Accuracy Disclaimer

⚠ Critical Notice

Candor deception scores are probabilistic estimates based on linguistic pattern analysis — they are NOT legal determinations of truth or deception, NOT forensic evidence, and NOT admissible as standalone proof in any legal, employment, or regulatory proceeding. Results must never be used as the sole basis for consequential decisions about individuals.

1. What Candor's Scores Actually Mean

Candor analyzes written text for statistical patterns associated with deceptive communication in the academic literature, including pronoun distancing, hedging language, detail specificity, cognitive complexity markers, and emotional leakage signals.

A high deception score means: this text exhibits linguistic patterns statistically associated with deceptive communication in research datasets.

A high deception score does not mean:

The author is lying
The statements are false
The author has deceptive intent
The document should be rejected, denied, or treated as fraudulent

2. Inherent Limitations of Linguistic Analysis

Deception detection through language analysis is an active area of research with significant known limitations. Users must understand these before relying on any output:

False Positives Are Common

Honest communications frequently score high on deception indicators. Causes include:

Stress and anxiety: Truthful people under pressure use linguistic patterns that overlap with deception markers (hedging, increased distancing, cognitive disfluency)
Non-native speakers: Language learners and non-native speakers exhibit structural patterns that may trigger deception signals for linguistic rather than deceptive reasons
Communication style: Formal writing, legal documents, and professionally edited text may score differently than spontaneous speech regardless of truthfulness
Cultural variation: Deception linguistic norms vary significantly across cultures; research datasets are primarily English-language and Western
Topic complexity: Discussing genuinely complex or uncertain topics produces hedging and qualification that resembles deception signals

False Negatives Also Occur

Sophisticated deceptive communications — particularly those crafted by experienced writers, legal teams, or individuals familiar with deception detection methods — may score low. Candor is not a reliable method for detecting prepared or coached deception.

No Ground Truth Validation on Your Data

Candor's models are trained and validated on research datasets. Performance on any specific real-world use case may differ significantly from laboratory accuracy metrics. The system has no access to external facts and cannot verify whether claims are true or false.

Accuracy estimate: In controlled research datasets, linguistic deception detection models achieve roughly 60–70% accuracy — meaningfully above chance, but far from reliable enough to justify adverse action against any individual. Real-world accuracy on your specific data may be lower.

3. Appropriate Use in High-Stakes Contexts

Candor is designed for professional use across a range of high-stakes industries — including insurance fraud investigation, employment screening, legal and compliance review, and trust & safety operations. These are legitimate and intended use cases for the product.

In these contexts, Candor functions as a supplementary analytical tool that flags communications warranting closer human review. It is not a decision-making system and must never replace the professional judgment of qualified investigators, claims adjusters, legal counsel, or HR professionals.

Core requirement: Candor scores must not serve as the sole or primary basis for any adverse decision affecting an individual — including claim denial, application rejection, disciplinary action, or legal determination. All adverse decisions require independent human judgment and corroborating evidence.

Specific safeguards that must be applied in professional contexts:

Insurance fraud investigations — Candor may be used to prioritize claims for review or flag communications for investigator attention. A flagged score alone does not justify claim denial; investigators must conduct independent inquiry before adverse action.
Employment and HR screening — Candor may be used to identify communications that merit follow-up questions or closer evaluation. Employment decisions must be grounded in lawful criteria and direct evidence — not automated scores.
Legal and compliance review — Candor may be used as a research and triage tool by legal professionals. Analysis outputs are not forensic evidence, do not establish credibility, and are not admissible as proof of deception in any proceeding.
Financial fraud detection — Candor may support fraud teams in prioritizing cases for investigation. It does not establish fraud; confirmed adverse actions require independent evidence.
Trust & safety operations — Candor may assist in identifying content warranting human review. Human reviewers must make final determinations on enforcement actions.

Users operating in regulated industries are responsible for ensuring their use of Candor complies with applicable law, including any requirements governing automated decision-making tools, fair lending, equal employment opportunity, or claims handling regulations in their jurisdiction.

4. Not a Replacement for Expert Judgment

Candor is a computational tool, not an expert system or professional service. It does not replace:

Qualified forensic linguistic analysis by a certified expert
Professional polygraph examination (which itself has contested scientific validity)
Legal counsel's assessment of the credibility of claims or documents
An insurer's claims investigation process
HR professional judgment in employment matters

For high-stakes decisions, seek qualified professional expertise. Do not substitute an automated score for human expertise in consequential matters.

5. User Responsibility

By using Candor, you accept sole responsibility for:

How you interpret analysis results
Decisions you make based in whole or in part on those results
Consequences to individuals affected by your decisions
Compliance with all applicable laws governing the use of automated decision-making tools in your jurisdiction and industry

Best practice: Use Candor as a triage and prioritization tool — to identify communications that merit closer human review — not as a decision-making system. Always apply independent judgment before taking action on any result.

6. No Legal or Professional Advice

Nothing on this website or in Candor's analysis outputs constitutes legal, forensic, medical, financial, or professional advice of any kind. Candor is a software tool. Always consult qualified professionals before making decisions with significant legal, financial, or personal consequences.

7. Limitation of Liability

Candor shall not be liable for any decisions made, actions taken, or harms caused based on analysis results. See our Terms of Service for the complete limitation of liability provisions, including the cap on damages equal to fees paid.

8. Contact

Questions about this disclaimer or Candor's methodology? Contact us at: legal@getcandor.com