llms.txt
+5 −3
https://everynumber.app/llms.txt
− - [LLMs shouldn't write your numbers](https://everynumber.app/blog/llms-shouldnt-write-your-numbers): In the product we're building, a language model only writes a small typed spec of the dashboard. A compiler turns it into SQL that PostgreSQL runs on the rows of your upload, with every value passed as a parameter, and the model is never called when you view a number, so the same data gives the same number. Today a person follows these rules by hand.− - [Anatomy of a receipt](https://everynumber.app/blog/anatomy-of-a-receipt): A receipt is the 6 facts behind one dashboard number (definition, filters, rows used, rows left out, source, and the date the data runs through). Each part prevents a specific mistake. In the product we're building, the receipt is created with the number and stored with a hash of its inputs, so any past number can be reproduced.− - [Why we start with a $99 concierge pilot](https://everynumber.app/blog/why-a-99-concierge-pilot): Before we automate anything, a person builds each pilot dashboard by hand, following the same rules the product will follow. The pilot costs $99 so both sides make a small, real commitment. It's fully refunded if we can't build the dashboard, and it counts toward your first month at launch.+ - [How to calculate refund rate from a Shopify orders export (and by sales channel)](https://everynumber.app/resources/shopify-refund-rate): Refund rate from a Shopify orders export can mean paid orders with any refund divided by paid orders, or refunded money divided by sales. Count each order once, since extra items get their own rows. Leave out test and canceled orders, and say which date you used. Shopify's sales reports record returns on the day they were processed.+ - [How we made exported Excel formulas count the same rows as PostgreSQL](https://everynumber.app/blog/excel-formulas-that-match-sql): Every Number's Excel export turns most dashboard numbers into live formulas that count the same rows PostgreSQL counted. Excel and SQL treat letter case, blank cells, wildcards and TRUE differently, so each formula spells out the database's rule. An evaluator we wrote matched all 35 test numbers and 26 sample numbers. We haven't checked inside Excel itself yet.+ - [LLMs shouldn't write your numbers](https://everynumber.app/blog/llms-shouldnt-write-your-numbers): In Every Number, a language model only writes small typed specs. A compiler turns them into SQL that PostgreSQL runs on the rows of your upload, with every value passed as a parameter, and the model is never called when you view a number, so the same data gives the same number.+ - [Anatomy of a receipt](https://everynumber.app/blog/anatomy-of-a-receipt): A receipt is the 6 facts behind one dashboard number (definition, filters, rows used, rows left out, source, and the date the data runs through). Each part prevents a specific mistake. In the app, the receipt comes from the same query as the number and carries an ID hashed from its definition and upload, so a past number can be reproduced.+ - [Why we start with a $99 concierge pilot](https://everynumber.app/blog/why-a-99-concierge-pilot): In the pilot, a person builds your dashboard by hand, following the same rules the app follows. The pilot costs $99 so both sides make a small, real commitment. It's fully refunded if we can't build the dashboard, and it counts toward your first month at launch.