Privacy

We collect the least we can get away with.

This page explains what happens to your information when you visit cultmath.com, subscribe to the newsletter or write to me. It is written in plain language on purpose. If something here is unclear, that is a fault in the writing and worth telling me about.

CULT+MATH LLC is a one-person firm in Houston, Texas. There is no data team, no marketing department, and no third party being sold anything. When you send something through this site, I am the person who reads it.

The short version

  • Personal information is collected only when you choose to send it, through a form.
  • Public pages carry a Google advertising tag, Google Analytics and the LinkedIn Insight Tag, which set cookies. All three are named and explained below.
  • Private client pages carry no tracking of any kind.
  • Your information is never sold, rented, or traded. Not to anyone, for any price.
  • You can ask for a copy of what is held about you, or ask for it to be deleted, and it will be done.

Effective November 21, 2026
CULT+MATH LLC · Houston, TX

What is collected, and why

There are eight situations in which this site handles information about you. Nothing outside these eight is collected.

1. Conversation requests (this form was retired on October 23, 2026)

Until October 23, 2026, a form at cultmath.com/request asked for your name, email address, company, role and company website, and five short questions about your business: the decision you are weighing, who buys from you and why, what they compare you to and whether your price sits higher, the same or lower, where sales stall, and what you would want customers to say about you. Everything was required. Nothing confidential was asked: no costs, contracts or exact prices. The page also recorded which page on this site sent you, if any. The form no longer exists; this section applies to requests already sent.

That information was used to read your request, prepare the positioning review, and reply. It is also looked at in aggregate, by me alone, to understand who asks and what they ask about, so the work and this site get better. It is not added to a mailing list, and it is never sold or shared.

Your request is emailed to me and saved in a request store: a database hosted in the United States (Upstash, through Vercel) that only I can open, behind the same sign-in as the rest of the site's private tools. To limit spam, the site counts requests per connection for one hour using a scrambled version of your IP address; the address itself is never stored.

Reader polls. If you vote on a question at /predict, your answer is added to an anonymous count in the same store. To allow one vote per browser, the site sets one cookie holding a random, signed identifier. It contains nothing about you, is HttpOnly and Secure, and is sent only to the voting service. The store keeps a scrambled version of that identifier with the answer you chose, so it can tell that a browser has already voted, and it counts votes per connection for one hour using a scrambled version of your IP address. Your name, email and IP address are never collected for a vote.

Partner referrals. If you arrive through a link shared by a CULT+MATH referral partner, the partner's tag (for example partner-name) is kept in your browser's local storage for 30 days and passed to Stripe with any brief you buy, so the partner can be paid a referral fee. It names the partner, never you, and is not sent anywhere else.

2. When you submit a form inside a Google ad

CULT+MATH sometimes runs advertising on Google. If an ad carries a form, the name, email address, company, and answers you provide there are passed to me by Google and used the same way as a request sent through this site: to read it and reply.

Google's own handling of that submission is governed by Google's privacy policy, not this one. Google also sets advertising cookies in its own products. It does not set any on this website.

3. When you open a private client or working page

Some pages under /clients and /for are password protected. When you sign in, a session cookie is set so that you are not asked for the password on every page. That cookie holds a signed session token and nothing else. It is HttpOnly, Secure, restricted to that one client area, and it expires on its own.

It is a functional cookie. It does not follow you, it cannot be read by other sites, and it is not used for advertising or measurement.

4. When you browse a public page

Public pages on this site carry a Google advertising tag, account AW-18400942856. It exists for one reason: CULT+MATH runs advertising on Google, and the tag reports back whether an advertisement led to a conversation request. Without it there is no way to tell a working advertisement from a wasted one.

The tag sets cookies in your browser and sends Google information about your visit, including your IP address, the page you are on, and whether you completed a request. Google may use this to measure and personalise advertising under its own terms.

If you buy a brief, the page you land on afterwards reports that purchase to the same account — the amount paid, the currency, and the Stripe checkout reference, which is sent so that one sale cannot be counted twice. That reference is an opaque identifier Stripe generates for the transaction. It is not your name, your email address or your card, none of which this site ever sends to Google.

The tag is deliberately absent from every private page. It does not run on client deliverables under /clients, on working pages under /for, or on the sign-in page. Client engagement material is not Google's business, and those pages are excluded from advertising measurement by design rather than by oversight.

Consent settings are applied before the tag loads. Visitors from the European Economic Area, the United Kingdom and Switzerland have advertising and analytics storage set to denied by default, so no cookies are set for them and advertising identifiers are stripped from any request. This site does not show a consent banner; it withholds the storage instead.

Google Analytics. Public pages also load Google Analytics 4 through the same file, so I can see which pages are read, which links lead to a conversation request or a purchase, and which sites and AI assistants send readers here. It records the page you are on, the section of the site it belongs to, the page that sent you, your approximate location and device type, and a small set of events: clicking a link to the Decision Briefs, clicking a payment link, completing a purchase, and reading a page (after you scroll, click, tap or type, either scrolling past half the page or keeping it open for 15 seconds). It also notes when a browser reports that it is being driven by automation software, so automated visits can be counted separately. The purchase event carries the amount, the currency and the same Stripe checkout reference described above. Nothing you type into a form is sent to Google Analytics, and no personal information is sent as an event detail. Google Analytics does not store IP addresses. It sets cookies in your browser, and the regional consent defaults above apply to it in the same way. It is absent from the same private pages as the advertising tag.

LinkedIn Insight Tag. Public pages also load LinkedIn's Insight Tag, account 10223876. It does two jobs. It tells CULT+MATH, as totals only, which kinds of companies and job titles visit and whether a LinkedIn advertisement led to a sign-up. And it lets CULT+MATH show LinkedIn advertisements to people who have visited this site, for example someone who read the DEOS page and did not start a trial. It sets cookies in your browser and sends LinkedIn your IP address, the page you are on, the page that sent you and a browser identifier. If you are signed in to LinkedIn, LinkedIn can match the visit to your account. CULT+MATH never learns who you are: LinkedIn reports only totals, and only once a group is large enough. LinkedIn uses this data under its own terms.

The LinkedIn tag is absent from private pages under /clients and /for, and from the DEOS app itself: your records and what you do in the app are never reported to LinkedIn. LinkedIn has no equivalent of Google's consent settings, so instead the tag does not load at all for a browser set to a European time zone, which covers the European Economic Area, the United Kingdom and Switzerland. It also does not load when your browser sends Do Not Track or Global Privacy Control.

If you would rather not be measured at all, any mainstream ad or tracker blocker will stop the tags, and nothing on this site breaks when it does. Google also offers its own ad settings and opt-out tools, and LinkedIn lets you turn off advertising based on other sites in its advertising settings.

5. When you subscribe to Sound Decisions

Many public pages show a small bar near the bottom with one field, once you have read some of the page. If you enter your email address and press Subscribe, this site saves the address, the page you were on, a note that you asked for the emails, and the time. That is all it saves. The bar is not shown on the request page, the purchase pages, or the private pages, and dismissing it keeps it away for 30 days. It remembers that in your browser's own storage, not on a server.

The address is stored in a database hosted in the United States (Upstash, through Vercel) that only I can open, and it is passed to Substack, which sends one confirmation email and then the newsletter. Substack holds the address under its own privacy policy. Every email carries an unsubscribe link, and you can also write to me to be removed from both places. To limit abuse, the site counts attempts from each connection using a scrambled, one-way code in place of your IP address. The code expires within the hour. The site does not say whether an address was already subscribed.

6. Ordinary server records

This site is hosted on Vercel. Like any web host, its servers keep short-lived technical records of requests, which include IP addresses and browser information. These are used to keep the site running and to investigate abuse or errors. I do not build profiles from them and do not connect them to any form submission.

Typography on this site loads from Google Fonts, which means your browser contacts fonts.googleapis.com and fonts.gstatic.com. Google receives your IP address as part of that request. It is a genuine third-party contact and worth naming rather than burying.

7. When an AI assistant uses the CULT+MATH connector

The research on this site is also offered to AI assistants (Claude, ChatGPT and others) through an MCP server at cultmath.com/mcp, with editions for app directories at /mcp/claude and /mcp/chatgpt. There is no account and no sign-in. The assistant sends what it needs to answer your question: usually a few keywords, or the address of a study.

For each request the server keeps, in weekly counts: which tool was used, the keywords or study address the assistant sent (shortened to 160 characters), how many results it found, and the name the assistant gives for itself, such as "claude-ai". These counts show which questions people bring and which ones the research does not yet answer. The hosting logs record the same tool and keywords for troubleshooting. Your conversation is never sent to the server, and the server never keeps an IP address, a user agent or a session identifier. To limit abuse, each connection is counted for one minute in the server's memory and then forgotten. To know roughly how many different connections use the connector each week, the server turns each connection's address into a scrambled, one-way code that changes every week, and adds it to a counter that can only report an approximate total. The code is not stored anywhere it can be read back or listed, and it cannot be turned back into an address. If you ask ChatGPT not to share your searches, your connection is not counted either. A connector address that carries an invitation code (ref=) is counted separately from everyone else's, so invited testers don't read as new demand.

In ChatGPT (the /mcp/chatgpt edition) less is kept, and you can turn it off. Instead of the keywords as sent, the server keeps only the topic words of a search, at most six of them: no numbers, email addresses, links, common first names or filler words, and never the question itself. To stop even that, tell ChatGPT not to share your searches with CULT+MATH. It then sends each search with recording switched off, and the server keeps nothing about it beyond the fact that a search was made.

Please do not put personal details into a question meant for the research. The server cannot always tell them apart from keywords, and will count them the same way until they expire.

8. When you use the DEOS app

DEOS, the app built on the DEOS Open standards, runs at cultmath.com/deos/app, in your browser or installed from your browser or an app store. The Android version is the same web app, opened by Chrome inside an app frame.

Your account. You sign in with Google. Google tells our server only that the sign-in is real and gives an account number; our server turns that number into a scrambled id it cannot turn back, and keeps with it the date your free month started and ends and, once you subscribe, your Stripe subscription number and its status. We do not keep your name or your email address. Signing in sets one cookie on cultmath.com that keeps you signed in on that device for 30 days; it is not used for anything else. Sign out under You.

The free month and the subscription. The app is free for your first 30 days and asks for no card. After that it costs $99 a month. You pay on Stripe's own page, not in the app; Stripe keeps your card and billing email under its own privacy policy, and tells our server only whether your subscription is active. If you do not subscribe, your records stay on your device and you can still read and download them.

Bringing in files from Google Drive. If you choose to, the app opens Google's own file picker. It can open only the files you pick there, never the rest of your Drive (Google's "drive.file" permission). Your browser downloads those files with a short-lived Google token and sends their text, or a PDF itself, to our server, which sends it to Anthropic's Claude API to sort and read. Neither the files nor their text are stored by CULT+MATH; our logs record only how many files ran and whether it worked. The records drafted from them, the links between them and any standards drafted from your records are saved only on your device, after you look them over.

Your records stay on your device. Every record you write, and every box you tick, is saved in your browser's storage on that device and nowhere else. CULT+MATH never receives it, cannot read it and cannot recover it. A backup file you download is yours and stays with you. To delete everything, delete the records in the app, clear this site's data in your browser, or uninstall the app.

Dictation. If you tap Speak beside a field, your browser turns your speech into text using its maker's speech service: Google's in Chrome and Edge, Apple's in Safari. Your voice may be sent to that company, under its own privacy policy, but never to CULT+MATH; only the text lands in your record, on your device. Nothing is recorded unless you tap Speak, and it stops when you tap Stop or leave the screen. Your phone keyboard's own dictation works the same way under your phone maker's policy.

Voice memo. When you tap Record a decision or start a record, the app records your voice on your device and sends it to cultmath.com in pieces under a minute. Our server sends each piece to Google Cloud Speech-to-Text to turn it into words, then sends the words to Anthropic's Claude API to fill in the record's fields. To find evidence, the app then searches CULT+MATH research with a few topic words the AI picks from what you said (never names of people), the same way a search you type does. When you review an earlier decision or turn a plan into a decision, the text of that earlier record is sent with the recording, so the AI can compare what was expected with what you say now; like the rest, it is not stored. Google processes these requests in memory and does not store the audio, and we keep Google's data logging off; Anthropic does not train on API data. CULT+MATH stores no audio, no words and no drafts, and our logs record only that a memo ran, how many seconds it was and whether it worked. Nothing is saved until you review the draft and tap Save, and then it is saved only on your device. If you use a beta invite code, it is kept on your device so you do not retype it; remove it under You.

The app carries no advertising tag and no analytics. Section 4 does not apply to it. It loads its fonts from Google Fonts, as section 6 describes.

Evidence search. If you search for evidence inside the app, the keywords in the search box are sent to the research server at cultmath.com/mcp/app. As in the ChatGPT edition (section 7), the server keeps only the topic words of a search, at most six, in weekly counts with no IP address and no identifier. Tick "Don't count my searches" and nothing about your searches is kept. Your record is never sent with a search. The sound check can suggest keywords from your record; they appear in the search box for you to edit, and nothing is sent until you tap Search. The rest of the sound check, which compares a record with your past records, runs on your device and sends nothing.

Handing a record to your assistant. If you choose "Open in Claude" or "Open in ChatGPT", the app opens that assistant with the standard and your draft in the link. That is the only time a record leaves your device, and only when you press the button. From then on the assistant's own privacy policy applies, not this one. "Copy the prompt" puts the same text on your clipboard and sends nothing.

Assistants in your browser. Where your browser offers it, the app lets the browser's own assistant read the standards and your records, start a record and check one, on that device. Nothing is sent to CULT+MATH when it does.

Buying. The subscription is paid on Stripe's page, as above. Links to Decision Briefs open cultmath.com, where sections 1 and 4 apply as on any public page.

What this site does not do

Stated plainly, because most policies bury it:

  • Named tools only. Public pages use Google Analytics 4, the Google advertising tag and the LinkedIn Insight Tag, all described above. There is no other analytics tool, no session recording, no heatmaps and no tag manager.
  • No tracking on private pages. Client and working areas carry no tag at all. Pages under /for additionally enforce a security policy that would block one at the browser level.
  • No sale of personal information. Your information is not sold, rented, licensed, or traded, and it never will be. The Google and LinkedIn advertising tags do let those companies show CULT+MATH advertisements to people who have visited this site, which California law calls sharing for cross-context behavioural advertising. You can stop it with a tracker blocker, Global Privacy Control or the settings linked above, and nothing on the site breaks.
  • No profiling and no automated decisions. Nothing about you is scored by a machine. A person reads what you send.
  • No sensitive information wanted. Please do not send health, financial account, government identifier, or similarly sensitive details through a form. If you send them anyway, they will be deleted.

If any of this changes, this page changes in the same commit as the code that changed it. A privacy policy that lags the software it describes is not a policy, it is a decoration.

Who else handles your information

Running even a small business means using other companies' software. These are the only ones that touch anything you send. Each is used for the single job listed and nothing more.

ServiceWhat it doesWhat it sees
VercelHosts the websiteTechnical request records, including IP address
ResendDelivers your request to my inboxThe contents of your form submission
Upstash (through Vercel)Stores requests so I can review themThe contents of your form submission
Upstash (through Vercel)Keeps the weekly counts of what AI assistants ask the connectorThe tool used, the keywords or study address sent, the assistant's name, any invitation code, and a weekly approximate count of distinct connections (one-way codes, not addresses)
Google WorkspaceEmailThe contents of your form submission, as delivered to my inbox
Google AdsAdvertising, and ad-based formsWhat you type into an ad form
Google tagMeasures whether an ad produced a requestIP address, pages viewed, conversion events
LinkedIn Insight TagMeasures LinkedIn advertising and lets CULT+MATH show LinkedIn ads to past visitorsIP address, pages viewed, the page that sent you, a browser identifier
Google FontsServes the typefacesYour IP address when a page loads

Beyond these, information is disclosed only where the law requires it, or where it is necessary to establish or defend a legal claim. If CULT+MATH LLC were ever sold or merged, information would transfer with the business, and this page would say so before it happened.

Some private client deliverables use additional services chosen for that engagement. Where that applies, it is set out in the agreement with that client rather than here, because it is their data and their contract governs it.

How long it is kept

  • Newsletter subscriptions are deleted from the subscriber store automatically two years after you subscribe, or sooner if you ask. Substack keeps its copy until you unsubscribe.
  • Conversation requests are deleted from the request store automatically two years after you send them. The emailed copy is kept for as long as the relationship is live, and for up to two years after the last contact, so that a later conversation can pick up where the previous one left off. Ask sooner and both go sooner.
  • Session cookies for private pages expire on their own within hours. Nothing persists after that. The DEOS sign-in cookie expires after 30 days.
  • DEOS accounts (the scrambled id, trial dates and subscription status) are kept while you use the app. Ask and the account is deleted; your records are on your device and are not affected.
  • Poll votes: the record that a browser voted, and its answer, is deleted automatically about 13 months after the vote, and the voting cookie expires on the same schedule. Vote totals contain nothing personal and are kept as a record of how readers called the question.
  • Advertising cookies set by the Google tag follow Google's own retention, typically up to two years, and can be cleared from your browser at any time. LinkedIn cookies set by the Insight Tag follow LinkedIn's own retention, set out in its cookie policy, and can be cleared the same way.
  • Server records are retained on the host's standard schedule, which is a matter of weeks, not years.
  • AI assistant counts from the connector are deleted automatically 400 days after the week they were counted in, so one year can be compared with the next.
  • Client engagement material is governed by the agreement with that client, which sets its own retention and deletion terms.

Your rights

Depending on where you live, you may have legal rights over your information under the Texas Data Privacy and Security Act, the California Consumer Privacy Act, the UK and EU General Data Protection Regulation, or a comparable law. Those rights generally include the following:

  • Access. Ask for a copy of what is held about you.
  • Correction. Ask for anything inaccurate to be fixed.
  • Deletion. Ask for it to be erased.
  • Portability. Ask for it in a form you can take elsewhere.
  • Objection. Ask that it stop being used for a particular purpose.
  • No retaliation. Exercising any of these rights costs you nothing and changes nothing about how you are treated.

These requests are honoured for everyone who asks, wherever they live. A firm this size may fall outside some of these statutes. That is not a reason to make people argue about jurisdiction to get their own information back.

There is no dashboard to navigate. Write to me, tell me what you want done, and it gets done. Expect a reply within thirty days, and usually much sooner.

Where the GDPR applies, the lawful bases relied on are your consent when you choose to send a form or subscribe, and legitimate interests in operating a secure website and responding to enquiries. Consent can be withdrawn at any time. Nothing in the five situations above involves an international transfer beyond the ordinary use of the United States services named above.

Security, and its limits

This site is served over HTTPS. Private client areas sit behind password authentication with signed, expiring, HttpOnly session cookies scoped to a single client's pages. Access is granted on a least-privilege basis, meaning each client area can reach only its own material. Credentials and keys are held as environment variables and are never committed to the code repository. A content security policy restricts what any page is permitted to load.

No system is perfectly secure, and anyone claiming otherwise is selling something. What is promised here is a specific standard of care, not a guarantee. If a breach ever affected your information, you would be told directly and promptly rather than through a legal notice buried on a webpage.

Children, and changes to this page

This site is built for business audiences and is not directed at children. Information is not knowingly collected from anyone under 18. If you believe a child has sent something through a form here, tell me and it will be deleted.

If this policy changes, the effective date at the top changes with it, and the substance of the change is described here rather than quietly slipped in. Material changes will be summarised in this section for a reasonable period after they take effect.

Change log · August 20, 2026: first published version. Amended the same day to disclose the Google advertising tag added to public pages, the cookies it sets, the regional consent defaults, and how to opt out. The earlier claim that this site carried no advertising tag was accurate when written and is no longer, which is why it was removed rather than left standing. September 25, 2026: added reader polls at /predict, the anonymous voting cookie, and how long a vote is kept. October 23, 2026: added a newsletter sign-up bar to public pages, and what it saves and where, and added Google Analytics 4 to public pages. The earlier statement that this site carried no analytics product was accurate when written and is no longer, which is why it was changed rather than left standing. November 14, 2026: Google Analytics now records the site section and a page-read event, and marks automated browsers; the retired request-form events were removed from this description. · November 21, 2026: the DEOS app gains Google sign-in, a free first month and a $99 monthly subscription paid on Stripe, and an optional import from Google Drive limited to the files you pick.

Contact

Questions about this policy, or a request to see, correct, or delete your information, should go to CULT+MATH LLC, Houston, Texas.

The fastest route is email: write to privacy@cultmath.com and put "privacy request" in the subject line. It will be handled ahead of everything else in the queue.