Result caching and freshness

Summary#

Metric Vault caches data at several levels so that repeating a piece of research is fast and does not hammer the data providers. The practical consequence is that running the same tool on the same target twice in a short window can return identical numbers. This page explains the layers, how long each tool's data lives, and what each layer means for your credits.

Purpose#

Search data does not change minute to minute. A domain's backlink profile is essentially the same today as it was yesterday; a keyword's monthly search volume is a monthly figure. Calling a paid data provider again for numbers that have not moved wastes money and makes the product slower for everyone.

Caching also protects your work. A saved result is a copy you can reopen at no cost for three months, which means a report you built for a client does not evaporate because you moved on to the next task.

The rule the design follows: never substitute cached data silently when you have asked for a fresh run, and never charge you twice for the same answer. Where the two goals conflict, you get told.

Overview#

Three layers matter to you.

A fourth, narrower one: PageSpeed keeps its own store, described below.

How it works#

Layer 1: your own saved results#

Every run that produces a result is saved to your workspace with its query, the cost at the time, and the rendered report. They are kept for 90 days and then purged.

Those saved runs power three things:

• The You already ran this prompt, which appears when you run a tool on a query you have used before. Choosing Open saved result returns the stored copy and charges nothing. See How to run a tool.
• The Library, where you can reopen, re-run or delete saved work. See Saved Work.
• Growth Actions, which reads your saved AI-visibility results rather than running anything new.

The prompt never substitutes the saved copy for you. It always offers the choice, because a saved result can be out of date and only you know whether that matters.

Layer 2: the shared data cache#

Behind the scenes there is one cache of provider data shared across all customers. It is keyed by the tool, the query, and your filters: country, device, date range, minimum volume and intent. Change any of those and it is a different key, so you get a fresh fetch.

When a run hits a fresh entry, you get the stored data immediately and no data provider is called. The credit is charged anyway. The saving is on our provider bill, not on your allowance, and the quota rules apply exactly as they would to a live run. This is why a repeat run of the same domain can come back in under a second and still cost 6 credits.

Because the cache is shared, a colleague's run and a stranger's run on the same domain both fill the same entry. Nothing about who ran it, or their account, is shared: only the provider data for that query.

How long each tool's data lives

The windows are not arbitrary. They track how fast the underlying data actually moves: backlink profiles are slow, rank positions are daily, and Content Optimizer gets the shortest window of all because it analyses a page you own and are actively editing.

What "Run fresh" actually does#

Run fresh in the already-ran prompt always charges the current credit cost and always sends a new request to the server. If the shared cache still holds a valid entry for exactly that tool, query and filter combination, the server serves that entry rather than calling the provider again.

So "Run fresh" reliably means "do not give me my old saved copy". It does not guarantee a brand-new provider fetch inside the tool's cache window. If you need data that is genuinely newer than the window, change something meaningful about the run: a different country, a different date range, or wait out the window.

Layer 3: provider call caches#

Each individual data-provider request is cached on its own, typically for one hour, one day or seven days depending on how volatile that particular dataset is. This layer sits below the shared cache and is invisible in the interface.

Metric Vault administrators can shorten or lengthen these per-tool lifetimes, and can purge a tool's cached data. Eight of the content tools share one grounding fetcher, so they cannot be tuned individually. See Tools and cache management and Caching architecture for the internal view.

PageSpeed is a special case#

PageSpeed & Vitals stores its own reports:

• A repeat check of the same URL and device within 6 hours returns the stored report.
• If Google's PageSpeed service is rate limiting or unavailable, a stored report up to 30 days old is served instead of an error.

The second rule is why an occasional PageSpeed result can be older than you expect. It is a deliberate trade: a slightly stale score is more useful than a failure.

What is never cached#

• Your credit balance and usage. Those are read live.
• The result of a tool that is not in the shared-cache table.
• Anything in the browser-only tools, which never call the server at all.

Screenshots#