# Website Categorization policy pack `policy_v2` — group reference Companion to §11 of the Website Categorization bulk feed data dictionary (), to the real-time API's `?pack=policy_v2` parameter and to the `policy_v1` reference [`website-categorization-policy-pack-v1.md`](https://website-categorization.whoisxmlapi.com/samples/website-categorization-db/website-categorization-policy-pack-v1.md), which this pack succeeds. Pack `policy_v2`, mapping version **2026.09.1**, taxonomy `iab_v3.1`, 47 groups. Branched from `policy_v1` mapping 2026.09.2. ## What changed from `policy_v1` Two things, and nothing else: 1. **The `news` group is retired; its nodes are `current_events`.** `policy_v1` published the three current-affairs topic roots — `Crime`, `Disasters`, `War and Conflicts` — under the id `news`, which reads like "news sites" and is not: IAB v3.1 has no news-publisher node, so no group contains newspapers or news portals, and a policy that allowed or blocked `news` allowed or blocked no news site. The same three roots are now `current_events`, an id that says what they are. **Migration:** replace `news` with `current_events` in any policy written against `policy_v1` ids. Every other node — 701 of 704 — resolves to exactly the group `policy_v1` 2026.09.2 gives it, and every other group keeps its id, name and `sensitive` flag. 2. **The projection reads the brand-safety axis first.** `policy_v1` computed `policy_group` from the primary IAB label alone. `policy_v2` carries a `safety_rules` table — brand-safety theme → group — and reads a row's flagged themes _before_ its labels, so a domain the adult, gambling or Sensitive-Topics detector flagged lands in the block-critical group **with provenance** (`policy_basis` = `safety`) unless its primary label already puts it in a different block-critical group — the safety axis never moves a row _out_ of one. The precedence is spelled out in [the next section](#the-safety-first-projection); it runs on the feed from schema 1.14.0 and on the API once the API carries the safety axis. Why a new pack id rather than a `policy_v1` version bump: a group **id never changes meaning inside a pack**. A rename is a retirement, and a retirement is a new pack id — so a policy written against `policy_v1` ids keeps working unchanged for as long as `policy_v1` is served, and adopting `policy_v2` is an explicit, one-word migration. The lineage is recorded in the pack itself (`previous_pack`, below) and checked against both packs' frozen mappings in CI. ### Selecting the pack - **API.** `?pack=policy_v2` together with `include=policy` on `/api/v4/categorize` and `/api/v4/categorize/bulk`, or on `/api/v4/taxonomy` for the full annotated tree, whose top-level `pack` object also lists every group's `block_with` union (see [Block-with unions](#block-with-unions)). `?pack=policy_v1` keeps working; without `?pack=` the API projects through its default pack (`policy_v1` today — a change of default is announced in the API changelog before it ships). An id the API does not bundle is a `422` (`policy_v3` is bundled too); a bundled pack the server has not loaded is a `503`, never a silently un-annotated `200`. `?pack=policy_v2` is served from the API image that carries the pack, announced in the API changelog when it ships; an earlier image does not know the id and answers `422`. **Today the API's `policy` block projects the primary label only** (`policy.basis` is `primary`, or `null`): the API does not yet carry the brand-safety axis, so the safety-first read below applies to the **feed** first; it reaches the API together with the API's safety block, announced in the API changelog when it ships. The same block also carries `confidence_basis` — what that row's `confidence` numbers actually are, one of `model_raw`, `filter`, `rule`, `curated`, `carry_over` or `none`, where `none` means there is no label to attribute a confidence to or the writer was not recorded. It also carries `abstained`, the same flag as the top-level `abstain.is_abstain`: when it is `true` the `group` stays populated but is **advisory** — we still project whatever was classified, but the classification itself is one we flagged in `abstain` (no real business purpose, or too little text to judge). - **Feed.** `manifest.json` → `policy_pack.id` names the pack a snapshot was exported under, and every row carries it as `policy_pack_version` (`policy_v2@2026.09.1`). `policy_v2` becomes the feed's pack with feed schema **1.14.0** — see the Website Categorization feed changelog for the date, the row-level effect and what `safety_flags` is populated from on a given artifact. A pack change ships via full snapshot only; the policy columns are derived, not tracked, so a delta never carries one. ## The safety-first projection `policy_group` (feed, schema ≥ 1.14.0) is read from three signals in a fixed order, and `policy_basis` names the one that answered. (The API's `policy.group` / `policy.basis` read the primary label only until the API carries the safety axis — see _Selecting the pack_ above.) | `basis` | Read from | Answers when | | -------------- | ----------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | `safety` | the row's flagged brand-safety themes (feed `safety_flags`, schema ≥ 1.14.0) | a flagged theme names a group, unless the primary label already sits in a _different_ sensitive group (except the theme-unknown bucket `sensitive_other` — see the rules below) | | `site_purpose` | the site-purpose label (`site_purpose_tier1..4`), resolved through the same rule table as the primary | it is populated and resolves, and no flagged theme names a group | | `primary` | the primary IAB label (`tier1..tier4`), deepest matching prefix wins | otherwise, when it resolves — and always when it already sits in a sensitive group a flagged theme does not name (except the theme-unknown bucket `sensitive_other` — see the rules below) | | `role` | — reserved — | never, in this pack; **reserved** for the role axis in a later release — no date is promised. Strict-enum clients: admit it now so a future artifact cannot break you. | `group` and `basis` are `null` together, exactly when nothing resolves: no flagged theme names a group, and neither the site-purpose label nor the primary label resolves (an unknown tier-1 with no site-purpose label, say). `basis` is a **closed vocabulary** — `safety`, `site_purpose`, `primary` — with `role` reserved; a client that validates the field against an enum should admit all four today. ### Safety rules | Theme (`safety_flags` value) | Projects to | Notes | | ---------------------------- | --------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `adult` | `adult` | Adult rule filter (pipeline/adult_filter), adjacent fires. Overrides any non-sensitive primary group; confirms an `adult` primary with basis `safety`. | | `gambling` | `gambling` | Gambling rule filter, licensed-casino target (the row's own label is `Attractions > Casinos & Gambling`). | | `harmful_gambling` | `harmful_content` | Gambling rule filter, HARMFUL target (illegal/unlicensed markers) — the same node the primary label lands on, `Sensitive Topics > Spam or Harmful Content`; deliberately not `gambling`. | | `sensitive` | `sensitive_other` | Sensitive-Topics filter: sensitive, theme unknown. A primary label that already names a specific sensitive theme (`Sensitive Topics > Adult & Explicit Sexual Content` → `adult`) answers instead; a bare `Sensitive Topics` label is the same group, so the verdict answers and is the row's provenance. | | `vendor` | — (no group; falls through) | Vendor rule: a ROLE verdict (the site sells the thing), not a topic — names no group; the projection falls through to site_purpose / primary. | One precedence, for every theme, exactly as the CI tests pin it: - **A flagged theme overrides any non-sensitive primary group.** `adult` on a row stored as `Entertainment > Movies` reads `adult` / `safety`; `sensitive` on the same label reads `sensitive_other` / `safety`. - **A flagged theme confirms a primary already in the same group.** `adult` on a row stored as `Sensitive Topics > Adult & Explicit Sexual Content` reads `adult` / `safety`; `sensitive` on a bare `Sensitive Topics` label reads `sensitive_other` / `safety`. The block is the same; the basis records that a detector, not only the label, put the row there — this is what makes the block set of the legacy filter rows reproducible from the safety axis alone. - **A primary already in a _different_ sensitive group answers itself.** The label is at least as specific as the verdict: `sensitive` on `Sensitive Topics > Adult & Explicit Sexual Content` keeps `adult` / `primary`, on `Sensitive Topics > Online Piracy` keeps `piracy` / `primary`; `gambling` on the adult label keeps `adult` / `primary`. The row stays in its block-critical group and the verdict stays on the row in `safety_flags`; a site-purpose label is **not** consulted in this case, so the row cannot be pulled out of the group. The one exception is the theme-unknown bucket: `sensitive_other` says only _sensitive, theme unknown_, so it is never as specific as a theme-specific verdict — `adult` on a bare `Sensitive Topics` label reads `adult` / `safety` (so does `["adult", "sensitive"]` on it). The row stays in a block-critical group, only a more specific one. - **`vendor` names no group.** It is a _role_ verdict (the site sells the thing) rather than a topic, so the projection falls through to `site_purpose` and then `primary`. - **Several flagged themes:** the first entry in the table order that may answer does; a theme that names the primary's own sensitive group wins over one that does not. A theme the pack does not know (a shadow theme, a typo in a flag) is ignored — the projection runs over every row and must always answer. What the projection does and does not change: - **Blocks are preserved, with provenance.** Every node `policy_v1` placed in a block-critical group is still in one under `policy_v2`, and the safety axis can only move a row _into_ a block-critical group — never out of one: a flagged row whose primary label already sits in a different sensitive group keeps that group, except the theme-unknown bucket `sensitive_other`, which a theme-specific verdict refines to that theme's group (basis `safety`). Rows the brand-safety filters flagged additionally arrive there with `basis = safety`. A block list keyed on group ids needs only the `news` → `current_events` edit. (The site-purpose axis sits between safety and primary exactly as the `policy_v1` reference documented it; it is empty on every row today, and when it is populated it outranks the primary label on a row with no applicable flagged theme.) - **It still cannot repair a wrong primary label on an unflagged row.** An adult site the detectors did not flag and the classifier stored as `Entertainment > Movies` reads `entertainment` / `primary`, exactly as under `policy_v1`. No accuracy figure attaches to a group by virtue of the pack; where per-group precision or recall is quoted, it is quoted with its frame and sample size, never here. Evaluate on your own domains before you block on it. - **`sensitive_other` shrinks only by a theme-specific verdict.** The Sensitive-Topics detector reports _sensitive, theme unknown_; its flagged rows project to `sensitive_other` unless the primary label already names the theme or a theme-specific detector flags the row (`adult` on a bare `Sensitive Topics` label reads `adult` / `safety`). Until such a detector runs on those rows, the bucket does not shrink. Worked examples on feed rows (`safety_flags`, site-purpose label, primary label → `group` / `basis`): | `safety_flags` | site_purpose | primary label | → group / basis | | ---------------------- | ---------------------------------- | ---------------------------------------------------- | ---------------------------- | | `["adult"]` | — | `Entertainment > Movies` | `adult` / `safety` | | `["harmful_gambling"]` | — | `Sensitive Topics > Spam or Harmful Content` | `harmful_content` / `safety` | | `["sensitive"]` | — | `Sensitive Topics > Adult & Explicit Sexual Content` | `adult` / `primary` | | `["gambling"]` | — | `Sensitive Topics > Adult & Explicit Sexual Content` | `adult` / `primary` | | `["sensitive"]` | `Entertainment` | `Sensitive Topics > Adult & Explicit Sexual Content` | `adult` / `primary` | | `["sensitive"]` | — | `Sensitive Topics` | `sensitive_other` / `safety` | | `["adult"]` | — | `Sensitive Topics` | `adult` / `safety` | | `["vendor"]` | — | `Business and Finance > Industries` | `business` / `primary` | | `[]` | `Attractions > Casinos & Gambling` | `Entertainment` | `gambling` / `site_purpose` | | `[]` | — | `Crime` | `current_events` / `primary` | | `[]` | — | _(unknown tier-1)_ | `null` / `null` | ## Versioning and lineage The mapping is versioned independently of the feed schema and the API, exactly as for `policy_v1` (see its Versioning section [`website-categorization-policy-pack-v1.md`](https://website-categorization.whoisxmlapi.com/samples/website-categorization-db/website-categorization-policy-pack-v1.md) for the CalVer rules, the changelog fields and the frozen per-version snapshots — all of which apply unchanged; this pack's snapshots are `policy_packs/data/history/policy_v2@.json`). Pin the version (feed: `manifest.json` → `policy_pack.version`, or the per-row `policy_pack_version`; API: `policy.pack_version`) if your policy depends on exact group membership. A change to a `safety_rules` entry is a mapping change too: it bumps the version and is recorded in the changelog like a moved path. So is a change to a group's `block_with` union: from feed schema 1.14.0 the union is expanded into every row's `policy_group_set`, so a union that changes changes rows. One block is new in kind. A pack that succeeds another carries **`previous_pack`** — the machine-readable cross-pack diff, in the same shape as a changelog entry, checked in CI against the predecessor's frozen mapping and this pack's own first-version mapping: | Field | Meaning | | -------------------------------- | -------------------------------------------------------------------------------------------------- | | `pack`, `version` | the pack and mapping version this id branched from (`policy_v1`, `2026.09.2`) | | `moved` | every taxonomy path whose group differs between the two, as `{path, from, to}` — the complete diff | | `added_groups`, `retired_groups` | group ids that exist only in this pack / only in the predecessor | | `notes` | the human explanation | ## Block-with unions A group's **`block_with`** names the other groups a policy that blocks it should block alongside it — the unions this reference has always described in prose ("block `gambling`, `gambling_games` and `harmful_content` to block everything gambling-shaped"), now carried by the pack itself so a client can expand a block list without reading this page. `policy_v2` declares three: | Block … | … with | Why | | ---------- | ----------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `gambling` | `gambling_games`, `harmful_content` | simulated casinos are a separate group by design (Option C), and the gambling detector's illegal-gambling markers land on `Sensitive Topics > Spam or Harmful Content`, i.e. in `harmful_content`, not in `gambling` | | `drugs` | `alcohol`, `tobacco` | IAB's single substances node holds the _sensitive_ framing of all four substances; the neutral beverage and cigar content of the same substances sits in the two consumer groups | | `weapons` | `hunting_fishing` | hunting and shooting sports are split from firearms, as in Cisco, BrightCloud, Fortinet and Palo Alto | Three rules keep the metadata honest: - **Symmetric.** Every partner declares the group back — `gambling_games` and `harmful_content` each name `gambling`, `alcohol` and `tobacco` name `drugs`, `hunting_fishing` names `weapons` — and the pack refuses to load otherwise. `block_with` lists a group's _direct_ partners, not the transitive closure: `gambling_games` names `gambling` only, so a policy keyed on `gambling_games` alone does not reach `harmful_content` — start from the head group of the union (`gambling`, `drugs`, `weapons`) to get all of it. - **Advisory, never a rule.** `block_with` moves no row: `policy_group` is exactly what the mapping and the safety-first projection give it. The union says which _other_ groups a block policy should key on; whether to pay for the wider block is your call (see the cost below). - **Versioned like the mapping.** From feed schema 1.14.0 every row also carries `policy_group_set` — its own group plus every group whose union names it — so a union that changes changes rows: a `block_with` edit bumps the mapping version like a moved path. Where to read it: `GET /api/v4/taxonomy?pack=policy_v2` → `pack.groups[].block_with` (every group is listed; a group with no union carries `[]`), and the **Block with** column of the [group table](#groups) below, which is generated from the pack. **What a union costs.** A union is wider than its head group on purpose, and the extra rows are not all the thing you set out to block: - **Gambling** adds `harmful_content` wholesale. On the 2026-09-02 reference partition (one of sixteen) the gambling-shaped union is 253,735 `gambling` + 0 `gambling_games` + 6,401 `harmful_content` = 260,136 rows; of the 6,401 `harmful_content` rows, 6,169 are the gambling rule filter's illegal-gambling markers and 232 are ensemble spam/scam rows with no gambling shape — the over-block is those 232 of the 6,401, in a union of 260,136. A policy that wants only real-money gambling keys on `gambling` alone; one that wants everything gambling-shaped accepts blocking spam and scam content with it. - **Drugs** adds neutral consumer content — breweries, wineries, wine and spirits retail (`alcohol`); cigar hobby and retail (`tobacco`) — that a sensitive-substances policy does not need unless it is a total substances block. - **Weapons** adds sport hunting, shooting sports and fishing (`hunting_fishing`), which a firearms-sales block does not otherwise touch. ## Groups The **Nodes** column counts the IAB v3.1 categories (of 704) that project to the group; the **Block with** column is the group's `block_with` union (blank for a group with no union — see [Block-with unions](#block-with-unions)); the **Rule paths** column lists the IAB paths the pack names explicitly (each rule also covers the path's descendants unless a deeper rule overrides it). Where a group's row reads exactly as in the `policy_v1` reference, that is because the mapping is the same. | Group id | Name | Sensitive | Nodes | Block with | Definition | Rule paths | | ------------------------- | ------------------------------- | :-------: | ----: | ----------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `adult` | Adult | yes | 2 | | Pornography and explicit sexual content (IAB `Sensitive Topics > Adult & Explicit Sexual Content`), plus adult-only video games. Vendors call this Adult / Pornography. Sex education and sexual health are `health`; lingerie retail is `fashion_beauty`; dating is `dating`. IAB v3.1 has no separate non-sexual-nudity node. | `Sensitive Topics > Adult & Explicit Sexual Content`
`Video Gaming > Video Game Genres > Adult Video Games` | | `gambling` | Gambling | yes | 3 | `gambling_games`, `harmful_content` | Real-money gambling: casinos and betting (`Attractions > Casinos & Gambling`), lotteries and scratchcards (`Shopping`), and poker and professional gambling (`Sports`). Casino-themed video games with no real-money play are the separate `gambling_games` group (Option C, 2026.09.2). Lotteries sit inside Gambling for most vendors; fantasy sports and horse racing stay in `sports` (Cisco, Palo Alto convention). Illegal-gambling markers detected by our gambling rule filter land on `Sensitive Topics > Spam or Harmful Content`, i.e. in `harmful_content` — block `gambling`, `gambling_games` and `harmful_content` to block everything gambling-shaped. | `Attractions > Casinos & Gambling`
`Shopping > Lotteries and Scratchcards`
`Sports > Poker and Professional Gambling` | | `gambling_games` | Gambling Games (simulated) | yes | 1 | `gambling` | Casino- and gambling-themed video games with no real-money play — social casino, virtual slots, poker-simulation games (`Video Gaming > Video Game Genres > Casino and Gambling Video Games`). Separate from `gambling`, where market and regulators draw the line at real money (see the design doc's market evidence). Blocking buyers who want everything casino-shaped, including real-money casinos the classifier may have stored under this genre, block `gambling` AND `gambling_games`. | `Video Gaming > Video Game Genres > Casino and Gambling Video Games` | | `piracy` | Piracy & Copyright Infringement | yes | 1 | | Sites distributing or facilitating unauthorised copies of copyrighted material — torrents, warez, cracks, illegal streaming (`Sensitive Topics > Online Piracy`). IAB v3.1 has no separate peer-to-peer node; legitimate freeware/shareware portals are `software_downloads`. | `Sensitive Topics > Online Piracy` | | `drugs` | Drugs | yes | 1 | `alcohol`, `tobacco` | IAB v3.1's single sensitive-substances node, `Sensitive Topics > Illegal Drugs, Tobacco, eCigarettes, Vaping, Alcohol`: illegal and abused drugs, and tobacco, vaping or alcohol content when the site is about the substance as sensitive content. The taxonomy does not separate the four (nor cannabis/CBD, which most vendors split out), so this group cannot either. Neutral beverage and cigar content is `alcohol` / `tobacco`; pharmaceuticals, addiction treatment and smoking cessation are `health`. | `Sensitive Topics > Illegal Drugs, Tobacco, eCigarettes, Vaping, Alcohol` | | `alcohol` | Alcohol | yes | 1 | `drugs` | Consumer alcohol content — beverage brands, breweries, wineries, wine and spirits retail, cocktail content (`Food & Drink > Alcoholic Beverages`). Bars and restaurants are `food_drink`; the sensitive framing of the same substance is `drugs`. | `Food & Drink > Alcoholic Beverages` | | `tobacco` | Tobacco | yes | 1 | `drugs` | Cigar and tobacco hobby and retail content (`Hobbies & Interests > Cigars`) — IAB v3.1's only tobacco-specific node. Vaping and e-cigarettes as sensitive content are `drugs`; smoking cessation is `health`. | `Hobbies & Interests > Cigars` | | `weapons` | Weapons | yes | 1 | `hunting_fishing` | Firearms, ammunition and weapon sales, reviews or advocacy (`Sensitive Topics > Arms & Ammunition`). Hunting and shooting sports are `hunting_fishing`; the defence industry is `business`. | `Sensitive Topics > Arms & Ammunition` | | `hate` | Hate & Discrimination | yes | 1 | | Hate speech, discrimination and acts of aggression against groups by race, ethnicity, religion, nationality, gender, sexual orientation, disability or age (`Sensitive Topics > Hate Speech and Acts of Aggression`). | `Sensitive Topics > Hate Speech and Acts of Aggression` | | `violence` | Violence & Gore | yes | 1 | | Graphic violence, death, injury and military conflict presented as sensitive content (`Sensitive Topics > Death, Injury, or Military Conflict`). Crime and disaster as news topics are `current_events`; criminal activity is `illegal_activities`. | `Sensitive Topics > Death, Injury, or Military Conflict` | | `illegal_activities` | Illegal Activities | yes | 1 | | Crime and harmful acts against individuals, society and human rights — instruction in or promotion of criminal activity, exploitation, abuse (`Sensitive Topics > Crime & Harmful Acts to Individuals, Society & Human Right Violations`). Vendors call this Illegal / Criminal Activities. | `Sensitive Topics > Crime & Harmful Acts to Individuals, Society & Human Right Violations` | | `terrorism` | Terrorism & Extremism | yes | 1 | | Terrorist and violent-extremist content, recruitment and propaganda (`Sensitive Topics > Terrorism`). | `Sensitive Topics > Terrorism` | | `harmful_content` | Spam & Harmful Content | yes | 1 | `gambling` | Spam, scams, deceptive or otherwise harmful content (`Sensitive Topics > Spam or Harmful Content`). Also where our gambling rule filter routes illegal/unlicensed gambling markers (satta/matka, Indonesian judi/slot SEO spam). A CONTENT label, not a threat verdict: IAB v3.1 has no malware, phishing, botnet or anonymizer category; threat intelligence is a separate WhoisXML product. | `Sensitive Topics > Spam or Harmful Content` | | `profanity` | Profanity & Tasteless | yes | 1 | | Obscene, vulgar, crude or profane content that is not sexual or violent (`Sensitive Topics > Obscenity and Profanity`). | `Sensitive Topics > Obscenity and Profanity` | | `social_issues` | Debated Social Issues | yes | 1 | | Contested social and moral issues — abortion, gun control, immigration, sexuality and gender debates and similar (`Sensitive Topics > Debated Sensitive Social Issues`). Vendors name Abortion / Social Issues here. Mainstream political coverage is `politics`. | `Sensitive Topics > Debated Sensitive Social Issues` | | `sensitive_other` | Sensitive (unspecified) | yes | 1 | | A label that stopped at the `Sensitive Topics` tier-1 without naming which sensitive theme applies, and — read first, via `safety_rules` — a theme-unknown `sensitive` brand-safety verdict on a row whose primary label names no more specific sensitive group. Treat as sensitive; the specific group is unknown. | `Sensitive Topics` | | `hunting_fishing` | Hunting & Fishing | | 2 | `weapons` | Sport hunting, shooting sports, gun clubs and fishing (`Sports > Hunting and Shooting`, `Sports > Fishing Sports`). Separate from `weapons`, as in Cisco, BrightCloud, Fortinet and Palo Alto; block both to cover everything firearm-related. | `Sports > Fishing Sports`
`Sports > Hunting and Shooting` | | `dating` | Dating | | 1 | | Online dating, personals and matchmaking (`Family and Relationships > Dating`). | `Family and Relationships > Dating` | | `social_networking` | Social Networking | | 1 | | Social networks and online communities (`Technology & Computing > Computing > Internet > Social Networking`). | `Technology & Computing > Computing > Internet > Social Networking` | | `search_engines` | Search Engines & Portals | | 1 | | Web search engines and portals (`Technology & Computing > Computing > Internet > Search`). | `Technology & Computing > Computing > Internet > Search` | | `communication` | Email, Messaging & Conferencing | | 3 | | Communication services and topics — web email (`… > Internet > Email`), web conferencing and online meetings (`… > Computer Software and Applications > Web Conferencing`) and the `Communication` tier-1 (messaging, telephony, postal services). | `Communication`
`Technology & Computing > Computing > Computer Software and Applications > Web Conferencing`
`Technology & Computing > Computing > Internet > Email` | | `software_downloads` | Software Downloads | | 1 | | Freeware and shareware download portals (`Technology & Computing > Computing > Computer Software and Applications > Shareware and Freeware`). Pirated software is `piracy`; software as a topic is `technology`. | `Technology & Computing > Computing > Computer Software and Applications > Shareware and Freeware` | | `artificial_intelligence` | Artificial Intelligence | | 1 | | AI and generative-AI tools, services and topics (`Technology & Computing > Artificial Intelligence`) — the node an enterprise "Generative AI" policy keys on. Robotics, AR and VR stay in `technology`. | `Technology & Computing > Artificial Intelligence` | | `technology` | Technology & Computing | | 40 | | Hardware, software, programming, networking, security, cloud, web hosting and consumer electronics — the `Technology & Computing` tree except the nodes split into `social_networking`, `search_engines`, `communication`, `software_downloads` and `artificial_intelligence`. IAB v3.1 has no hacking, proxy/anonymizer, VPN or peer-to-peer node; `Information and Network Security` is defensive security content. Includes `Web Hosting`, where hosting-provider placeholder pages have historically been over-labelled (feed changelog 2026-08-24). | `Technology & Computing` | | `games` | Games | | 26 | | Video games on any platform, esports and non-gambling game genres (`Video Gaming` except the casino and adult genres), plus tabletop, card and role-playing games (`Hobbies & Interests > Games and Puzzles`). No real-money play. | `Hobbies & Interests > Games and Puzzles`
`Video Gaming` | | `entertainment` | Entertainment | | 82 | | Movies, music, television and radio (`Entertainment`, `Genres`), celebrity and pop culture (`Pop Culture`), events and conventions (`Events`), and nightclubs (`Attractions > Nightclubs`). IAB v3.1 has no streaming-service node: a streaming platform is labelled by its content. | `Attractions > Nightclubs`
`Entertainment`
`Events`
`Genres`
`Pop Culture` | | `arts_literature` | Arts & Literature | | 14 | | Fine and performing arts (`Fine Art`) and books, comics and poetry (`Books and Literature`). Arts-and-crafts hobbies are `lifestyle`. | `Books and Literature`
`Fine Art` | | `current_events` | Current Events | | 3 | | Current-affairs topic roots: `Crime`, `Disasters` and `War and Conflicts` — the three roots policy_v1 published under the misleading id "news", renamed here. IAB v3.1 has no "news publisher" or military node — a general news site is labelled by its dominant topic, most often `politics`. WARNING — this group does NOT contain newspapers, news portals or news publishers, so a policy that allows or blocks `current_events` does NOT allow or block news sites; there is no group that does. The rename is why policy_v2 is a new pack id: a group id never changes meaning inside a pack (design §4.3), so the old id was retired with its pack rather than re-used. | `Crime`
`Disasters`
`War and Conflicts` | | `politics` | Politics & Advocacy | | 4 | | Politicians, parties, elections, civic affairs, policy and advocacy (`Politics`). | `Politics` | | `government_legal` | Government & Legal | | 1 | | Legal topics and services (`Law`). IAB v3.1 has no government-services or military node; civic affairs sit under `Politics` (→ `politics`). | `Law` | | `business` | Business | | 61 | | Companies, industries, B2B services, marketing and advertising, logistics and management (`Business and Finance` except `Economy`). IAB v3.1 has no ad-server/tracker node; the advertising nodes here are about the advertising business. | `Business and Finance` | | `finance` | Finance | | 41 | | Personal finance, banking, insurance, investing, loans and taxes (`Personal Finance`), and the economy, markets and currencies (`Business and Finance > Economy`). | `Business and Finance > Economy`
`Personal Finance` | | `careers` | Jobs & Careers | | 9 | | Job search, career advice, vocational training and remote work (`Careers`). | `Careers` | | `education` | Education | | 18 | | Schools, universities, online learning, homework and language learning (`Education`). IAB v3.1 has no cheating/plagiarism node. | `Education` | | `science` | Science | | 10 | | Natural sciences, space, geography, environment and weather (`Science`). | `Science` | | `health` | Health & Medicine | | 54 | | Medical conditions, treatments, pharmaceuticals, sexual and reproductive health, mental health, addiction treatment, fitness, nutrition and wellness (`Medical Health`, `Healthy Living`). Sex education lives here (vendor convention), not in `adult`. IAB v3.1 has no self-harm node. | `Healthy Living`
`Medical Health` | | `lifestyle` | Society & Lifestyle | | 66 | | Family and parenting (`Family and Relationships` except `Dating`), hobbies, crafts and collecting (`Hobbies & Interests` except `Cigars`, `Games and Puzzles` and `Paranormal Phenomena`), pets (`Pets`), celebrations (`Personal Celebrations & Life Events`) and holidays (`Holidays`). | `Family and Relationships`
`Hobbies & Interests`
`Holidays`
`Personal Celebrations & Life Events`
`Pets` | | `food_drink` | Food & Dining | | 13 | | Cooking, recipes, diets, restaurants and non-alcoholic beverages (`Food & Drink` except `Alcoholic Beverages`; `Attractions > Bars & Restaurants`). | `Attractions > Bars & Restaurants`
`Food & Drink` | | `travel` | Travel & Attractions | | 35 | | Travel, hotels, transport and destinations (`Travel`), and visitor attractions — theme parks, museums, parks, zoos, landmarks (`Attractions` except casinos, bars, malls and nightclubs). | `Attractions`
`Travel` | | `shopping` | Shopping | | 11 | | Retail, marketplaces, coupons, gifts and groceries (`Shopping` except `Lotteries and Scratchcards`; `Attractions > Malls & Shopping Centers`). | `Attractions > Malls & Shopping Centers`
`Shopping` | | `real_estate` | Real Estate | | 12 | | Property listings, buying, selling and renting (`Real Estate`). | `Real Estate` | | `automotive` | Vehicles | | 41 | | Cars, motorcycles, parts, repair, buying and selling (`Automotive`). | `Automotive` | | `sports` | Sports | | 66 | | All sports, teams, leagues and equipment (`Sports` except `Poker and Professional Gambling`, `Hunting and Shooting` and `Fishing Sports`), including fantasy sports and horse racing. | `Sports` | | `fashion_beauty` | Fashion & Beauty | | 44 | | Clothing, accessories, intimates, beauty and personal care (`Style & Fashion`). | `Style & Fashion` | | `home_garden` | Home & Garden | | 12 | | Home improvement, décor, gardening, appliances and smart home (`Home & Garden`). | `Home & Garden` | | `religion` | Religion | | 10 | | Religions, spirituality, atheism and agnosticism (`Religion & Spirituality` except `Astrology`). | `Religion & Spirituality` | | `cults_occult` | Cults, Occult & Astrology | | 2 | | Astrology and horoscopes (`Religion & Spirituality > Astrology`) and paranormal phenomena (`Hobbies & Interests > Paranormal Phenomena`) — the closest IAB v3.1 comes to vendors' Cult & Occult / Alternative Beliefs category. IAB has no cult or satanism node. | `Hobbies & Interests > Paranormal Phenomena`
`Religion & Spirituality > Astrology` | ## Open mapping decisions These are recorded so a change is a visible version bump, never a silent drift: 1. **Gambling scope — decided 2026-09-03 (Option C, inherited from `policy_v1` 2026.09.2).** The three real-money nodes (`Attractions > Casinos & Gambling`, `Shopping > Lotteries and Scratchcards`, `Sports > Poker and Professional Gambling`) project to `gambling`; `Video Gaming > Video Game Genres > Casino and Gambling Video Games` (simulated gambling, no real-money play) projects to the separate `gambling_games` group. Block **both** groups to block everything casino-shaped. The gambling detector's `harmful_gambling` theme (illegal/unlicensed markers) projects to `harmful_content`, not `gambling` — block all three to block everything gambling-shaped. The pack encodes this as `gambling`'s `block_with` union ([Block-with unions](#block-with-unions)), as it does the substances and weapons unions below. 2. **Substances.** IAB's single node `Illegal Drugs, Tobacco, eCigarettes, Vaping, Alcohol` cannot be split, so `drugs` covers all four substances when framed as sensitive content; neutral beverage content is `alcohol`, cigar content is `tobacco`. 3. **News — resolved by this pack.** The current-affairs roots are `current_events`. There is still no news-publisher group, because IAB v3.1 has no such node; a general news site is labelled by its dominant topic, most often `politics`. 4. **Theme-aware sensitive detection.** `sensitive_other` keeps every theme-unknown `sensitive` verdict whose primary label names no theme. Splitting it needs a detector that reports the theme; when one ships, its themes join `safety_rules` in a new mapping version. ## Changelog - **2026.09.1** (2026-09-09) — First release of the `policy_v2` id. `previous_version: null`; nothing has moved _inside_ this pack. - **Lineage** (`previous_pack`): branched from `policy_v1` **2026.09.2**. `moved`: `Crime`, `Disasters`, `War and Conflicts` — each from `news` to `current_events`. `added_groups`: `current_events`. `retired_groups`: `news`. No other path changed group. - Adds the `safety_rules` table (`adult` → `adult`, `gambling` → `gambling`, `harmful_gambling` → `harmful_content`, `sensitive` → `sensitive_other`, `vendor` → no group) and the safety-first projection with the `basis` values `safety` / `site_purpose` / `primary` (`role` reserved). - Declares the three `block_with` unions (`gambling` ↔ `gambling_games`, `harmful_content`; `drugs` ↔ `alcohol`, `tobacco`; `weapons` ↔ `hunting_fishing`), served on `/api/v4/taxonomy?pack=policy_v2` as `pack.groups`. Added to this first version before its first publication — no feed artifact and no deployed API image carried 2026.09.1 without them — so they ride the version unbumped; a later union change bumps it. - **What it moves, in rows.** Every row `policy_v1` published as `news` reads `current_events`; on the feed, every flagged row whose primary label is not already in a different sensitive group reads its theme's group with `basis = safety`. The API serves the pack (primary-label projection) on `?pack=policy_v2` from the image that carries it (see the API changelog; an earlier image answers `422` for the id); the feed with the first full snapshot exported under schema 1.14.0 (a pack change never rides a delta) — see the feed changelog for the dates.