Skip to content

Limits and Permissions

2 min read

Oracle is deliberately narrower than it first appears. This page is the full account.


Never more than you can. Oracle reads its permissions from the same rules that guard each screen, so if your role cannot open Defects, you cannot ask about them.

That extends to conditions. A question like test cases that have an open defect against them returns test cases, but it reads defects to answer — so a role without defect access cannot ask it, and cannot learn case-by-case which ones have defects. The same applies to readiness columns: they are computed from defects, so a role that cannot read defects cannot get one in a column or sort by it.

Records you cannot read are not filtered out of the answer. They are never part of the vocabulary Oracle is given, so the question cannot be formed in the first place.


It never sees a record. No test case title, no defect description, no comment, no step text. Those never reach the model.

What it is shown is your project’s vocabulary: what the fields are called and the values they hold — your folder names, your label names, your release titles, your team members — together with your question. That is what lets it turn “in the Checkout folder” into a real condition.

Everything after the question is turned into a query — validation, execution, permissions, the explanation — runs with no model involved at all.


QuestionAnswer
Which tester is fastest?Unanswerable — nothing records it
How many defects will we have next month?Unanswerable — it does not predict
Show me the important onesOne question back, with answers to select
Defects assigned to Sarah, where nobody is called SarahA refusal, not an empty table

The last one matters most. An empty table reads as good news; a refusal reads as a typo.


  • No arithmetic. No sums, averages or grouping. Oracle returns the records or counts them. Calculations are Observatory.
  • Conditions combine with “and”. There is no “or” and no bracketed logic. A question needing two answers is two questions.
  • One sort column.
  • No pagination. A hundred rows by default, five hundred if you ask for more. Deeper than that is an export.
  • Rich text is not searched. Test case step bodies and shared step text are not part of what a condition can match.

Oracle’s paid operations carry per-workspace daily caps, reset each day:

OperationAllowance
Asking a question1,000 a day
Drawing a chart500 a day
AI readings of a readiness explanation100 a day

These are shared across everyone in the workspace, not per person, and exist to catch a runaway loop rather than to ration normal use.

Corrections are free and do not count. Removing a condition, sorting, asking for more rows, folding in readiness columns, and redrawing a chart another way all re-run without a model. Answering a clarification does count, because it plans the question again.