From Google Analytics

Get the export

Two shapes work, because the importer resolves columns through a table of aliases rather than parsing a format:

• The BigQuery event export (one row per event, the richest source.
• A Data API or interface export: aggregate rows, one line saying this page had 412 views on this day.

CSV either way. Nothing needs reshaping first.

Import it

Upload it on the site’s Imports screen, or point the importer at a URL. Then watch the progress; a large export takes a while, and it resumes rather than restarting if it is interrupted.

Two properties are worth knowing before you start.

It is idempotent, by replacement. Every row it writes carries an import_id. Re-running the same import deletes that id’s rows first and writes them again, so an import run twice (by a retry, by a second press of the button, by a worker that died halfway) leaves exactly one copy. That column is also what makes an import reversible: undoing one is a delete.

It is resumable, through a staging table. The file is streamed into imports_raw verbatim first, one row per source record with its position in the file, and only then normalised into events. A job that died at row 400,000 resumes at 400,001, and a mapping bug is re-runnable without asking you to export again.

Aggregate rows become events

Micaforge stores pageviews, not counts. A row saying “412 views of /pricing on 3 March” expands into 412 events at the start of that day, capped at 10,000 per row and marked in sdk_version as an import, so an expanded row can always be told from a real one.

Which means: imported days have no hourly shape. Everything lands at midnight. If your export was aggregated, do not read the hour-of-day chart for the imported period, because it is an artefact of the import, not of your readers.

What will not match, and why

Set the expectation now, because the two tools are not measuring the same thing.

Sessions. GA4 restarts a session at midnight and on a campaign change. Micaforge uses a 30-minute inactivity gap and nothing else. Micaforge will usually show fewer sessions.

Users. GA4 has a cookie. Micaforge has a daily-rotating hash, so a returning reader tomorrow is a new visitor. Daily uniques are comparable; anything longer is not, and Micaforge will read higher over a month.

Bots. GA4 drops “known bots” silently. Micaforge names them, counts them and keeps them out of the human numbers unless you ask for them. Your human pageviews may go down, and that is the correction rather than a loss.

Answer engines. GA4 files ChatGPT and Perplexity referrals as referral. Micaforge makes them a top-level channel. Your referral number will drop and a new one will appear beside search.

After the import

Run both for a fortnight. Do not reconcile the totals; reconcile the shapes: the same pages in the same order, the same weekday rhythm, the same campaigns. When the shapes agree, the difference in the totals is the definitions, which is exactly what you were told above.

Then turn off the GA tag. Two analytics scripts is one script’s worth of page weight wasted.

What does not come over

Google’s own modelled data: thresholded rows, modelled conversions, anything GA4 estimated rather than observed. Micaforge imports what your export contains. It will not invent the parts Google withheld from it, and neither will it pretend the gap is not there.