Ask AI
Understand the scope and limits of AI explanations.
45. What data is included when I use Ask AI?
Ask AI uses selected sequences from the chosen pattern to prepare summarized evidence, such as common attribute values and relationships.
Cortex receives that aggregate evidence and your prompt. The generated evidence payload does not include raw sequence traces, internal sequence IDs, step positions, or the matched timeline as sequence metadata.
Source payload attributes can still contain sensitive values. Aggregation is not a blanket guarantee of anonymization.
46. How do I control the number of sequences AI analyzes?
In Ask AI, set No. of sequences to include (limit). Available choices are 5, 10, 20, and 50; the default is 10.
The limit selects up to that many sequences for evidence preparation. Fewer available sequences means fewer are used. Ten sequences can contain many more than ten records.
47. Does changing the AI sequence limit change my query results?
No. It changes the evidence prepared for the AI request. It does not rerun the original sequence query, remove matches, or change its results.
It can change the AI explanation because a different amount of source evidence is summarized.
48. Are the examples given to AI randomly selected?
No. Sequence selection is ordered by EventOwner and sequence identifier, then limited to the requested count.
It is a deterministic selection rule, not a representative statistical sample. Different sequences can come from the same owner.
49. Can AI’s explanation be treated as a conclusion about every matching sequence?
No. It is based on evidence from the selected sequences, which may not represent the whole pattern.
The evidence focuses on payload attributes rather than raw chronological traces. Do not assume it proves causation or explains step timing that was not supplied in the evidence.
50. How can I verify an AI explanation against the underlying records?
Inspect sample sequences and their payloads, then use Copy SQL to examine more of the result data if needed.
Check whether the stated attributes appear in the records and whether the claim holds beyond the selected examples. For claims about order or delay, inspect timestamps and event timing directly.