoretical baseline

The baseline artifact is version 1 under p3-v0.1-candidate. Quick and full banks each use 20,000 simulated response sets. The generator seeds its random stream once with 49,734,321 (0x02f6e2b1), builds quick first, then continues the same stream for full; the two versions have separate channel statistics, not independently restarted seeds. For every scored item, it draws from a normal distribution centered at 2 with standard deviation 1.05, rounds, and clips to 0–4. It skips attention checks, applies production normalization and mappings, and records the mean and population standard deviation for each scored channel.

A fixed seed makes the artifact reproducible with the same bank, generator, sample count, and implementation, and prevents routine build drift. Reproducibility does not validate the assumptions. Independent synthetic answers do not model real correlations, community composition, language effects, skipped items, or human context.

The baseline is theoretical, not a surveyed population norm. A 90th-percentile result means the raw signal is high relative to this simulation. It does not mean that the person is “more BDSM” than 90 percent of people, that 90 percent of people answered differently, or that an activity has a 90 percent probability of fitting them.

Worked fictional dimension example and role-rule illustration

This teaching example uses invented prompts, mappings, and baseline statistics; it does not reveal or approximate live item weights. Four fictional answers—4, 1, skipped, and 3—normalize to 1, −0.5, 0, and 0.5. If the second item is reverse-keyed, it becomes +0.5. With illustrative weights 0.8, 0.6, 0.4, and −0.7, the contributions are 0.8, 0.3, 0, and −0.35, producing a fictional raw dimension signal of 0.75.

With an invented baseline mean of 0.10 and standard deviation 0.50, the standardized value is (0.75 − 0.10) ÷ 0.50 = 1.30. The normal cumulative distribution gives about 90.32 percent, rounded to 90. This is a relative location, not agreement, certainty, or a recommendation.

The archetype branch runs in parallel. Imagine fictional raw role signals above the result floor and percentiles of 68 for Dominant and 64 for Submissive. Both reach 60 and differ by 4, so the role rule returns Switch as primary and the strongest non-Switch role as secondary. If every raw archetype signal were below 2, it would report no clear tendency. The label summarizes a pattern; it does not assign identity.

Quick and full version boundaries

The quick test contains 24 scored prompts plus 2 attention checks and is presented as about four minutes. The full test contains 51 scored prompts plus 2 attention checks and is presented as about ten minutes. Both report the same published nine dimensions and 18 candidate archetypes, but they do not use identical item coverage. Each version therefore has its own 20,000-response baseline.

The versions are comparable at the level of broad themes, not as interchangeable measurements. A quick score of 72 and a full score of 64 should not be described as an eight-point decline. Differences may reflect added coverage, different raw-score distributions, skipped items, context, or normal response variation. Use the quick version for a lighter first pass and the full version when broader coverage is worth the additional time. Neither version is a clinical scale, and the full version is not “more true” simply because it is longer.

Published versions and change records

Question banks, scoring logic, and baselines are versioned separately because each can change a result. Comparisons should record the test version and date. Future material changes should add a visible entry.

Research still needed

• Reliability: internal consistency where appropriate, test–retest stability, and the effect of state and context on repeated answers.
• Validity: whether dimensions and candidate labels relate to the constructs they claim to summarize without merely restating item wording.
• Quick/full linking: empirical work to determine which comparisons across versions are defensible and how much uncertainty should be shown.
• Cross-language equivalence: translation review, measurement invariance, and checks for terms that carry different community meanings in English and Simplified Chinese.
• Sample representation: voluntary human samples with transparent recruitment, demographics, context, missingness, and privacy protections.
• Subgroup bias and accessibility: differential item functioning, disability and neurodiversity considerations, relationship-context effects, and usability with assistive technology.
• Response-quality sensitivity: how skipping, all-midpoint responses, straight-lining flags, and intentional impression management affect interpretation.

Until this work is completed and independently reviewed, BDSM Prism should not be described as scientifically validated, clinically reliable, diagnostic, or representative of a population.

Editorial responsibility and corrections

The BDSM Prism content team maintains this explanation and checks it against the scoring source, generated banks, baseline generator, and published artifact. Review is editorial and implementation-focused; no clinician, psychometrician, medical professional, or independent safety specialist is claimed. Any future reviewer, scope, and date should be named.

Material corrections should update this page, its version record, and dependent guides. Product behavior remains the source of truth for implementation facts; this page must not promise unsupported formulas, privacy behavior, or validation.