How Raveki understands your data
Not all data is born equal. Understand the difference between historical, observed, and volatile — and why it matters for trusting what you see.
Not all data has the same origin — so it shouldn't carry the same weight either. A reading based on 3 months of history is different from one based on 3 days of live monitoring, and you should be able to tell them apart.
The three origins of a data point
- Historical — your account’s past, imported the moment the integration was connected. It’s real and useful, but hasn’t been observed “live” yet.
- Observed — data directly monitored, day after day, since the integration was connected — already closed and stable.
- Volatile — very recent data (usually today or yesterday) that the source platform itself — Google, Meta, etc. — might still revise over the following days. It isn’t wrong; it just hasn’t settled yet.
There's a fourth label, but only at the comparison level (which joins two periods, never a single data point): if the two sides don't share the same origin — for example, this week, already observed, against the same week last year, which only exists as historical — the comparison is marked Calibrating. It isn't a fourth origin for a data point; it's what happens when the three above cross paths within the same comparison.
These labels show up throughout the platform — on a signal, on a comparison, on a card in your workspace — to indicate exactly what each conclusion is standing on.
Why this matters
Consider a Google Analytics account connected with 3 months of history. Presenting that as “still no activity” would be inaccurate — there’s already plenty to analyze. But treating that history as live monitoring would be equally misleading, since nothing has actually been observed since connecting. So the two stay distinct: historical describes the past; observed describes ongoing watch from now on.
There’s a third case. In today’s or yesterday’s numbers on a Google Ads account, for example, it’s common for the values themselves to keep being adjusted over the following hours or days — a click logged late, a conversion attributed out of order. That data is labeled volatile: it isn’t hidden or held back until it settles, but it’s flagged, so no final conclusion gets drawn before it does.
How a data point matures
Today
Volatile
Confidence window closes
varies by platform
Closed
Observed
The confidence score
Every signal and every comparison carries a confidence score, which rises with the proportion of observed, stable data behind it. A reading built on volatile data starts with lower confidence, precisely because it can still change; once those days close and become observed, confidence follows.
That number literally reflects how much has been seen and how settled it is — never an assumption drawn from the past. It’s also why precision improves over time: not through some mysterious learning process, but through the daily accumulation of observed, stable facts.