First, find which layer stopped moving
A Looker Studio chart sits at the end of a chain. A source records the activity: a GA4Google's event-based analytics platform. property, an ad platform, a booking or order system. A load, export or scheduled query may copy it into a warehouseA database designed for analysis that stores data loaded from many systems, such as analytics, ad platforms and the CRM, so it can be joined and queried. table, an extract may keep a stored snapshot, the data source caches results for as long as its freshness setting allows, the date control picks the days, and a scheduled delivery may turn the result into an email attachment. No layer can be newer than the one it reads.
Work back from the reader and note the newest date at each step: the attachment, the live report before and after the refresh action, any extract, the reporting table and its inputs, and the same day in the source's own interface. The first layer older than the one beneath it is where the chain stopped, and the repair goes there, not in the chart.
Charts showing an error, or showing nothing, are a separate problem: the site's Looker Studio guides note that an exhausted connector quota produces an error rather than old figures, and such charts belong to the diagnosis for charts with errors or no data.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Figures move after a refresh, or a partner-sourced chart trails the platform | Figures before and after refreshing; the vendor's last sync | A cache, freshness setting or partner sync slower than the reader expects | The GA4 mismatch diagnosis's cache fix; for a partner, the vendor's sync |
| One page stands still while live pages move | Whether it reads an extract with refresh switched on | An extract whose scheduled refresh stopped or was never switched on | The GA4 mismatch diagnosis's extract fix, timed after its input, with an owner |
| Every page on a warehouse source stops on one date | The latest date in the table and in each input | A warehouse table whose scheduled load, query or export stopped | Repair the step, rerun missing load or query days, add a freshness test |
| The report opens on the same dates every time | The date control's default | A date control or default fixed to a past period | Default to the period each page is meant to answer |
| The newest day is missing or short, then fills in | The same day in the source later; how recent days load | Recent days the source has not finalised yet | Read closed days; reload a trailing window; mark recent days provisional |
| The attachment lacks the latest day, or has stopped | Start time against each layer's update; the creator's account | A scheduled delivery sent before the sources updated, or stopped with its creator | Start after the last update; recreate under a role account |
Fix, verify and prevent each cause
The first two causes share an owner: the section "A cache or extract behind the property" of the diagnosis for Looker Studio figures that do not match GA4, with its row "Cached or extracted data out of date", holds their fix, and that page also owns closed days that disagree with GA4 rather than stand still.
A cache, freshness setting or partner sync behind the source
The data sources and connectors guide explains that how long cached results count as fresh is set on each data source rather than the report, with options and defaults in the help centre, and that the refresh action goes past the cache for whoever uses it. It also describes partner sources that keep their own scheduled copy of a platform's data and answer Looker Studio from it, so a chart shows no more than the vendor held at its last sync.
Fix: for the cache and the freshnessWhether a table, feed or report holds data as recent as it should; a freshness check fails when the latest date is older than the source's expected arrival. setting, apply the mismatch diagnosis's fix. For a partner source, check the vendor's account for its last sync and for a platform login that has stopped working, and take a stalled sync to its support.
Verify: the chart's newest closed day matches the platform's own interface.
Prevent: record each source's freshness setting and any vendor's sync timetable in the dashboard inventory, and give each partner account a named owner.
An extract that is not refreshing
The performance and caching guide describes an extract as a stored copy of chosen fields, optionally filtered and limited to a date range, that can refresh on an automatic schedule; the mismatch diagnosis adds that readers see the snapshot until the next scheduled refresh. If that refresh was never switched on, or has stopped, the page keeps the day the copy was made.
Fix: the mismatch diagnosis covers the refresh itself. Add two things: time it after the layer the extract reads has updated, as general practice, since an extract refreshed before its table loads copies the day before; and name an owner for the schedule.
Verify: after the next scheduled run, the page shows the same newest closed day as a live source on the same data.
Prevent: when a page moves onto an extract for speed, record its schedule and owner in the inventory as part of that change.
A warehouse table whose load, query or export stopped
A warehouse table is no more recent than the last successful run of the job that builds it, or than that job's inputs, such as a GA4 export or a booking system's feed. The tutorial on moving a report onto a warehouse warns that a query nobody owns stops refreshing after an upstream change while the report goes on showing old figures, and the GA4 property handover checklist notes that an export billed to an outgoing owner's project stops when that project closes.
Fix: query the newest date in the table and then in each input until the dates stop matching; repair the step that failed, such as a query broken by a changed column, and rerun the missing days. An export relinked in a project the business controls starts again with the first full day after the new link, since nothing earlier is backfilled, so the days it missed stay missing.
Verify: the table's newest date is the one its schedule promises, and the report shows it once cached results expire.
Prevent: add a freshness test from the data tests tutorial, which flags a table whose newest date has fallen behind its source's arrival pattern and alerts the table's owner; set the source's freshness to match the table's refresh schedule, as the warehouse tutorial advises; and run each query after its inputs are expected to land, checking first that the input table exists, which the diagnosis for an export that does not match GA4 recommends for its reconciliation queries.
A date default pinned to a past period
A date control left on a fixed custom range, such as the period used to reconcile totals at launch, shows the same days to every viewer who touches nothing. The charts are full, which separates this from the empty-range cause in the chart diagnosis.
Fix: default the control to the period the page exists to answer, with a matching comparison, as the dashboard design guide sets out.
Verify: in a fresh session as a viewer, the default period has moved on since the day before.
Prevent: keep fixed ranges for like-for-like checks, out of viewers' defaults.
Recent days the source has not finished
The guide to ad platform data in a warehouse explains that a platform dates each conversion to the ad interaction it credits and keeps crediting that day until its attribution window closes. For GA4, the diagnosis for a property showing no data notes that standard reports read processed data and processing takes time, and the BigQuery export basics guide says a daily table is written only once Google completes processing that day, at no fixed hour. The lasting fault is a warehouse that loads each day once, which the ad platform guide shows holds every day at its first, incomplete figure.
Fix: read closed days for decisions, and reload a trailing window of recent platform days on every run instead of appending each day once.
Verify: a day that looked short at first matches the platform's figure once settled.
Prevent: write how long a day stays open to restatement into the metric's definition and mark those days provisional on the dashboard, as the metrics layer guide recommends, and compare periods at the same age, as the guide to comparing periods explains.
A delivery sent too early or orphaned by its creator
The sharing and embedding guide explains that a scheduled PDF is rendered with its creator's access, fixes the data at the moment of fetching, and stops once the creator's account is suspended after they leave. A delivery timed before the sources update carries the previous day or provisional figures.
Fix: move the start time after the last layer behind the delivered pages updates; recreate a leaver's schedule under a role account the business controls, sent to a group; and check any date range the schedule applies, which the guide notes its owner can set, because a fixed range pins the email as a fixed default pins the report.
Verify: the next attachment shows the same newest closed day as the live report at the time of sending.
Prevent: list each schedule in the inventory with its creator, start time and the layer it waits for.
A prevention routine: a freshness test and an owner for each layer
The remedy for every cause above is the same: a test or check on each layer, and a named person who answers for it.
| Layer | Test or check | Owner |
|---|---|---|
| Source and export | Closed days in its own interface; the export's status in the property's link settings | The property or platform owner |
| Load and query | A freshness test after each load, blocking the reporting tables on failure | The query's owner |
| Extract | Refresh on, timed after its input | Whoever built it |
| Data source, report and delivery | Freshness matched to the layer it reads; date defaults that move; a start time after the last update | The report's owner |
The data tests tutorial, linked under the warehouse cause, runs the freshness test between the load and the reporting tables, sends failures to a named owner, and marks affected dashboards stale when a blocking test fails, so readers hear about a stalled layer from the report itself. Keep a headline KPIA metric tied directly to an objective. off any layer that has no test.
The dashboard governance guide keeps an inventory with each dashboard's owner, sources and refresh timing; give each layer behind a dashboard its own line, so that no extract or delivery is left without an owner. Run the dashboard launch QA checklist at launch and whenever a layer changes, since it asks whether the cache fits the freshness the audience needs and whether the default date range is right. When a layer stalls, the five whys keeps asking until the answer is a missing check rather than a person: not "the load failed" but "an export changed and nothing watched the load".
A karting chain's Monday dashboard stops at Friday (illustrative)
Zantovrik Karting finds two stalled layers
Zantovrik Karting, an invented chain of 4 indoor karting tracks, opens its bookings dashboard at 09:00 on a Monday and finds the headline page ending on Friday; a page of bookings by track ends 3 weeks earlier. A refresh brings nothing newer, and the reporting table behind the headline page holds nothing after Friday: the booking system's nightly export added a column from Saturday's file onwards, and the scheduled query that loads the table has failed on both runs since. The track page reads an extract built at launch whose automatic refresh was never switched on. The analyst fixes the query, reruns Saturday and Sunday, and adds a freshness test that blocks the reporting table and alerts her when the newest booking date is more than 1 day old. The extract's refresh is switched on to run each morning after the load, with the analyst as owner. The inventory now lists 4 layers, each with a test or check and a named person, from the booking system's administrator for the export to the marketing lead for the data source.
Other measurement symptoms, from tags that never fire to numbers that disagree, start from the analytics troubleshooting collection.
Common mistakes
- Treating every old figure as a cache problem. Find the layer that stopped before changing any setting.
- Shortening freshness to cure a stalled load. It only re-reads a table that has not changed.
- Reading yesterday as final. The source may still be adding to it.
- A query, extract or schedule with no owner. When it stops, nobody hears.
- Timing a delivery before the load. The attachment carries the day before.
