Statistics

Statistics turns your Audit Logs into visual insight. Instead of reading executions one by one, you select a rule, a date range, and a field, and the page shows you how that field behaves across every matching execution: how values are distributed, how they change over time, and which individual logs sit behind the result. It is built into the workspace, so there is nothing to set up and nothing to export first.

Audit Logs enabled on the rule you want to analyze. That is the whole prerequisite. Statistics reads the logs your rules already produce, so any rule that has been logging has data ready to explore. The rule picker even marks it for you: a green dot means logs exist for your selected date range, an orange dot means there is nothing to analyze in that window yet.

It is the list of everything you can analyze. DecisionRules reads your Audit Logs and organizes the fields it finds into three scopes: Input, the data sent to the rule, Output, the data the rule returned, and Metadata, the technical detail around the execution such as rule name, version, status, etc. Search for a field, click it, and the statistics load.

Business questions and operational ones, from the same place. What share of applications get approved? Which customer segment dominates? How did the average quoted price move after last week's pricing change? Are error rates climbing? How long does this rule take at peak hours? If the value passed through the rule, it is available.

No. Picking a rule, a date range, and a field is the entire workflow. The charts are generated for you and the summary metrics adapt to the field you chose automatically.

Three, and they work together. Frequency Distribution counts how often each unique value appears, which shows you the composition of your decisions. Time Series groups executions into intervals and plots a metric across them, which shows you movement and trends. Raw Data Table lists the individual Audit Logs behind the analysis, which lets you verify anything you see in the charts.

Three headline numbers, chosen automatically based on the type of field you selected. Numbers give you minimum, maximum, and average. Booleans give you true count, false count, and true rate. Dates give you earliest, latest, and count. Strings give you unique values, most common, and count. Objects and arrays report how many values are filled and how large they typically are.

Yes. Enable the comparison switch, select a second rule, and both are visualized side by side. You can give the secondary rule its own date range, which makes period-over-period analysis just as easy as rule-over-rule. If the selected field only exists in one of the two rules, the page shows the primary rule alone rather than guessing.

Relative or absolute, the same options you already use on the Audit Logs page. The selection is shared between both tabs, so changing it in one applies it to the other. Every chart and every calculation reflects the period you picked.

Yes. The Data Dictionary looks inside objects and arrays and surfaces the fields it finds within them, so an array of customers gives you access to customer ID, name, or status directly. For bulk-style inputs, statistics can count values from each item in the array rather than treating the whole array as a single value, which is what you want when a single execution carries a batch of records.

Types are detected from your actual audit data rather than declared in advance, so they reflect what is really in the logs. Mixed means the same field appeared with different types across records. Unknown means the value was missing, empty, or could not be determined confidently. Both are usually worth a look: they often point at a data quality issue upstream that is easier to fix once you can see it.

Yes. Frequency Distribution, Time Series, and Raw Data Table all export to CSV. You get a ZIP containing the data itself plus a metadata file describing the export, so the numbers stay traceable once they leave the platform.

Yes, through the Business Intelligence API. The Integrations panel gives you the setup information for Power BI and generates REST API snippets in cURL, JavaScript, Python, PHP, .NET, and Java, already filled in with your selected BI API key, rules, and date range. Copy, paste, run.

Yes. The same BI API that powers the integration snippets can be called on a schedule to pipe audit data into your own data lake, SIEM, or reporting stack. Statistics is the fast way to explore, the API is the way to operationalize.

Start with your outputs. The decision itself, the score, the tier, the routing outcome. A frequency distribution on your main output field is the fastest way to see whether your rule behaves the way the business believes it does, and it very often surfaces a surprise in the first minute. Then look at inputs to understand who you are actually receiving, and metadata to check performance and errors.

Compare. Put the new version and the previous one side by side, or keep one rule and compare last week against this week. A shift in the distribution of outcomes right after a publish is exactly the signal you want to catch on day one rather than at quarter end.

Narrow your range. The Data Dictionary is built from a sample of recent logs, and the charts sample once the matching set grows past ten thousand records, which keeps the page fast on large workspaces. That sample is meant to be representative, and for exploring trends it is exactly right. When you need precision on a specific model version or a specific incident, tighten the date range or filter by rule version, and for exact figures on the full dataset use the Business Intelligence API.

Trust it, then verify it. That is what the Raw Data Table is for. When a distribution shows something unexpected, drop into the underlying logs and read the actual records. The path from an aggregate to the individual execution that caused it is one click, which is the whole point of having statistics inside the rule engine instead of three systems away.