Search systems
Google search systems, sorted by how much we actually know
59 systems, algorithms and retrieval theories. Each page carries one evidence label, verbatim quotes from primary sources, and a linked audit check you can run on your own site. Where a check can only measure a proxy for the system, the check says so.
Dated updates from Florida (2003) to 2026: the Google update timeline
Which systems has Google confirmed?
Named by Google itself: the Search Central ranking systems guide, Google blog posts, or on-record statements by Google staff. Includes systems Google has since retired and the infrastructure that crawls, renders and indexes the web.
RankBrain
RankBrain is a machine learning system Google uses to understand how words relate to concepts, so it can rank pages that are relevant to a search even when they do not contain every word in it. Google confirms it in its ranking systems guide and describes it as its first deep learning system in Search. It is not something a site can optimise for directly. What a site can do is make sure each page clearly covers the concept behind the searches it is shown for.
BERT
BERT is a transformer based language model, published by Google researchers in 2018, that Google Search uses to understand how combinations of words express meaning and intent. Google confirms it in its ranking systems guide and said in 2020 that BERT was used in almost every English query, for both ranking and retrieval. It rewards nothing a site can add as markup. What it changes is that words like to, for and no now count, so pages must answer the query as actually phrased.
MUM (Multitask Unified Model)
MUM, the Multitask Unified Model, is a Google AI model announced in 2021 that can understand and generate language across 75 languages and more than one format. Google confirms it exists and is used in Search, but states plainly that it is not currently used for general ranking. Its confirmed uses are narrow: COVID-19 vaccine searches, detecting personal crisis searches, spam protection and featured snippet callouts. Sites should not optimise for MUM as a ranking system.
Neural matching
Neural matching is an AI system Google uses to understand the concepts represented in queries and pages and match them to one another. Google introduced it in 2018, confirms it in its ranking systems guide, and describes it as a critical part of how it retrieves candidate documents. It means a page can be found for searches phrased differently from the page. Sites cannot optimise for it directly, but pages written only in internal jargon give it less to work with.
Passage ranking
Passage ranking is an AI system Google uses to identify individual sections of a page to better understand how relevant the page is to a search. Google confirms it in its ranking systems guide. It launched for US English queries in February 2021. Despite its early name, passage indexing, Google does not index passages separately; it still indexes and ranks pages. For sites, it means a specific answer deep in a long page can help that page rank, if the passage is clear.
Hummingbird
Hummingbird was a major rewrite of Google's overall ranking systems, in use from August 2013 and announced in September 2013. It was designed to pay more attention to the meaning of the whole query rather than individual words. Google now lists it under retired systems in its ranking systems guide, noting its systems have evolved since. There is nothing to optimise for today, but the shift it started, from keywords to meaning, runs through RankBrain, BERT and neural matching.
Knowledge Graph
The Knowledge Graph is Google's database of facts about people, places, organisations and things, launched in May 2012. Google confirms it and says it held over 500 billion facts about five billion entities by 2020. It is infrastructure rather than a ranking system: it powers knowledge panels, answers factual questions and is named as a data source for AI Mode. For a site, the practical questions are whether its organisation is a recognised entity and whether the facts Google shows about it are correct.
AI Mode and query fan-out
AI Mode is Google's AI search experience, launched in Labs in March 2025 and rolled out in the US in May 2025. Google confirms it uses a query fan-out technique: it breaks a question into subtopics and issues many related searches at once, then combines the results into one response with links. Google says there are no extra technical requirements beyond being indexed and eligible for a snippet. For sites, fan-out means visibility depends on answering the subquestions behind a topic, not only the head query.
Link analysis systems and PageRank
Link analysis systems are the Google ranking systems that read how pages link to each other to work out what pages are about and which are most useful for a query. PageRank is the best known of them. Google confirms both in its ranking systems guide and says PageRank has evolved a lot but remains part of its core ranking systems. For a site, this means crawlable links, descriptive anchor text and an internal link graph that points weight at the pages that matter are still worth auditing.
Penguin
Penguin was a Google system designed to combat link spam. Google announced it in April 2012 and integrated it into its core ranking systems in September 2016, when it became real-time and started devaluing spam links rather than demoting whole sites. Google now lists Penguin among retired systems. Its job continues through link spam handling in core ranking and SpamBrain, so sites should audit for manipulative links and over-optimised anchors, not for a Penguin penalty.
SpamBrain and spam detection systems
SpamBrain is Google's AI-based spam-prevention system and the named part of its spam detection systems. Google says it launched in 2018 and is regularly improved through spam updates. Since the December 2022 link spam update it also detects sites buying links and sites used to pass outgoing links, and neutralises those links. For a site, the practical consequences are that unqualified paid or affiliate links, scaled low-value content and hacked pages are all within its scope.
Site diversity system
The site diversity system is a confirmed Google system that generally shows no more than two web listings from the same site in the top results, unless its systems judge more to be especially relevant. Subdomains are usually treated as part of the root domain. Google launched it in June 2019 and said it is about display, not ranking. For a site, this means several pages targeting the same query compete for two slots, so consolidating overlapping pages matters more than publishing more of them.
Panda
Panda was a Google ranking system announced in February 2011 to reduce rankings for low-quality sites and reward sites with original, useful content. Google confirms it by name in its ranking systems guide, which lists it as retired: it became part of the core ranking systems in 2015. There is no separate Panda update to recover from today. The practical lesson still holds, because Google said low-quality content on part of a site can affect the whole site's rankings, so thin and duplicated sections remain a sitewide risk.
Helpful content system
The helpful content system was a Google ranking system launched in August 2022 to demote content made mainly to attract search traffic rather than to help people. Google confirms it in its ranking systems guide, which lists it as retired: in March 2024 it became part of the core ranking systems, and Google says no single signal now does this job. For a site, the practical point is unchanged. Sections written for search demand outside the site's real purpose and audience are the pattern Google describes as unhelpful.
Reviews system
The reviews system is a confirmed Google ranking system that aims to reward reviews offering insightful analysis and original research by experts or enthusiasts. It began as the product reviews update in April 2021 and was broadened to reviews of any topic in 2023. Google says it evaluates first-party review content, mainly page by page, but can assess a whole site with substantial review content. For a site, this means review and best-of pages need visible first-hand evidence, not rewritten manufacturer specifications.
Original content systems
Original content systems are confirmed Google ranking systems that aim to show original content, including original reporting, ahead of pages that merely cite it. Google's ranking systems guide names them and notes support for canonical markup that identifies the primary page when content is duplicated in several places. In 2019 Google said it had changed its products to keep significant original reporting visible for longer. For a site, the useful action is making sure copies of its work elsewhere point back to the original.
Medic update
The Medic update is the industry nickname for Google's broad core algorithm update confirmed on 1 August 2018. Google confirmed a broad core update but did not name it Medic: Barry Schwartz of Search Engine Roundtable coined the name after a survey in which health and medical sites were heavily represented among those affected. Google said the update involved all searches and that there was nothing specific to fix. For sites, the lasting point is that health and money topics are where Google says expertise and trust carry more weight.
Core updates
Core updates are significant, broad changes that Google makes to its search algorithms and systems several times a year and announces on its Search Status Dashboard. Google confirms them and says they do not target specific sites or pages. It also documents how to assess a drop: wait a week after the rollout ends, compare equal periods in Search Console, separate small drops from large ones, and only for large, sustained drops review the site against its helpful content self-assessment. Quick fixes are discouraged and recovery can take months.
Reliable information systems
Reliable information systems are the confirmed Google ranking systems that aim to surface authoritative pages, demote low-quality content and elevate quality journalism. When reliable information is lacking, Google shows content advisories on rapidly changing topics or when it lacks confidence in the overall quality of results. Google says it places even more emphasis on expertise and trustworthiness for health, finance, civic and crisis topics. For a site, the practical test is whether its claims are sourced, current and consistent with authoritative consensus.
Crisis information systems
Crisis information systems are confirmed Google systems that show helpful, timely information during crises. For personal crises, Google detects searches about suicide, sexual assault, poison ingestion, gender-based violence or drug addiction and shows hotlines and content from trusted organisations. For natural disasters and wide-spread emergencies, SOS Alerts show updates from local, national or international authorities. Google said in 2022 it uses MUM to detect a wider range of personal crisis searches. Sites cannot rank into these features, but publishers on these topics can signpost help clearly.
Page experience system
Page experience is Google's label for how users perceive interacting with a page: Core Web Vitals, secure delivery, mobile display, ads and interstitials. Google confirms that its core ranking systems look to reward good page experience, but since 2023 it has said this was never a separate ranking system or a single combined signal, and since 2024 its documentation says only Core Web Vitals are used directly in ranking. For a site, page experience is a tie-breaker when relevance is close, measured on real-user field data.
Page speed system
The page speed system was Google's use of page load speed as a ranking signal: first for desktop searches in 2010, then for mobile searches through the July 2018 Speed Update. Google confirmed both, and described both as small, affecting only the slowest pages or a small share of queries. It was removed from Google's ranking systems guide in April 2023. Speed still matters through Core Web Vitals, which Google says its ranking systems use, so the practical job is fixing the slowest templates on mobile field data.
Mobile-friendly ranking system
The mobile-friendly ranking system was Google's use of mobile-friendliness as a ranking signal for mobile searches, expanded worldwide from 21 April 2015 and nicknamed Mobilegeddon by the industry. Google confirmed it. In April 2023 it was removed from the ranking systems guide, and Google retired the Mobile-Friendly Test and Mobile Usability report from December 2023. Google now says aspects of page experience other than Core Web Vitals do not directly help rankings, so mobile display is a user requirement first.
Secure sites system (HTTPS)
The secure sites system was Google's use of HTTPS as a ranking signal, announced in August 2014 as a very lightweight signal affecting fewer than 1% of global queries. Google confirmed it and later listed HTTPS as a page experience signal. In April 2023 the secure sites system was removed from Google's ranking systems guide, and Google now says page experience aspects other than Core Web Vitals do not directly help rankings. HTTPS remains a baseline for users, browsers and clean site migrations rather than a ranking lever.
Mobile-first indexing
Mobile-first indexing means Google uses the mobile version of a site's content, crawled with the smartphone agent, for indexing and ranking. Google confirmed it, began experiments in 2016, declared the migration complete in October 2023 and stopped crawling the remaining sites with desktop Googlebot after 5 July 2024. It is infrastructure, not a ranking boost. For a site, anything missing from the mobile version (text, links, images, structured data, robots directives) is missing from Google's index.
Web Rendering Service (WRS)
The Web Rendering Service is the Googlebot component that runs a page's JavaScript in a headless, evergreen version of Chromium so Google can index the rendered HTML. Google confirms it by name in its JavaScript documentation and announced the evergreen Chromium change in May 2019. It is infrastructure, not a ranking system. For a site, it means content injected by JavaScript can be indexed, but only if rendering succeeds, resources are crawlable, and the WRS is not served stale cached scripts.
Caffeine
Caffeine is the web indexing system Google completed in June 2010. It replaced a layered, batch-refreshed index with one that updates continuously as pages are crawled, and Google said it delivered 50 percent fresher results. A Google paper presented at OSDI 2010 describes the Percolator system underneath it. Caffeine is infrastructure, not a ranking signal. For a site, it means new and changed pages can reach the index quickly, provided Google can discover and crawl them.
Freshness systems
Freshness systems are Google's query deserves freshness systems: they decide when a query needs recent results and then favour fresher content. Google confirms them in its ranking systems guide. The idea was reported as QDF in 2007, and Google's November 2011 freshness update, built on the Caffeine index, affected roughly 35 percent of searches. Freshness applies per query, not per site. For a site, it rewards genuinely updated content on time-sensitive topics and clear, consistent dates, not date changes without substance.
AI Overviews
AI Overviews are Google's generated summaries shown at the top of some search results, with links to supporting web pages. Google tested them as the Search Generative Experience in Search Labs from May 2023 and launched them to everyone in the US in May 2024. Google says there are no special requirements to appear: a page must be indexed and eligible to show with a snippet. For a site, visibility depends on normal Search eligibility plus passages that directly answer the query.
Deduplication systems
Deduplication systems are Google's confirmed ranking systems that stop near identical pages from filling the results. When several pages are very similar, Google shows the most relevant one, and since January 2020 a page shown as a featured snippet is not repeated on the first page. Canonicalisation is the related indexing step that picks one URL from a set of duplicates. For a site, duplicate URLs compete with each other and only one is likely to be shown, so consolidate them and signal a clear canonical.
Exact match domain system
The exact match domain system is a confirmed Google ranking system that limits the credit given to domains built to match a query, such as a domain made of the words of a search. It traces to the EMD update announced by Matt Cutts on 28 September 2012, which he said affected 0.6% of English US queries. Words in a domain remain one minor relevance factor. For a site, a keyword domain is not a ranking shortcut, and a thin site on one gains nothing from the name.
Local news systems
Local news systems are confirmed Google ranking systems that identify local sources of news and surface them when relevant, through features such as Top stories and Local news. Google announced a local news carousel in Search in November 2021 and said in 2022 that authoritative local sources now appear more often alongside national publications. Google has not published which signals mark a source as local. For publishers, clear transparency about who you are, where you report and when, plus consistent local coverage, are the inputs you control.
Removal-based demotion systems
Removal-based demotion systems are confirmed Google ranking systems that use high volumes of valid removals against a site as a signal to demote its other content. They cover legal removals, such as copyright, defamation, counterfeit goods and court orders, and personal information removals from sites with exploitative removal practices. Google began using copyright notices as a ranking signal in August 2012 and extended protections against exploitative sites in June 2021. For a site, a pattern of valid removals can pull down pages that were never the subject of a complaint.
Site reputation abuse
Site reputation abuse is a Google spam policy, announced in March 2024 and enforced from May 2024, against third-party content published on an established site mainly to benefit from that site's ranking signals. Google confirms it in its spam policies, which since August 2026 call it the site reputation policy. Outside the European Economic Area a violation can bring a manual action on the affected section; inside the EEA that section may instead be separated and ranked on its own merits. Sites hosting partner, coupon or affiliate sections need visible editorial integration.
Scaled content abuse
Scaled content abuse is a Google spam policy, introduced in March 2024, against generating many pages mainly to manipulate rankings rather than help users. Google confirms it in its spam policies and says it applies whether pages are produced by generative AI, scraping, templates or people. It replaced the narrower automatically generated content policy. For a site, the risk sits in any large page set where individual pages add little beyond a template, a feed or a model output.
Expired domain abuse
Expired domain abuse is a Google spam policy, introduced in March 2024, against buying an expired domain and repurposing it mainly to rank low-value content on the strength of its previous owner's reputation. Google confirms it in its spam policies and says using an old domain for a new, original, people-first site is fine. For an auditor, the signal is a mismatch between a domain's history and backlinks and what it publishes now.
Which systems are known from the DOJ trial or the 2024 leak?
Known from sworn testimony and exhibits in United States v. Google, Judge Mehta’s opinions, or the May 2024 Content Warehouse API documentation leak. Google has not described these publicly, so each page says exactly which evidence it rests on.
Navboost
Navboost is a Google ranking signal that memorises which results users clicked for which queries. Google has never documented it publicly; it is known from sworn testimony and exhibits in United States v. Google, Judge Mehta's 2024 opinion, and the May 2024 Content Warehouse leak. Testimony puts its training window at 13 months of click data since 2017. For a site, it means the way a result earns and keeps clicks after it ranks is plausibly part of ranking, not only a traffic outcome.
Glue
Glue is a Google system that records how users interact with the whole results page, not only the ten blue links, and feeds that into decisions about search features. It is known from sworn trial testimony, Judge Mehta's 2025 remedies opinion and the 2024 Content Warehouse leak; Google does not document it publicly. Pandu Nayak called it Navboost for all of the other features on the page. For sites, it means engagement with images, video and other features is plausibly recorded, not only clicks on web results.
RankEmbed and RankEmbedBERT
RankEmbed is a Google deep learning model that places queries and documents in the same embedding space so it can retrieve and rank by meaning rather than exact words; RankEmbedBERT is its later version. It is known from trial testimony, Judge Mehta's 2024 and 2025 opinions and a leaked module name, not from Google's public documentation. For sites, it means pages can be retrieved for queries they never state verbatim, so clear topical meaning matters more than keyword repetition.
DeepRank
DeepRank is Google's name for BERT when it is applied to ranking, according to Pandu Nayak's sworn testimony in the US antitrust trial. It is a transformer model trained partly on user data that re-scores a small set of top results for language understanding. Google has publicly announced BERT in Search but, as far as we can verify, not the DeepRank name. For sites, it means the meaning of every word in a query, including small words, can decide whether a page is the best answer.
QBST (Query-based Salient Terms)
QBST, Query-based Salient Terms, is a Google ranking signal that learns which words and word pairs should appear prominently on pages relevant to a given query. It is known from Eric Lehman's trial testimony as summarised in Judge Mehta's 2024 opinion, and a QBST feature appears in the 2024 leak; Google does not document it publicly. It is trained on about 13 months of user data. For sites, it means the terms that consistently characterise good answers to a query are a real relevance input.
Twiddlers
Twiddlers are Google functions that re-rank a set of results after the main scoring system has selected them. They are known from the May 2024 Content Warehouse leak, which names twiddlers for freshness, site boosting, deduplication and spam sandboxing, and from 2025 remedies phase call notes with Pandu Nayak. Google does not document them publicly. For sites, it means date signals, duplication and spam indicators can change a result's position late in the process, after relevance scoring.
Tangram (formerly Tetris)
Tangram, formerly called Tetris, is the Google system that decides which search features appear on a results page and where, using Glue and other signals. It is known only from Pandu Nayak's October 2023 trial testimony and 2025 remedies phase notes of a call with Google engineer HJ Kim; Google does not document it publicly and it is not visible in the leak. For sites, it means feature visibility is its own ranking decision, separate from blue link ranking.
ABC signals (Anchors, Body, Clicks)
ABC signals are Anchors, Body and Clicks: links pointing to a page, the terms on the page, and how users behave after clicking it. Notes of a February 2025 call with Google engineer HJ Kim, filed in the US antitrust remedies phase, call them the three fundamental signals and the key components of topicality, T*. Google does not document them publicly. For sites, it means anchor text, body copy and post-click satisfaction should describe the same topic.
T* (Topicality)
T* (Topicality) is the name a Google engineer used, in a call summarised by the US Department of Justice and filed as a trial exhibit in 2025, for Google's base score of how relevant a document is to a query. It combines three signals, known as the ABC signals: anchors, body and clicks. Google has not documented T* publicly. For a site, it means relevance is judged from the words on the page, the words others use to link to it, and how searchers respond to it.
Q* (Quality score)
Q* (Q star) is Google's measure of the quality of a document, described in US Department of Justice records of 2025 calls with Google engineers that were filed as exhibits in the search antitrust case. The records describe it as largely static and related to the site rather than the query, with PageRank as one input. The 2024 Content Warehouse leak names several attributes as applied in Qstar. Google has not documented Q* publicly. For a site, it means quality is judged largely at site level and changes slowly.
Site authority (siteAuthority)
siteAuthority is an attribute in the Google Content Warehouse API documentation leaked in May 2024, described as converted from a quality signal and applied in Qstar, Google's quality score. Googlers had said for years that Google does not have an overall domain authority score. The two can both be true: siteAuthority is a quality input, not a third-party style link metric. Google has not confirmed how it is used. For a site, it means reputation accrues at site level and new hosts start from little.
Which ideas come only from Google patents?
Described in a Google patent. A patent shows what Google engineers designed, not what runs in production, and Google has not confirmed using any of these.
Reasonable surfer model
The reasonable surfer model is described in Google patent US7716225, filed in 2004 by Jeffrey Dean, Corin Anderson and Alexis Battle. It replaces PageRank's random surfer, who clicks any link with equal probability, with one who is more likely to follow prominent, relevant links than footer or advert links. Google has not confirmed it uses this model. The practical lesson is still testable: put links to important pages where readers will actually use them, not only in boilerplate.
Site quality score patent
The site quality score patent (US9031929B1), granted to Google in 2015 with Navneet Panda and April R. Lehman as inventors, describes scoring a site as the ratio of queries that refer to the site by name to queries that lead to clicks on the site's pages. A related Panda patent predicts site quality from phrase use. Google has not confirmed either is used in ranking. For a site, the practical reading is that branded demand relative to generic visibility is worth tracking.
Information gain patent
Google's patent 'Contextual estimation of link information gain' (US11354342B2, granted 2022) describes an information gain score: how much new information a document offers beyond documents the user has already viewed, predicted by a machine learning model. It is framed around an automated assistant and a user's own reading history. Google has not confirmed it is used in web ranking. For a site, the useful lesson is to make each page add something the reader has not already seen.
Historical data patent
'Information retrieval based on historical data' (US7346839B2) is a Google patent filed in 2003 and granted in 2008, naming engineers including Matt Cutts, Jeff Dean and Paul Haahr. It describes scoring documents using history: when a document first appeared, how its content changes, and how links to it grow or spike. Google has not confirmed which parts, if any, are used. For a site, it is the origin of many freshness and link velocity ideas, and a reminder that cosmetic date changes are easy to discount.
Phrase-based indexing
Phrase-based indexing is a family of Google patents by Anna Lynn Patterson, with a 2004 priority date, describing an index built from meaningful phrases rather than single words. Phrases are judged good if they predict the presence of other phrases, and documents are ranked and checked for spam by the related phrases they contain. Google has not confirmed production use. For a site, the practical reading is that topical depth shows up as the related phrases a subject naturally involves, and that stuffing them is a documented spam pattern.
Which are general information retrieval theory?
Information retrieval concepts that are not Google specific, such as TF-IDF and BM25, often credited to Google in SEO writing without evidence. Useful for understanding search, not proof of how Google ranks.
Topic-sensitive PageRank
Topic-sensitive PageRank is a 2002 research method by Taher Haveliwala at Stanford that computes a separate PageRank score for each of a set of topics, then blends them by how well a query matches each topic. It is general information retrieval theory. Google acquired Haveliwala's startup Kaltix in 2003 and holds later personalisation patents by the same team, but has never confirmed using topic-sensitive PageRank. The useful takeaway for sites is to earn links from pages about your topic.
TrustRank
TrustRank is a link analysis method from a 2004 paper by Zoltán Gyöngyi and Hector Garcia-Molina of Stanford and Jan Pedersen of Yahoo. It starts from a small set of human-reviewed trustworthy seed pages and propagates trust through links, so pages far from the seeds are more likely to be spam. It is not a Google system. Google filed a trademark for the name for an anti-phishing filter, and Matt Cutts said Google has nothing specifically called trust rank. Proximity to reputable sites is still a sensible link audit.
Hilltop algorithm
Hilltop is a ranking method from Krishna Bharat and George Mihaila, developed at Compaq's Systems Research Center and published at WWW10 in 2001. It ranks pages by links from independent expert pages, meaning topic directories that link to many unaffiliated sources. Bharat later joined Google, but the Hilltop patent is held by HP's successors, not Google, and Google has never confirmed using Hilltop. The durable lesson is that links from independent, topical resource pages carry more meaning than links from related sites.
TF-IDF
TF-IDF (term frequency, inverse document frequency) is a classic information retrieval weighting that scores a word highly when it is frequent in one document but rare across a collection. It underpins the vector space model, where documents and queries are compared as vectors. Google has never confirmed using TF-IDF as a ranking factor, and John Mueller has called it a fairly old metric. For a site, the practical lesson is to cover a topic in its own natural vocabulary rather than tuning copy to a TF-IDF tool score.
Okapi BM25
Okapi BM25 is a ranking function from the probabilistic relevance framework developed by Stephen Robertson, Karen Spärck Jones and others. It scores documents on query term matches, with term frequency saturation and document length normalisation, and its BM25F variant weights fields such as title, body and anchor text. Google has never said it uses BM25, and the US v. Google opinions we reviewed do not name it. For sites, the lesson is that lexical matching still matters: the words a searcher uses should appear in the title, headings and body.
Latent semantic indexing (LSI)
Latent semantic indexing is the information retrieval name for latent semantic analysis, a technique that uses singular value decomposition on a term and document matrix to group words and documents by shared concepts. It dates from the late 1980s. Google has never said it uses LSI, and John Mueller has said on record that there is no such thing as LSI keywords. For sites, lists of so called LSI keywords are at best harmless and at worst keyword stuffing; covering the topic properly is what matters.
Word embeddings and word2vec
Word embeddings represent words as dense vectors so that words with similar meanings sit close together. Google researchers popularised the approach with word2vec in 2013. Google has not said word2vec itself runs in Search ranking, but it confirms machine learning systems that match concepts rather than exact words, such as RankBrain and neural matching, and US v. Google evidence describes an embedding based ranking model called RankEmbed. For sites, pages are matched on meaning, so clear coverage of the concept beats repeating one phrasing.
The field notes
Frameworks, builds and what is changing in AI search.
Sent occasionally, never noise. The thinking behind the work, and the experiments before they ship.
No spam. Unsubscribe anytime.