Limitations & Disclaimers

Last updated: September 26, 2026

Not a diagnosis

Clarity does not diagnose manipulation, abuse, mental health conditions, or anyone’s intent. A high or low score does not prove what someone meant or whether a relationship is safe.

Not a safety assessment

Clarity cannot determine whether you are in danger. If you feel unsafe, contact local emergency services or a crisis line, see Resources.

Pattern matching limits

Clarity uses a fixed set of language patterns. Many forms of pressure, coercion, or abuse may not be detected. Neutral or kind messages may still feel difficult. Context, history, and tone matter more than any text score.

Scores are not evidence

Analysis output is not suitable as evidence in legal, workplace, or disciplinary proceedings. Do not treat Clarity output as proof of wrongdoing or innocence.

Not professional advice

Response suggestions are starting points for reflection, not scripts from therapists, lawyers, or mediators. Adapt any language to your situation and relationships.

Media Lens

Media Lens is a separate, experimental preview for public articles, ads, speeches, and campaign material. It is not part of this Clarity site. It produces no overall score for an article, and no field describes a person or outlet as manipulative. It does not fact-check: every claim's support is always "not checked" in this preview, with no path yet to mark a claim otherwise. Coverage information (origin, independent reporting, duplication) appears only in labeled fixture examples. The live preview reads no coverage data, so coverage is marked insufficient rather than guessed, and comparing how different outlets covered the same story is not live. Examples labeled as fixtures or samples are made up. Quoted-speech detection is heuristic and can miss unusual formatting. Media Lens is English-only in this preview. No accuracy claim is made for its output. Nothing it shows is evidence of an outlet's or a person's intent, reliability, or character.

A limited Jev-only live URL preview runs on a separate operator host, ml-jev.manipulationscore.com. It is experimental and not production-ready, and the operator can pause it at any time. It accepts one public page at a time from a short list of hosts the operator allows. The Media Lens page sends only https addresses; the worker behind it also accepts http addresses sent directly to its API, under the same host list and redirect rules. Pages that are too long, cannot be fetched in time, or cannot be extracted are left unanalyzed. On a page Media Lens recognizes as paywalled, only the visible excerpt is analyzed. Fetching a URL does not fact-check the article, does not bypass a paywall, and does not make Media Lens output evidence of an outlet's or a person's intent or reliability. The operator host requests the page you name, and prepared public span text is sent to TypeSafe's Jev classifier. See Privacy for the data flow. Live pasted-text analysis remains prohibited. Private messages still belong in Clarity.

When to seek help

Consider speaking with a counselor, advocate, trusted friend, or support organization if communication consistently leaves you feeling afraid, controlled, or diminished, regardless of what Clarity detects.