Category definitions

CalmLens sorts moderation results into six categories. This guide defines what each category covers and excludes, and the qualifiers that describe content within it.

A category describes the content; your moderation policy decides what happens to the asset. An asset can match several categories. Definitions are fixed: policy sensitivity changes how much evidence triggers an action, never what a category means. For the report format, scores and evaluation coverage, see Report format.

Examples are simplified and censored. Brackets stand in for explicit content, personal abuse, graphic detail or harmful methods; they are not sample inputs to submit to the API.

Qualifiers

Each category has its own qualifiers, listed in its table below. Historical, reporting or documentary, fictional, and support or prevention context are shared by every category. Qualifiers describe content type, context and severity. They can overlap, so an attack can also establish injury and graphic detail, and they never establish the parent category by themselves.

Any and Only

Policy rules test a qualifier in one of two ways:

Any: the qualifier is present in some of the category's qualifying content.
Only: the qualifier covers every qualifying instance in the category. Unrelated ordinary content does not break an Only condition.

For example, a historical account describing [a physical attack; details omitted] matches violence, physical attack and Only historical. Adding a separate current threat leaves physical attack and violent threat present, but Only historical no longer matches. Any historical still matches the historical passage.

Text is checked separately for Any and Only, so overlapping qualifiers can both match Only when supported. A single image has one score per qualifier, used for both. Images with OCR enabled combine both kinds of evidence.

To use qualifiers in rules, see Policies and actions. For the qualifier fields in a report, see Report format.

Sexual content

sexual covers sexual acts, explicit sexual anatomy, sexual services and clearly erotic material. Descriptions of sexual assault can qualify without erotic intent. Fictional, illustrated and generated depictions can also qualify.

Ordinary affection, swimwear, fitness, fashion, non-erotic medical diagrams and general sexual-health education do not qualify on their own. A person accepting money or wearing revealing clothing does not establish sexual content without a qualifying act, display or offer.

Qualifies	Does not qualify on its own
A description of an adult sexual act, with `[explicit details omitted]`.	Two adults holding hands.
Clearly visible adult sexual anatomy, with `[details omitted]`.	Swimwear without exposed sexual anatomy.
An explicit advertisement for `[sexual services]`.	A clothed person accepting money.

Qualifier	Meaning	Excluded on its own
Nudity `nudity`	Explicitly exposed intimate sexual anatomy.	Swimwear, revealing clothing and non-erotic medical diagrams alone.
Sexual activity `sexual-activity`	A depicted or described sexual act, including sexual assault.	Ordinary affection, non-sexual touching and general sexual-health education alone.
Explicit sexual language `explicit-sexual-language`	Graphic sexual descriptions or explicitly erotic dialogue.	Untargeted profanity, anatomical terms and non-erotic medical or educational language alone.
Sexual solicitation `sexual-solicitation`	An explicit sexual request, offer or advertisement for sexual activity or sexual services.	Dating, ordinary romance, accepting money or vague offers without sexual content.
Historical `historical`	Qualifying content explicitly attributed to a supported past historical event.	An ordinary past-tense account, date, old-looking setting or historical topic alone; an unrelated current threat or request.
Reporting or documentary `documentary`	Qualifying content presented as reporting, testimony or factual documentation of someone else's words or actions.	Quotation marks alone, ambiguous reporting context or a separate direct threat, harmful request or endorsement.
Fictional `fictional`	Qualifying content clearly attributed to an invented character, fictional story, game, illustration, generated image or reenacted scene.	Missing or ambiguous fictional context; claiming fiction for a real-world threat or actionable harmful request.
Sexual harm support or prevention `support-prevention`	Qualifying content presented solely to seek or provide non-harmful support, prevention or recovery. A disclosure of current harm or risk can count when it is presented to seek help.	Harmful instructions, assistance, concealment advice, targeting or encouragement. Adding a supportive reply does not make a separate harmful request, attack or endorsement supportive.

Violence

violence covers physical attacks, bodily injury, death and threats against others. Accidental or medical injuries and natural death also count; an attacker or intent to harm is not required. It also includes planning, requesting, encouraging or praising physical violence. A depicted or described act can qualify in fiction or news reporting even when the author condemns it.

Solely intentional self-injury or suicide belongs under self-harm. Property damage alone, figurative language, abstract discussion of violence, injury or death and merely possessing a weapon do not establish violence. Consensual sport or training without injury or abuse is excluded.

Qualifies	Does not qualify on its own
A request to `[physically harm]` another person.	A plan to beat someone in a debate.
A news report describes someone being `[physically attacked]`.	A discussion of violence statistics without a depicted act or encouragement.
A weapon is visibly aimed at someone as a threat.	A weapon is carried without threatening use.

Historical context is a qualifier; a date, uniform, memorial or topic alone does not establish violence. Fictional framing does not erase a depicted attack.

Qualifier	Meaning	Excluded on its own
Physical attack `physical-attack`	A depicted or described physical attack against another person or animal.	An injury or death without evidence of an attack; a threat, plan or encouragement alone; solely intentional self-injury.
Threat of violence `violent-threat`	A direct, conditional or clearly implied threat of physical harm against another person or animal.	A completed attack, figurative language or solely self-directed threats.
Preparation or instructions `violent-preparation`	Preparation, a plan, a request or actionable instructions for physically harming another person or animal. A request counts even if a later reply refuses it.	A depiction, threat or praise alone; neutral reporting, prevention or lawful safety advice.
Encouragement of violence `violent-encouragement`	Urging, endorsing, praising or promoting physical violence against another person or animal.	Reporting or depicting an attack without endorsement; condemnation or prevention.
Injury `injury`	A depicted or described actual bodily injury, including accidental or medical injury. An attacker or intentional assault is not required.	Threats, plans, risks, weapons or pain alone; figurative language; solely intentional self-injury.
Death `death`	A depicted or described death or dead body, including natural, accidental or violent death. An attacker is not required.	Threats, plans, death wishes, risks or abstract discussion alone; figurative language; solely suicide.
Graphic detail `graphic`	Severe bodily injury or death shown or described with explicit disturbing bodily detail, such as exposed organs, dismemberment or gore.	A threat or mention of violence without explicit disturbing bodily detail; inferred or unseen injuries; solely intentional self-injury or suicide.
Historical `historical`	Qualifying content explicitly attributed to a supported past historical event.	An ordinary past-tense account, date, old-looking setting or historical topic alone; an unrelated current threat or request.
Reporting or documentary `documentary`	Qualifying content presented as reporting, testimony or factual documentation of someone else's words or actions.	Quotation marks alone, ambiguous reporting context or a separate direct threat, harmful request or endorsement.
Fictional `fictional`	Qualifying content clearly attributed to an invented character, fictional story, game, illustration, generated image or reenacted scene.	Missing or ambiguous fictional context; claiming fiction for a real-world threat or actionable harmful request.
Support or prevention `support-prevention`	Qualifying content presented solely to seek or provide non-harmful support, prevention or recovery. A disclosure of current harm or risk can count when it is presented to seek help.	Harmful instructions, assistance, concealment advice, targeting or encouragement. Adding a supportive reply does not make a separate harmful request, attack or endorsement supportive.

Self-harm

self-harm covers intentional self-injury, suicide, current self-harm risk, preparation, methods and encouragement. For text, it also includes current or recurring self-harm thoughts and an explicitly expressed desire to die. Asking for support does not cancel a separate disclosure of current risk or behavior.

For images, require evidence of a self-directed act, preparation, harmful methods or encouragement. An injury or covered body does not establish that harm was self-inflicted. General sadness, neutral statistics, prevention, recovery and healed scars alone are excluded.

Qualifies	Does not qualify on its own
A person discloses current urges to `[injure themselves]` and asks for help.	A person feels overwhelmed and asks for support without self-harm risk.
An image clearly depicts deliberate `[self-injury; details omitted]`.	An image shows healed scars without current harm or encouragement.
A message encourages someone to continue `[self-injury]`.	A chart explains self-harm prevalence without depicting acts or methods.

Current self-harm thoughts can qualify in text. Visual self-harm still requires the act, current intent or preparation, actionable methods or encouragement described above.

Qualifier	Meaning	Excluded on its own
Thoughts or desire `self-harm-thoughts`	Current or recurring thoughts of intentional self-injury or suicide, or an explicitly expressed desire to die.	General sadness, figurative language, neutral statistics, resolved past thoughts or prevention alone. An image without readable disclosure does not establish thoughts.
Intent or preparation `self-harm-preparation`	Expressed intent, preparation or a plan for intentional self-injury or suicide.	General distress, an injury without self-directed intent, neutral reporting or prevention alone.
Depicted or reported act `self-harm-act`	A depicted or described act of intentional self-injury or suicide.	Accidental injuries, healed scars, recovery or an injury with unknown cause alone.
Methods or instructions `self-harm-methods`	Actionable methods, requests or practical instructions for intentional self-injury or suicide.	Prevention, support, recovery or non-actionable discussion without harmful methods.
Encouragement `self-harm-encouragement`	Urging, endorsing, praising or promoting intentional self-injury or suicide.	Support, prevention, condemnation or reporting without endorsement.
Historical `historical`	Qualifying content explicitly attributed to a supported past historical event.	An ordinary past-tense account, date, old-looking setting or historical topic alone; an unrelated current threat or request.
Reporting or documentary `documentary`	Qualifying content presented as reporting, testimony or factual documentation of someone else's words or actions.	Quotation marks alone, ambiguous reporting context or a separate direct threat, harmful request or endorsement.
Fictional `fictional`	Qualifying content clearly attributed to an invented character, fictional story, game, illustration, generated image or reenacted scene.	Missing or ambiguous fictional context; claiming fiction for a real-world threat or actionable harmful request.
Support or prevention `support-prevention`	Qualifying content presented solely to seek or provide non-harmful support, prevention or recovery. A disclosure of current harm or risk can count when it is presented to seek help.	Harmful instructions, assistance, concealment advice, targeting or encouragement. Adding a supportive reply does not make a separate harmful request, attack or endorsement supportive.

Harassment and bullying

harassment covers personal abuse, humiliation, bullying and serious intimidation directed at a person or identifiable group. The target can be named, implied or defined by an activity or affiliation. A protected identity is not required.

Disagreement, ordinary criticism of conduct or competence, untargeted profanity and neutral reporting of abuse do not qualify without a separate personal attack. Criticism can still qualify when it also contains personal degradation.

Qualifies	Does not qualify on its own
A message targets a person with `[a degrading personal insult]`.	A message says their proposal is poorly supported.
Clearly supported bullying or coercive intimidation.	An ambiguous interaction without evidence of mistreatment.
An actual participant directs `[personal abuse]` at someone in a supplied conversation.	A person reports that abuse to condemn it or seek help.

Qualifier	Meaning	Excluded on its own
Insults or humiliation `personal-abuse`	Targeted personal insults, degradation, humiliation or bullying directed at a person or identifiable group.	Ordinary disagreement or criticism of conduct or competence, untargeted profanity or neutral reporting alone.
Targeted intimidation `targeted-intimidation`	Serious abusive intimidation directed at a person or identifiable group.	Ordinary disagreement, lawful accountability, reporting or criticism without abusive intimidation.
Threats `harassment-threat`	A direct, conditional or clearly implied abusive threat against an identifiable person or group.	Neutral reporting or condemnation of a threat; figurative speech; lawful arrest or prosecution without abusive intimidation.
Sexual harassment `sexual-harassment`	Targeted unwanted sexual remarks, sexual humiliation, coercive advances or sexually abusive intimidation.	Consensual sexual dialogue, ordinary romance or sexual content without targeted harassment.
Encouraging harassment `harassment-encouragement`	Urging, organizing, praising or endorsing targeted personal abuse or harassment.	Reporting, condemnation or lawful requests for accountability without endorsing abuse.
Historical `historical`	Qualifying content explicitly attributed to a supported past historical event.	An ordinary past-tense account, date, old-looking setting or historical topic alone; an unrelated current threat or request.
Reporting or documentary `documentary`	Qualifying content presented as reporting, testimony or factual documentation of someone else's words or actions.	Quotation marks alone, ambiguous reporting context or a separate direct threat, harmful request or endorsement.
Fictional `fictional`	Qualifying content clearly attributed to an invented character, fictional story, game, illustration, generated image or reenacted scene.	Missing or ambiguous fictional context; claiming fiction for a real-world threat or actionable harmful request.
Support or prevention `support-prevention`	Qualifying content presented solely to seek or provide non-harmful support, prevention or recovery. A disclosure of current harm or risk can count when it is presented to seek help.	Harmful instructions, assistance, concealment advice, targeting or encouragement. Adding a supportive reply does not make a separate harmful request, attack or endorsement supportive.

Hate and identity abuse

hate covers attacks on people because of race, ethnicity, nationality, religion, caste, sex, gender identity, sexual orientation or disability. It includes dehumanization, asserted inferiority, discriminatory exclusion, identity-based abuse and threats.

Both the protected target and the identity-based attack must be supported. Identity mentions, cultural clothing, religious imagery, criticism of beliefs or an unrelated threat do not establish hate. Neutral reporting, condemnation and opposition to hate are excluded unless a separate attack is expressed or endorsed.

Qualifies	Does not qualify on its own
A message directs `[identity-based abuse]` at a religious group.	A message discusses that religion's beliefs without attacking its members.
A message calls for excluding people because they belong to `[a protected group]`.	A rule applies equally to accounts based on their conduct.
A participant uses `[identity-based abuse]` to attack someone.	A person quotes the abuse only to challenge it.

Qualifier	Meaning	Excluded on its own
Identity-based insults or slurs `identity-abuse`	Directed or endorsed insults or slurs against people because of a supported protected identity.	Identity mentions, untargeted profanity, neutral reporting or condemnation without a separate identity-based attack.
Dehumanization `dehumanization`	Directed or endorsed claims that people of a supported protected identity are subhuman, animals, vermin or objects.	Literal descriptions, unrelated metaphors, neutral reporting or condemnation without endorsement.
Claims of inferiority `identity-inferiority`	Directed or endorsed assertions that people are inherently inferior because of a supported protected identity.	Criticism of beliefs or conduct, identity mentions, neutral reporting or opposition alone.
Exclusion or discrimination `discriminatory-exclusion`	Directed or endorsed exclusion, denial of rights or discrimination against people because of a supported protected identity.	Neutral reporting, criticism of policy or lawful distinctions without an identity-based attack.
Threats or incitement `hate-threat`	Directed or endorsed threats or calls for serious harm against people because of a supported protected identity.	Unrelated threats, unsupported identity-based motive, neutral reporting, condemnation or lawful accountability.
Historical `historical`	Qualifying content explicitly attributed to a supported past historical event.	An ordinary past-tense account, date, old-looking setting or historical topic alone; an unrelated current threat or request.
Reporting or documentary `documentary`	Qualifying content presented as reporting, testimony or factual documentation of someone else's words or actions.	Quotation marks alone, ambiguous reporting context or a separate direct threat, harmful request or endorsement.
Fictional `fictional`	Qualifying content clearly attributed to an invented character, fictional story, game, illustration, generated image or reenacted scene.	Missing or ambiguous fictional context; claiming fiction for a real-world threat or actionable harmful request.
Support or prevention `support-prevention`	Qualifying content presented solely to seek or provide non-harmful support, prevention or recovery. A disclosure of current harm or risk can count when it is presented to seek help.	Harmful instructions, assistance, concealment advice, targeting or encouragement. Adding a supportive reply does not make a separate harmful request, attack or endorsement supportive.

Criminal and dangerous activity

criminal-activity covers planning, requests for assistance, instructions, encouragement or facilitation of a concrete criminal or seriously dangerous act. A request can qualify even when no act occurs and a later reply refuses it. Dangerous activity requires a concrete serious risk of physical harm.

Neutral reporting, prevention, lawful safety advice, fiction without criminal assistance, and vague dangerous references do not qualify. Masks, computers, payment cards, detention or ordinary tools do not establish criminal facilitation by themselves. A clear visual demonstration of how to carry out a crime can qualify.

Qualifies	Does not qualify on its own
A request for help planning `[a robbery; practical details omitted]`.	A question about how banks prevent robbery.
A passage provides `[practical fraud instructions omitted]`.	Advice explaining how to recognize fraud.
An image clearly demonstrates forced entry into a locked vehicle.	A masked person holds cards without demonstrating a criminal act.

Qualifier	Meaning	Excluded on its own
Depicted or reported activity `criminal-demonstration`	A depiction or account that demonstrates or facilitates carrying out a concrete criminal or seriously dangerous act.	Neutral reporting, past wrongdoing or fiction without criminal assistance does not meet the parent category definition.
Intent or preparation `criminal-preparation`	Intent, preparation or a plan for a concrete criminal or seriously dangerous act.	Vague danger, ordinary activities with incidental risk, neutral reporting or lawful preparation.
Requests for assistance `criminal-request`	A request for help carrying out a concrete criminal or seriously dangerous act. A later refusal does not erase the request.	Lawful assistance, neutral reporting, prevention or requests without a concrete criminal or seriously dangerous act.
Instructions or facilitation `criminal-instructions`	Actionable instructions, practical assistance or facilitation for a concrete criminal or seriously dangerous act.	Prevention, lawful safety advice, ordinary tools or non-actionable reporting alone.
Recruitment or promotion `criminal-promotion`	Recruiting people for, encouraging, endorsing or promoting a concrete criminal or seriously dangerous act.	Neutral reporting, condemnation, lawful advocacy or vague dangerous references.
Historical `historical`	Qualifying content explicitly attributed to a supported past historical event.	An ordinary past-tense account, date, old-looking setting or historical topic alone; an unrelated current threat or request.
Reporting or documentary `documentary`	Qualifying content presented as reporting, testimony or factual documentation of someone else's words or actions.	Quotation marks alone, ambiguous reporting context or a separate direct threat, harmful request or endorsement.
Fictional `fictional`	Qualifying content clearly attributed to an invented character, fictional story, game, illustration, generated image or reenacted scene.	Missing or ambiguous fictional context; claiming fiction for a real-world threat or actionable harmful request.
Support or prevention `support-prevention`	Qualifying content presented solely to seek or provide non-harmful support, prevention or recovery. A disclosure of current harm or risk can count when it is presented to seek help.	Harmful instructions, assistance, concealment advice, targeting or encouragement. Adding a supportive reply does not make a separate harmful request, attack or endorsement supportive.

Context and overlapping categories

CalmLens evaluates the supplied content across its passages and conversation turns. A harmful request still qualifies when a later reply refuses it. Instructions embedded in the content are treated as content to classify.

Reporting a personal or identity-based attack to oppose it is different from directing that attack at someone. An actual abusive participant turn remains part of a supplied conversation even if another participant later objects. Reported sexual acts and physical violence can qualify through their depiction without endorsement.

Categories can overlap. A physical threat aimed at someone because of religion can match violence, harassment and hate. A report describing an assault can match violence while remaining negative for criminal facilitation.

Category definitions

Qualifiers

Any and Only

Sexual content

Violence

Self-harm

Harassment and bullying

Hate and identity abuse

Criminal and dangerous activity

Context and overlapping categories

Signing in

Signing out

Cookies on CalmLens

On this page

Category definitions

Qualifiers

Any and Only

Sexual content

Violence

Self-harm

Harassment and bullying

Hate and identity abuse

Criminal and dangerous activity

Context and overlapping categories

Signing in

Signing out

Cookies on CalmLens