# Decision Table Hit Policies: Choose What Multiple Matching Rules Should Mean

<a id="main-content"></a>

Operations &amp; Systems

October 28, 2026  ·  8 min read  ·  Victor Valentine Romo

On this page

1. [Write the Question Before Choosing a Policy](<#write-the-question-before-choosing-a-policy>)
2. [Compare the Five Documented Policies](<#compare-the-five-documented-policies>)
3. [Trace One Fictional Document Through Three Rules](<#trace-one-fictional-document-through-three-rules>)
4. [Separate Unique From Any With an Identical-Output Case](<#separate-unique-from-any-with-an-identical-output-case>)
5. [Move One Row and Recalculate First](<#move-one-row-and-recalculate-first>)
6. [Make the Consumer State What It Can Accept](<#make-the-consumer-state-what-it-can-accept>)
7. [Fill In a Hit-Policy Decision Record](<#fill-in-a-hit-policy-decision-record>)
8. [Preserve Cases That Challenge the Selection](<#preserve-cases-that-challenge-the-selection>)
9. [Sources](<#sources>)

Three rules match one document. One reviewer expects a single label. Another expects a list of required checks.

Both can read the conditions correctly and still expect different results. The table needs a decision about what matching means.

A hit policy supplies that evaluation rule. Choose it after the owner explains the result the workflow needs. A setting can resolve how rows produce an answer. It cannot establish whether those rows express an authorized business policy.

This guide uses [Camunda's decision-table documentation, version 8.10](<https://docs.camunda.io/docs/components/modeler/dmn/decision-table/>) for the table structure. Its worksheets and selection [procedure](</articles/sop-documentation-b2b/>) are proposed methods. Every document, rule, and result below is fictional. No example configures an engine or changes a live workflow.

<a id="write-the-question-before-choosing-a-policy"></a>

## Write the Question Before Choosing a Policy

Begin with the noun in the output. Is this table choosing one classification, listing required checks, or calculating a quantity?

Those questions can use similar inputs while requiring different answers. A document can have one primary review path and several independent review requirements. Putting both ideas in a field called review makes the result ambiguous before the first row runs.

Write the expected result shape in plain language. For example: one primary label, or a list of every applicable check. State whether no answer is permitted. State whether the sequence of several answers carries meaning.

Then name the policy owner. The person choosing the engine setting should work from that recorded intent. If intent remains disputed, keep the setting proposed.

The [automation workflows guide](</articles/automation-workflows-b2b-operations/>) describes work moving between steps. This review happens inside one decision. The next step still needs to know what its answer represents before acting on it.

Avoid choosing a policy because it makes today's error disappear. Preserve the overlapping case and ask what it means. A visible conflict provides information about the policy the workflow still lacks.

<a id="compare-the-five-documented-policies"></a>

## Compare the Five Documented Policies

[Camunda 8.10 documents these hit policies](<https://docs.camunda.io/docs/components/modeler/dmn/decision-table-hit-policy/>):

| Policy | Meaning for matching rules |
| --- | --- |
| Unique | At most one rule may match. Multiple matches violate the policy. |
| Any | Multiple matches are allowed only when their outputs agree. |
| First | Return the first matching row's output. |
| Rule Order | Return all matching outputs in table order. |
| Collect | Return all matching outputs without a defined order. |

Under Unique, matching rows with identical outputs still violate the one-match condition. Under Any, conflicting outputs violate the agreement condition. Camunda 8.10 also permits Collect aggregators and restricts an aggregated table to one output. This example uses plain Collect without aggregation. These documented behaviors are version-specific; confirm the engine contract before implementation.

Zero matches are valid under Unique. It does not guarantee exactly one answer. If the business requires one answer for every valid input, [review coverage separately](</articles/business-decision-table-gaps-overlaps/>).

The remaining examples reason from those definitions. They propose no default policy for every business decision.

<a id="trace-one-fictional-document-through-three-rules"></a>

## Trace One Fictional Document Through Three Rules

Suppose an editorial team classifies incoming documentation for review. Its synthetic document has category guide, twenty-four pages, and diagrams present.

The table contains three rows in the displayed order. Each row produces one text label.

| Rule | Category | Pages | Diagrams present | Output label |
| --- | --- | --- | --- | --- |
| R1 | Guide | Any positive count | Either value | Editorial check |
| R2 | Either permitted category | At least 20 | Either value | Extended check |
| R3 | Either permitted category | Any positive count | Yes | Diagram check |

The permitted categories are guide and reference. The document matches R1 because it is a guide. It matches R2 because twenty-four is at least twenty. It matches R3 because diagrams are present.

These labels describe possible checks. They do not authorize publication, accept document quality, or assign a reviewer. Keep the output boundary narrow while comparing evaluation policies.

Write the match set once: R1, R2, R3. Every comparison below uses that same input and those same conditions. Changing the input halfway through would make the policies harder to compare.

| Applied policy | Result of this paper trace |
| --- | --- |
| Unique | Policy violation: three rows match |
| Any | Policy violation: the three output labels differ |
| First | Editorial check |
| Rule Order | Editorial check, Extended check, Diagram check, in that sequence |
| Collect | All three labels; no sequence should be inferred |

A policy violation is not a label named manual review. The workflow may need a controlled response to the violation. Define that response separately before an implementation uses it.

<a id="separate-unique-from-any-with-an-identical-output-case"></a>

## Separate Unique From Any With an Identical-Output Case

Keep the same synthetic document and conditions. For this comparison only, suppose all three rows output specialist review.

There are still three matching rules. Unique still has a conflict because the match count exceeds one. Any has agreeing outputs and can supply specialist review.

The comparison isolates two different requirements. One asks whether conditions are mutually exclusive. The other asks whether overlapping conditions agree about the answer.

If the business expects exactly one primary classification, ask whether overlap should be forbidden or tolerated when labels agree. [Write that choice down](</articles/decision-journal-business-operators/>). Similar output text does not settle whether duplicated policy conditions are acceptable.

Also inspect what agreement means for the whole output. If the eventual table has several output fields, do not compare only the visible label in a [spreadsheet](</articles/spreadsheet-formula-audit-business/>). Require implementation evidence for the complete result under the chosen engine.

This example does not choose Any for the team. It demonstrates the question the owner must answer. Keep the identical-output case beside the conflicting-output case so the reviewer can see both conditions.

Avoid renaming three different review requirements specialist review merely to make them agree. That change would erase the distinction the original table was trying to express.

<a id="move-one-row-and-recalculate-first"></a>

## Move One Row and Recalculate First

Return to the original three different labels. Move R3 above R1 and R2 without changing a condition.

The synthetic document still matches the same three rules. Under First, the selected label changes from editorial check to diagram check. The earlier result depended on row order.

Under Rule Order, the returned sequence becomes diagram check, editorial check, extended check. All three outputs remain present. Under Collect, the outputs remain the same collection; the displayed row change provides no ordering promise.

Keep these three observations separate in the policy record. Selecting one answer, ordering several answers, and collecting several answers express different requirements.

If the owner wants First, ask why diagram review should outrank [editorial review](</articles/editorial-workflow-b2b-teams/>) in this proposed order. A maintainer sorting rows alphabetically should not become the person deciding that precedence.

Record an explicit order-change case in the review evidence. Show the original row sequence, revised sequence, original result, and revised result. The owner can then inspect the consequence of the proposed edit.

For Rule Order, ask whether output sequence merely expresses priority or instructs the downstream workflow to perform steps in order. Returning an ordered list alone does not establish how another system executes it.

The [API-first operations guide](</articles/api-first-business-operations/>) covers integration design. Keep this result contract attached when the list crosses a system boundary.

<a id="make-the-consumer-state-what-it-can-accept"></a>

## Make the Consumer State What It Can Accept

A list is not interchangeable with a single text value. Ask the receiving workflow [what result shape it expects](</articles/ai-structured-outputs-schema-business-validation/>) before choosing a multiple-result policy.

For the fictional document team, a primary-path field can hold one label. A review-checklist field can hold several labels. Decide whether these are separate decisions rather than forcing both into the same output.

If every applicable check is required, choosing First would leave two of the example's checks out of the answer. If one primary label is required, returning all three labels would leave selection unresolved for the consumer.

Use the synthetic input to make the disagreement concrete. Ask the owner to write the answer another person should see for this document. Ask the maintainer to show how that answer reaches its destination.

Plain Collect leaves ordering unspecified. If a consumer displays the first item as the main recommendation, that consumer has introduced another selection rule. Name and review it separately.

Likewise, avoid assuming a collected list represents unique business tasks. Two different rows can describe the same intended work. Decide how task identity should be handled before a downstream step creates work from each returned entry.

The policy record should identify these consumer assumptions. It cannot prove that a downstream implementation follows them. Keep that evidence as a separate review item.

<a id="fill-in-a-hit-policy-decision-record"></a>

## Fill In a Hit-Policy Decision Record

Use one record for each decision version. Keep the worksheet beside its synthetic cases and the approved plain-language rules.

| Record field | Question to answer |
| --- | --- |
| Decision question | What specific answer does the workflow need? |
| Output meaning | What does each returned value represent? |
| Allowed result count | Zero or one, exactly one, or several? |
| Overlapping-case evidence | Which input matches which rule identifiers? |
| Output agreement | Do the matched rules return the same complete output? |
| Order significance | Does row order select, prioritize, or have no business meaning? |
| Selected policy | Which documented policy fits the approved intent? |
| Consumer contract | Which field or step receives which result shape? |
| Owner and approval state | Who can authorize the choice, and is it approved? |
| Engine reference | Which product version and configuration will be verified? |

For a fictional checklist decision, a proposed entry could read: return every applicable check; several results allowed; document D24 matches R1, R2, R3; order has no business meaning; plain Collect proposed.

The documentation owner still has to accept that intent. The workflow maintainer still has to demonstrate the actual result shape in the selected implementation. A complete worksheet does not supply either approval automatically.

Use [consulting deliverable acceptance criteria](</articles/consulting-deliverable-acceptance-criteria/>) to define review evidence for the handoff. For example, require the trace and result to agree for each recorded case. Keep business acceptance distinct from a successful evaluation.

<a id="preserve-cases-that-challenge-the-selection"></a>

## Preserve Cases That Challenge the Selection

Carry more than the convenient three-match case into review. Add a document matching one row, a document matching no rows, and the identical-output comparison.

For this synthetic table, a reference document with ten pages and no diagrams matches no rows. A ten-page guide without diagrams matches only R1. Record these facts before specifying the surrounding workflow's response.

Missing category, unknown page count, or [unknown diagram status](</articles/null-blank-zero-business-data/>) needs its own case. Do not assume the absent input means no condition matched. Require evidence from the actual engine and an approved policy for the surrounding workflow.

Keep policy violations visible in expected results. Replacing every expected violation with a generic fallback can hide the very disagreement the review is meant to expose.

When rows change, repeat the cases under the recorded policy. [Preserve a version reference](</articles/spreadsheet-version-control-business-decisions/>) so a reviewer can tell which conditions and ordering produced each result.

The review is complete when the owner can explain the selected answer and the maintainer can demonstrate it against the agreed cases. [Broader automation](</articles/operations-automation-audit/>) still requires its own implementation review.

Choose the result the work requires.

<a id="sources"></a>

## Sources

- [Camunda 8.10: Decision-table overview](<https://docs.camunda.io/docs/components/modeler/dmn/decision-table/>). Inputs, outputs, and rules. Checked 2026.10.28.
- [Camunda 8.10: Decision-table hit policy](<https://docs.camunda.io/docs/components/modeler/dmn/decision-table-hit-policy/>). Unique, Any, First, Rule Order, Collect, and aggregator constraints. Checked 2026.10.28.
