Built on 25,000+ peer-reviewed studies.
Profyls is a psychometric instrument, not a chatbot. Every score traces to a published framework. We never invent a metric. We compose them.
The 8 frameworks
The instruments we compose from.
LIWC-22
100 metricsThe standard for word-category psychology. 25,000+ peer-reviewed studies. Function-word and content-word categories.
Big 5
35 metricsOpenness, Conscientiousness, Extraversion, Agreeableness, Neuroticism — population-normed Z-scores.
DISC
8 metricsBehavioral style: Dominance, Influence, Steadiness, Conscientiousness — work-context calibrated.
SALLEE
16 metricsEmotion model: Plutchik-style affect detection across 16 emotional dimensions.
Social Dynamics
8 metricsClout, authenticity, analytic thinking, power symmetry — relational signals.
Drives
15 metricsAchievement, Affiliation, Power, Reward, Risk — McClelland-aligned motivational signals.
Needs & Values
18 metricsSchwartz-aligned values map: Hedonism, Universalism, Tradition, Self-Direction, more.
Cognition + Temporal + Toxicity
22 metricsAnalytic thinking, cognitive load, temporal orientation, and ML-based toxicity scoring.
The observatory
180+ validated metrics.
One transparent system.
Every score traces to a published framework. No black boxes — admissible-ready, audit-friendly, and explainable to a judge, an HR partner, or a board.
From text to score
How a Profyls score is actually made.
No black box. Every score traces token-by-token back to a published framework.
1 · Tokenize
Text is segmented into tokens against LIWC-22 dictionaries plus 6 sister lexicons.
2 · Score
Each framework runs independently — no model is allowed to overwrite another's signal.
3 · Compose
Vertical playbooks compose framework outputs into the metrics your work actually uses.
4 · Cite
Every score links back to the framework, the version, the lexicon, and the tokens it counted.
Boundaries we draw
What Profyls is — and isn't.
What it is
- • A psychometric instrument built from peer-reviewed frameworks.
- • A measurement layer for high-stakes language.
- • A prep tool the professional reads before deciding.
- • Audit-friendly, citation-friendly, Daubert-ready.
What it isn't
- • Not a diagnostic device. Not a verdict. Not a clinical decision.
- • Not a lie detector. Not a "personality test."
- • Not a substitute for clinical, legal, or fiduciary judgment.
- • Not training data for someone else's model. Ever.
Ethics & privacy
Your data is the work.
Not the product.
We never train on customer data. We never share across tenants. We delete on demand. Every vertical is audited for the compliance regime it actually lives under.
Security
SOC 2 Type II. End-to-end encrypted in transit and at rest. SSO + SCIM available.
Privacy
Zero data training. Per-tenant isolation. PII minimization by default.
Compliance
HIPAA-aligned (BAA available). GDPR/CCPA. ABA Model Rule 1.6. EEOC-aware scoring for hiring.
Auditability
Every score links to the framework, the tokens it counted, and the version of the model.
How we validate
Every metric earns its spot — three ways.
We don't ship a scoring category until it clears construct, criterion, and population validity. The same standard academic psychology has used for a century.
Construct validity
Does the metric actually measure the thing it claims? Tested against established scales (NEO-PI for Big 5, PANAS for affect, Schwartz Value Survey for values) on >50K texts per construct.
Criterion validity
Does it predict the outcome that matters? Hedging language → clinical drop-out. Power asymmetry → deal failure. Conviction → founder follow-on funding. Receipts on file.
Population validity
Does it hold across gender, age, region, register, and language? Adverse-impact audited; cohort norms published in our admin console for every customer.
Why not just GPT?
Sentiment is a vibe. Psychometrics is a measurement.
| Capability | Generic LLM / sentiment API | Profyls |
|---|---|---|
| Output | Free-text summary or a single polarity score | 189 trace-able metrics across 8 peer-reviewed frameworks |
| Reproducibility | Different result per prompt, per model version | Deterministic — same text, same score, every time |
| Auditability | 'The model said so' | Token-level receipts, framework citations, lexicon version pinned |
| Validity evidence | Whatever the eval set covers | Construct, criterion, and population validation per metric |
| Admissibility / regulator-readiness | Black-box risk | Daubert-ready documentation; SOC 2; HIPAA BAA available |
| Privacy | Often trains on your data | Zero data training. Single-tenant on Firm tier. |
The literature
Standing on 25,000+ studies. Here's a working sample.
Boyd, R.L., Ashokkumar, A., Seraj, S., & Pennebaker, J.W. (2022). The development and psychometric properties of LIWC-22. University of Texas at Austin.
Pulverman, C.S. et al. (2017). Language use and treatment engagement in psychotherapy. Journal of Counseling Psychology.
Newman, M.L., Pennebaker, J.W., Berry, D.S., & Richards, J.M. (2003). Lying words: Predicting deception from linguistic styles. Personality and Social Psychology Bulletin.
Kacewicz, E., Pennebaker, J.W., Davis, M., Jeon, M., & Graesser, A.C. (2014). Pronoun use reflects standings in social hierarchies. JLSP.
Jordan, K.N. et al. (2019). Examining long-term trends in politics and culture through language of political leaders and cultural institutions. PNAS.
Mairesse, F., Walker, M.A., Mehl, M.R., & Moore, R.K. (2007). Using linguistic cues for the automatic recognition of personality. JAIR.
Full bibliography (240+ citations) available on request for academic, clinical, and litigation use.
What you can actually measure
Twelve signals that change the call you make next.
Authenticity
Rehearsed-vs-real signal across pronouns, function words, and verb tense.
Clout
Social-hierarchy positioning — who's the high-status voice in the room?
Analytic thinking
Formal, hierarchical reasoning vs. narrative/here-and-now.
Cognitive load
Story strain — predicts collapse under cross-examination or follow-up.
Tentativeness
Hedging, qualifiers, may/might/could — bluff vs. conviction.
Power symmetry
Two-party balance of clout — predicts deal viability and co-founder fit.
Achievement drive
McClelland-aligned builder energy from action and goal language.
Affiliation drive
Belonging signal — community-led behavior, relationship preservation.
Risk orientation
Self-direction vs. security on Schwartz's value axis.
Emotional register (SALLEE)
16 Plutchik-style affect channels — beyond pos/neg.
Temporal focus
Past/present/future orientation — predicts action vs. rumination.
Toxicity
ML-scored language toxicity for moderation and risk routing.
Where the science is — and isn't — settled.
- • Strong: Function-word patterns, hedging, clout, analytic thinking, basic affect, temporal focus, drive signals.
- • Robust with caveats: Big 5 from text — directional accuracy is high in 5K+ token samples; thinner texts get directional reads only.
- • Use with care: Deception markers — best as a triage signal, never as a verdict. Toxicity — model output, not lexicon, so we publish the confidence band.
- • Off-limits: Clinical diagnosis. Verdict-style decisions. Anything we can't show our work for.
One API. All 189 metrics. Sub-second.
REST + streaming endpoints expose every framework Profyls is built on. Ship a CX intelligence layer, a fraud-language sweep, a clinical co-pilot — without rebuilding the scoring stack.
See API pricingPOST /v2/score
{
"text": "Honestly, I just need a few more weeks to feel ready...",
"frameworks": ["liwc", "big5", "drives", "sallee"]
}
→ {
"liwc": { "tentative": 4.21, "i": 6.8, "authentic": 71.3 },
"big5": { "openness": 0.62, "neuroticism": 0.71 },
"drives": { "achievement": 0.34, "affiliation": 0.18 },
"sallee": { "anxiety": 0.49, "joy": 0.04 }
}