Redundant and Conflicting Rules
As a policy set grows, rules start to overlap. Two administrators write the same rule a month apart. A narrow rule sits behind a broad one that always wins first. HYPR AgentPass™ compares your rules against each other and reports how they relate, both when you save a draft and across the whole set in the Redundancy view.
The comparison is behavioral, not textual. Two expressions that differ only in the order of their operands are the same rule. Two that differ by a single character in a string can be entirely different rules. AgentPass therefore replays both rules over the same window of your own traffic, and compares what each would have matched.
Reviewing redundancy is an administrator task.
How Two Rules Can Relate
Open Oversight → Policies, and click the Redundancy tab. Policies are compared periodically, by replaying recent traffic. Until the first pass has run, the tab reads Policies have not been compared yet.
Two rules can relate in four ways. The Redundancy tab tags each pair with a short label. The AI wizard's Review step words the same relationship as a phrase about your draft. The following table lists both.
| Relationship | Tag on the Redundancy tab | Phrase in the wizard | What it tells you |
|---|---|---|---|
| Identical | Same rule | does the same thing as this draft | Both rules match the same requests. The second adds no new coverage |
| Narrower | Contained by | is narrower than this draft | Everything one rule catches, the other already catches |
| Broader | Contains | is broader than this draft | One rule also catches requests the other misses |
| Overlapping | Overlaps | overlaps this draft | Both match some of the same requests, and each catches some the other misses |
In the wizard, each finding offers Update that rule instead. It closes the wizard and opens the existing rule on the policy form, so you can adjust one rule instead of saving two competing ones. Nothing from your draft is carried over. You can still save your draft as a separate policy. See Check for Rules You Already Have.
The Two Kinds of Claim
AgentPass makes two different claims, and the wording tells you which one you are reading:
- "These are the same rule." The comparison reached this from the expressions themselves, independently of any traffic. It holds whatever your agents do next.
- "Both would have matched the same requests over the last window." The comparison replayed both rules against recorded traffic. That is evidence about what happened, not proof about what will happen. Two rules that behaved identically for a week can diverge on the eighth day.
Neither claim is stronger than its wording. Read the second as a measurement of your traffic, not a statement about the rules.
Read the Redundancy Tab
Each row is a pair of rules. The table has the following columns.
| Column | What it shows |
|---|---|
| Rules | The two rules, with a sentence describing how they relate |
| Relation | The tag from the preceding table |
| Evidence | The rules match when the verdict came from the expressions themselves, or N shared requests when it came from replaying traffic |
| Flags | Never runs when one rule can never fire behind the other, and Conflict when both rules are enforced and disagree about traffic they both catch |
Rules that never run are listed first, then conflicts. Banners above the table count them, such as One rule can never fire. When the comparison could not reach every pair, a banner reads Checked N of M possible pairs. Anything not listed here was not compared.
Where your role is limited to your own policies, the tab lists only pairs where you own both rules.
Conflicts and Rules That Never Run
Policies are evaluated in priority order, and the first enforcing block or hold wins. A rule can therefore be enabled and listed while doing nothing at all. The Redundancy view reports this separately from overlap, because it is the finding most worth acting on.
A rule is reported as never running when a higher-priority rule that blocks or requires approval matches everything that rule would have matched. Both rules must be enforcing. The higher-priority rule must also have matched all of the lower one's traffic in the window. The report names the rule that wins.
An explicit Allow never stops evaluation. To the engine, a policy that matches with Allow is the same as a policy that did not match, so evaluation continues to the next rule, whatever the priority. A higher-priority Allow therefore cannot stop the rule behind it from running. Only a higher-priority Block or Require Approval can.
Never Matched Is Not a Recommendation to Delete
A rule that has not fired may be protecting against something that has not happened yet, which is what prevention looks like from the inside. It may also be scoped to a tool nobody uses anymore, or written against a condition that can never be true. To see whether a rule has matched, expand its row on the Rules tab. See See Whether a Rule Has Ever Fired.
Deciding which case applies needs knowledge of your organization that the gateway does not have.
When the Result Is Undetermined
Undetermined is a real answer, not a finding. It means the comparison could not gather the evidence to judge the pair, not that the rules are clean. A pair the comparison could not judge does not appear in the table.
Some rules act on a provider's streamed response, and recorded responses are stored up to a size limit. When a rule depends on a part of a response that was cut at that limit, a replay cannot reproduce what the live gateway saw. The comparison then reports that it could not judge, instead of reporting that nothing matched.
The view counts these pairs in a banner above the results, which reads N pairs could not be compared, because the recorded traffic was too incomplete to replay. Changing the rule does not help. An empty findings table is therefore never mistaken for a clean policy set.
Where Comparison Runs
Comparison runs in two places, covering different ground:
- When you save a draft. A bounded check runs against your existing rules, so a duplicate is caught before it exists, not found later. It compares a limited number of the closest candidates, and tells you how many rules it compared, so you know when the answer is partial.
- On a schedule, across the whole set. A pass over every pair of policies produces the Redundancy view. It replays the last seven days of traffic, and refreshes about every six hours, not on demand.
Both are advisory. AgentPass never merges, deletes or reprioritizes a rule for you.
Related
- Understanding Your Policies — plain-language summaries and match counts
- Policy Coverage — connector and tool pairs no rule reaches
- Policies — priority, modes and how evaluation order works