> ## Documentation Index > Fetch the complete documentation index at: https://docs.tryprofound.com/llms.txt > Use this file to discover all available pages before exploring further. # Why past numbers change > Answers are stored once. Topics, tags, and brands are applied from your current configuration on every request, so editing your setup changes historical numbers. * Profound stores each answer once. The response text, citations, and date do not change. * Topic, tags, and brand attribution are not stored with the answer. They are read from your current configuration each time a report loads or the API is called. * Editing your configuration therefore changes past numbers. Moving a prompt to a new topic moves all of its past answers. Excluding a brand under Configure Competitors removes its past mentions from every report. * Reports reflect the current configuration only. An older configuration cannot be viewed. ## How it works Your configuration defines which prompts run and how their answers are labeled. Collection runs each prompt and stores the answer. Reports and the Answers API join stored answers to the current configuration on every request. | Stage | Contents | Mutability | | ------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------ | | Configuration | Category, topics (one per prompt), prompts (permanent id, editable text), tags (many-to-many with prompts), brands and aliases, the Configure Competitors include or exclude list, models, regions, personas | Changes at any time | | Collection | One run per prompt, model, region, persona, and day. Brand names in the response are extracted once and stored. | Written once, never modified | | Serving | Reports and the Answers API join stored answers to the configuration as it is now | Evaluated on every request; changes apply to all dates | Each stored run carries `run_id`, `date`, `prompt_id`, `model`, `region`, `persona`, `response`, `citations`, `search_queries`, and the brand names extracted from the response. At request time, `prompt_id` resolves to the current topic, tags, text, and status, and the extracted brand names resolve to the current brand list. Deleted prompts are excluded. Paused prompts are included. A **run** is one prompt sent to one model from one region on one day. Each run produces one answer and one row in the Answers API. A **mention** is a brand named in an answer. A **claim** is a statement about a brand extracted from an answer, scored positive or negative. See the [data model](/cookbook/setup/data-model) for how categories, topics, prompts, tags, and assets relate. Prompts run once a day per model and region. Days are counted in Eastern Time. ## Effect of configuration changes on past data | Change | Effect on past data | | ------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Move a prompt to another topic | All of its past answers move to the new topic. | | Add or remove a tag | All of the prompt's past answers gain or lose the tag. | | Rename a topic or tag | The name changes. Numbers do not. | | Edit a prompt's wording | Numbers do not change. Past answers display the new wording even though they were generated from the old wording. | | Pause a prompt | The prompt stops running. Its past answers remain in reports. | | Delete a prompt | The prompt stops running and its past answers are removed from reports and the API. Restoring the prompt restores them. | | Exclude a brand under Configure Competitors | Its past mentions stop counting on every date. Share of voice changes for every brand. Your visibility score rises if past answers named only the excluded brand and no other, because those answers leave the denominator. Your average position improves in answers where the excluded brand was named before your brand. Switching to an include list does the same for every brand not on it. | | Add a prompt, topic, model, or region | Data begins on the first day it runs. | Profound does not re-run or re-analyze old answers. Only the labels applied to them change. Answers API rows reflect configuration changes immediately. Some summary reports refresh on a schedule and can lag briefly. ## Store your own configuration snapshots The Answers API returns the current topic, tags, and brand list, so metrics you compute from it change whenever your configuration changes. To keep fixed historical numbers, save the following once a day and join to the saved copies instead: | Save | Reason | | -------------------------------------------------- | -------------------------------------------------------------------------------------------- | | The topic of each `prompt_id` | A moved prompt does not change your past topic totals. | | The tags of each `prompt_id` | The same, for tags. | | Each prompt's wording | You know which text generated an old answer. | | Your Configure Competitors include or exclude list | The Answers API does not apply it, so you need the list as of that date to match the report. | | The rows you have fetched | Deleted prompts are removed from the API. Saved rows are not affected. | ## Recommendations * Pause prompts instead of deleting them to keep their history. * Move prompts between topics and tags deliberately. Each move changes past numbers for both the old and new group. * Expect share of voice to change across all dates when you exclude a brand or switch to an include list. * If a number must not change, for example in a quarterly report, export it at the time of reporting. The live report always reflects the current configuration.