--- type: page id: privacy lang: en status: unverified canonical: https://tomola.org/privacy original: https://tomola.org/privacy.md --- # Privacy notice (trial edition) Last updated: November 21, 2026 This page explains how your information is handled during the tomola trial. Japanese is the original. Other language editions are translations for reference. ## 1. Operator tomola is operated by the general incorporated association AI Alignment Network (ALIGN). - Address: 405 Yamato Building, 1-6-16 Kandaizumicho, Chiyoda-ku, Tokyo, Japan - Representative Director: Koichi Takahashi - Contact: privacy@tomola.org for requests and complaints about your information; contact@tomola.org for other enquiries. ## 2. Information collected and its purposes The information tomola handles and its purposes are listed below. We do not use it for other purposes. - **Email address**: signing in, verifying your identity, notification emails and replies from the contact desk. - **Username, introduction, language settings and country of residence**: presentation in the forum, translation, notification language and checking eligibility for the trial. - **Display names (optional)**: the forms in your own language and in Latin letters are used to display your name as an author. On public posts, they are also visible to people who are not signed in. - **Avatar (your own image)**: displayed to identify you as an author. On public posts, it is also visible to people who are not signed in. - **Posts, replies, drafts, images and chat messages**: providing and displaying the forum, translation, search, making content available for your AI to read, and handling reports and moderation. - **How much drafts from AI tools are edited**: for drafts created from AI tools (the API and MCP), we measure how much they were edited before publication, in aggregate rather than per member, to improve how tomola is run. - **Stances, Insightful marks, follows, Read later and signatures**: providing the forum's features. Stances are counted without showing who chose them. The operator does not look up who chose which stance. Read later is visible only to you. - **When you last viewed each list**: to mark posts added since you last looked at the home page, a language's list or a village's page, we remember, for each list, the creation time of the newest post shown when you last viewed it. For signed-in members we save this with the account, so the marks are the same on every device. - **Article reads**: We count site-wide daily totals by original language, reading language, source category (such as search, social networks or shared links) and reading method (such as browser, Markdown or AI tools). These site-wide totals have no fixed retention period. We also count reads for each public article by day, reading method, and whether the original or a translation was shown. We do not link these article totals to readers’ accounts or devices. We use them to review site operations and to show counts to eligible signed-in members on article pages. - **What we do not measure**: We do not use tracking cookies, stored IP addresses or device fingerprints to follow people who only read, and we do not count unique visitors. Existing authentication and abuse-prevention mechanisms are separate from reading counts. We do not use external analytics or advertising pixels. We do not record time spent reading, scrolling, completion or search terms. Our servers do not build a history of who read which articles, including for members. Article-level counts are unavailable to signed-out readers, and view counts do not affect the order of article lists. Operators can review article-level counts and reading trends to monitor site operations. - **Browser storage**: To reduce repeated notices from the same tab, the browser stores the article, displayed language, whether it showed the original or a translation, and an expiry time in `sessionStorage`. Each record expires after 30 minutes. The browser does not send those stored records to the server. - **Reports, appeals and moderation records**: running the forum safely and reviewing appeals. - **AI preliminary checks and spam detection**: AI may read reported posts and messages and the surrounding content to prepare priorities and an assessment. AI may also detect spam in public posts. Chat spam is detected mechanically from message frequency and repetition, without sending its text to AI (sections 3 and 9). - **Nushi**: When you call Nushi, we send the calling post and the selected public posts from that village to Google’s Gemini API to generate an answer. We save the answer as a public AI reply and show it to people who can read the topic. If you report an error, we save the question, answer and consulted material for the operator’s investigation only after you agree. Error reports are normally deleted after 90 days, subject to an applicable retention hold. If a supporting post is corrected or withdrawn, we stop displaying the answer and its translations. When you call Nushi in a village chat, we send your calling message and the selected public posts from that village. Its reply target and surrounding messages are not sent. The answer is saved as a message in that chat and shown to people who can read the chat. If you delete the calling message, or a supporting post is corrected or withdrawn, we stop displaying the answer. Reports about answers in chat are handled like reports about chat messages. If a village owner sets a speaking style for Nushi, we also send that text when Nushi is called. The speaking-style text and its change history (when, who, and what it was changed to) are public. - **Sign-in state, passkeys, access records (including IP addresses) and rate-limit records**: verifying your identity, preventing fraud and abuse, and investigating failures. We do not use your information for: - Advertising. - Selling information. - Training AI. tomola does not use your posts or messages to train AI. Translation providers also commit not to use it for training (section 3). - Research. Before using information for research, we will identify the researchers and the research and ask for your agreement separately. ## 3. Providers abroad that handle information tomola uses the following providers to handle information to the extent needed to run the forum. Information is handled mainly in the United States. We may change the AI providers we use for translation, converting text for search, Nushi answers, village address suggestions, first checks of reports, and spam detection, depending on circumstances. We use only providers that have committed not to use what is sent to train AI and that process it for the operator under contract. Before we change a provider, we update this page and notify you when you sign in. - **Google** - Purposes: translating posts and chat, reading and translating text in figures, converting text for search, generating Nushi answers, suggesting village addresses, preliminary report checks and detecting spam in public posts (Gemini API). - Information sent: titles and text of public posts and replies; chat messages from people who agreed to the chat notice, including the two preceding messages and, for a reply, the message it replies to (unless deleted or hidden), as context for translating other people's messages; search queries; when you call Nushi, the calling post or chat message, the selected public posts from that village and the Nushi speaking style set by the village owner; when you type a village name containing non-Latin letters (such as kanji, kana or Hangul) on the village creation page, that name, to suggest an English address, even if you do not go on to create the village; and reported posts or messages and surrounding content for preliminary checks. Email addresses are not sent. - To translate text in figures, we send images in public posts and replies (images their authors uploaded themselves). We read the text and its positions in each image and translate that text. Translated figures are drawn on this site's server (no image-generating AI is used). - For spam assessment, the titles and text of suspicious public posts and replies may be sent. - When a report is made, the reported post, reply or village description (including hidden content), surrounding posts, and the reporter's written explanation may be sent for preliminary checks. Only chat messages written by people who have given consent are sent. - Timing and method: when you write, when translations are created and when you search, through encrypted connections. - Retention: Google commits not to use what is sent to train AI. It may retain it for a limited period to monitor abuse. Google's terms do not specify a number of days. - Processing country: not specified in the terms; it is expected to be mainly the United States. - **Postmark** (AC PM, LLC, United States) - Purpose: sending email from tomola. - Information sent: email addresses and email text, including sign-in codes, invitations and notifications. - **Cloudflare, Inc.** (101 Townsend St, San Francisco, CA 94107, USA) - Purposes: relaying site traffic and forwarding contact emails. - Information passing through: communications between you and the site, and emails sent to the contact desk. - **Device notification delivery services** (Apple, Google, Mozilla or Microsoft, whichever your browser uses) - Purpose: delivering notifications to the devices of people who have turned on device notifications. Nothing is sent for people who have not turned them on. - What is sent: the delivery identifier, the encrypted notification, the sender, the time, the number of notifications and their size. A notification shows only the kind of notice; it does not include the body or title of a post, or the author's name or handle. - Withdrawal: notifications a delivery service has already accepted cannot be withdrawn. - **Google (Google Workspace)** - Purpose: the contact desk's email inbox. - Information passing through: emails sent to the contact desk. - Processing country: it may be handled in the United States and other countries where Google has facilities. tomola's database is held in Japan on a machine managed by the operator. For the US personal information protection system, see the summary by Japan's Personal Information Protection Commission (https://www.ppc.go.jp/enforcement/infoprovision/laws/offshore_report_america/). We use these providers under their standard agreements. Each agreement includes data processing terms under which the provider handles personal information on our behalf as a processor. For the Gemini API, this is Google's "Data Processing Addendum for Products Where Google is a Data Processor"; for Google Workspace, the "Cloud Data Processing Addendum"; for Postmark, its "Data Processing Addendum"; and for Cloudflare, the "Cloudflare Data Processing Addendum". Under their own terms, the providers also use some records themselves: Google keeps what we send to the Gemini API for a limited period to detect misuse, Postmark may use message content to improve its service, and Cloudflare may use traffic data to run and improve its services. At least once a year, we check that these agreements are being kept and whether the laws of the destination countries have changed. We will share the results on request. You cannot opt out of sending information to these providers, because translation and email are core features. If you do not want information sent, read without posting or chatting, or request to leave through the contact desk in section 6. Device notification delivery services are used only if you turn on device notifications. ## 4. Retention and deletion - **Posts and replies you delete**: immediately hidden from everyone, including their translations. Their text, previous versions and mention snapshots are removed from records 30 days after deletion, to allow responses to enquiries about accidental deletion. After that, only a deleted-post marker remains to preserve reply connections. - **Previous versions of published posts and replies**: when you edit a post, the version before the edit is kept, and anyone who can read the post can see its previous versions and the changes (in the original language; no translations are made). Edits made within 30 minutes of publishing, before any reply or Insightful reaction, are folded into the first version and do not create a version. Edits made while the post, or the post a reply belongs to, is hidden, or while a report on the post is open, always create a version. You can delete your previous versions. The title, text and translations of a deleted version are immediately hidden from everyone and removed from records; only its date and a “Deleted” marker remain. Passages quoted in a report about a translation stay in the report records until 30 days after the post itself is deleted. You cannot delete previous versions while the post, or the post a reply belongs to, is hidden, or while an appeal is under review. The version from before a moderator’s correction cannot be deleted until any appeal has been decided, and is kept if you do not appeal. - **Text read from figures, its translations and translated figures**: stored in this site's database. Translations that cannot be drawn inside a figure are shown as text below it. When an original image is deleted, the translated text and figures made from it are deleted too. Reading results shared by images with the same content are deleted when every image using them has been deleted. When a reading is uncertain, the original figure is shown, but the text that was read is still stored. - **Posts hidden by moderators**: kept while hidden, for appeals and reversal of moderation. - **Evidence copies of display names and avatars (including village icons)**: names and images saved before moderation removes them, and copies saved when they are reported, are deleted after 180 days. If an appeal is still open, we do not delete the copies until it ends. If you leave, they are deleted immediately. - **Chat messages**: kept until you delete them. A copy of a reported message, and only that message, is kept for 180 days after the report is handled, even if its author deletes it, for reviewing appeals and checking repeated conduct. It is then deleted. Surrounding messages and the reported message's reply target are not copied; only messages still present when accessed are read. - **Records of operator access to chat**: one year. - **Notifications**: read notifications are deleted after 90 days; unread notifications after 180 days. - **Device notification registrations**: we keep the delivery address and keys, the estimated browser and OS names, the registration date and the delivery status. When you turn notifications off, we delete the delivery address and keys immediately (the device label and registration date remain until you leave). When you leave, the registrations are deleted. - **When you last viewed each list**: one per list; we keep the 50 most recent and delete older ones. Deleted immediately when you leave. - **Sign-in state**: deleted on expiry. The lifetime is 30 days, with an absolute maximum of 90 days even with continued use. - **Sign-in email and code records, and rate-limit records**: deleted on expiry. - **Invitations**: the recipient email address is deleted 30 days after an unused invitation expires. - **Nushi error reports**: the question, answer and consulted material saved with your agreement are normally deleted after 90 days. - **AI usage records**: 13 months. Information identifying a member is removed after 90 days. - **Per-article reading counts**: kept while the public article remains available. When an author deletes their own article, its per-article reading counts are deleted immediately. For articles hidden by moderators, counts are kept without being displayed while the article remains hidden. For articles deleted by an operator or moderator, counts are kept without being displayed for appeal review and deleted together with the article’s records 30 days after deletion when no appeal is open. Site-wide reading totals have no fixed retention period. - **Access records (server logs)**: query text and similar information are redacted; older records are removed when the configured size limit is exceeded. - **Database backup copies**: the latest ten generations are kept. Older copies are deleted as generations are replaced. After restoring a backup, already deleted information is deleted again. If an investigative authority lawfully requests preservation of records, deletion may be suspended for that period. ## 5. Safeguards - Communications with the site are encrypted. - Sign-in uses emailed codes and passkeys, without passwords. - Operator and moderator roles are separated and their access is limited. Access to chat records who read it, when and why. - Before using providers abroad, the operator checks the systems in those countries (section 3). ## 6. Your requests You may ask us to: - Tell you what information we hold (disclosure). - Correct mistakes (correction). - Stop using or delete information (cessation of use or deletion). - Stop sending information to providers abroad. Send requests from your registered email address to the contact desk in section 1. To verify your identity, we may ask you to reply from that address. There is no charge. We normally respond within 30 days. You can retrieve many records yourself through Export in Settings. Complaints are accepted at the same contact desk. ## 7. Leaving and deletion You may leave at any time using Leave in Settings. You may also request this through the contact desk in section 1. - Deleted information: your email address, sign-in state and passkeys, settings, drafts, stance records, notifications and chat messages. - Posts and replies: when leaving, you choose whether to delete them or keep them without identifying you as their author. If you do not choose, they remain without your author attribution, to preserve reply connections. Deletion follows the same process as posts you delete yourself (section 4). If you keep them without your name, their previous versions remain as well, and you can no longer delete them after leaving. - Retained information: moderation and appeal records without member-identifying information until the appeal period ends, and copies of reported chat messages for the period in section 4. If you leave while your account is suspended, a one-way hash of your email address is kept for one year, to prevent immediate re-registration. - Your username: for 90 days after you leave, no one else can use the same username. - Before leaving, you can retrieve your records through Export in Settings. ## 8. Age, country of residence and stances - The tomola trial is open to people aged 18 or over. - During the trial, residents of the EEA (the EU, Norway, Iceland and Liechtenstein), Switzerland, the United Kingdom, Gibraltar, Jersey, Guernsey, the Isle of Man and mainland China cannot register. You are asked to declare your country of residence when registering. - Stances, such as agreeing or disagreeing with a question, may include political views or beliefs. We therefore ask for your separate agreement before your first stance selection. ## 9. When the operator reads content or discloses it outside tomola - The operator and moderators read public posts and replies when reports are made, when reviewing appeals and when checking the forum's safety. - Chat is read only when a message is reported or an appeal is reviewed, by the operator and moderators responsible for that language, within the scope you agreed to when entering chat. Failure investigations use records such as times and counts without reading message text. - Before a person reads a report, AI may read the reported posts or messages and surrounding content to prepare priorities and an assessment. People decide whether to hide content. AI access is recorded in the same way as human access. - Automated systems and AI may detect and automatically hide spam and mechanical abuse, such as rapid mass posting, repeated content and promotional links. In chat, mechanical checks examine frequency and repetition without sending the text to AI. We notify you when content is hidden automatically and accept appeals. - In response to enquiries from investigative authorities, bar associations and others, we do not disclose post or chat content, authorship and timing records, IP addresses or similar records without legal procedures such as a court warrant. - When disclosing under legal procedures, we notify you unless the law prohibits notification. - If someone's life or physical safety is in imminent danger, information may exceptionally be disclosed to the extent necessary to prevent harm. ## 10. Cookies tomola uses cookies for sign-in and identity checks, sign-in progress, and remembering interface language and display preferences. Depending on the action, cookies may also be set for people who are not signed in. - **Sign-in state**: 30 days. Renewed during use, but the sign-in session itself ends after at most 90 days. - **Email sign-in progress**: 15 minutes. These remember the browser that requested the link, a pending code, the recipient email address and resend information. The pending-code and recipient cookies are cleared when sign-in is completed or restarted. The browser marker remains until expiry. - **Passkey identity checks**: 5 minutes. A marker links a registration or sign-in request to its response. - **Signed-out notice**: at most one minute. Cleared on the next page or data read. - **Return to a post read before registration**: one day. Set for public readers and cleared when sign-in or registration is completed. - **Later on the passkey prompt**: 30 days. Remembers that the prompt should be deferred for that member. Remains until expiry even after sign-out. - **Explicitly selected interface language**: at most 400 days. Selecting again renews it. Merely viewing a language specified by a link does not write this cookie. tomola does not use advertising or external analytics cookies, and does not load scripts from outside providers. ## 11. Changes to this notice We update this page's date when it changes. For substantial changes, we notify you when you sign in.