# Business Decision Tables: Find Gaps and Overlapping Rules Before Automation

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

Operations &amp; Systems

October 28, 2026  ·  7 min read  ·  Victor Valentine Romo

On this page

1. [Name One Decision and Its Authority](<#name-one-decision-and-its-authority>)
2. [Define the Input Domains Before Writing Rows](<#define-the-input-domains-before-writing-rows>)
3. [Put a Deliberately Defective Table on Paper](<#put-a-deliberately-defective-table-on-paper>)
4. [Trace Below, At, and Above the Boundary](<#trace-below-at-and-above-the-boundary>)
5. [Keep No Match Separate From Manual Review](<#keep-no-match-separate-from-manual-review>)
6. [Use a Coverage Worksheet With Open Questions](<#use-a-coverage-worksheet-with-open-questions>)
7. [Repair the Policy Before Translating Its Syntax](<#repair-the-policy-before-translating-its-syntax>)
8. [Review Changes as New Decisions](<#review-changes-as-new-decisions>)
9. [Sources](<#sources>)

A document has twenty pages. One rule covers documents below twenty pages. Another covers documents above twenty pages.

The person reviewing the [spreadsheet](</articles/spreadsheet-formula-audit-business/>) can guess what happens next. The rules provide no answer.

A business decision table gives that missing answer a place to surface before someone [connects the workflow to an API](</articles/webhook-architecture-b2b/>). Write the inputs, enumerate the rules, and trace cases through every condition. Record where nothing matches and where several rules match.

[Camunda's decision-table overview, version 8.10](<https://docs.camunda.io/docs/components/modeler/dmn/decision-table/>) describes a table with inputs, outputs, and rules. The worksheet below proposes a business review method around that structure. It is not an engine configuration guide. Every document and policy in the example is fictional.

<a id="name-one-decision-and-its-authority"></a>

## Name One Decision and Its Authority

Start with one question: which review path should this document enter?

Keep document storage, reviewer assignment, notification, and acceptance outside this table. Those activities may follow the answer. Combining them with classification makes it harder to identify which policy produced a disputed result.

Name the person authorized to approve the review policy. Separately name the person translating it into expressions. The implementer can identify an uncovered case. The implementer does not acquire authority to decide its business treatment by finding it first.

Use a heading that travels with the table:

| Decision record | Fictional entry |
| --- | --- |
| Question | Which review path should an incoming document enter? |
| Policy owner | Documentation owner |
| Implementation owner | Workflow maintainer |
| Proposed version | DOC-REVIEW-1 |
| Effective state | Proposed, not approved for use |
| Inputs | Document category and page count |
| Outputs | Standard review, extended review, manual review |
| Exclusions | Reviewer assignment, document acceptance, and publication |

An effective version identifies the rules that apply to a particular evaluation. A draft can have a version without being authorized. Keep its approval state visible beside the version.

The [API-first operations guide](</articles/api-first-business-operations/>) covers connections between systems. Define this decision before choosing the connection that carries its answer. A working integration can transmit an unresolved policy faithfully.

<a id="define-the-input-domains-before-writing-rows"></a>

## Define the Input Domains Before Writing Rows

For this example, category has two permitted values: `Guide` and `Reference`. These spellings match the table cells exactly. Page count is a positive whole number. Count the document's pages using the agreed submitted version.

That domain excludes a page count of zero, a negative count, a fraction, and an unreadable file with unknown length. It also excludes a category labeled template. Exclusion does not establish the correct handling of those records. Write a separate intake question for each condition that could arrive.

State whether case, surrounding spaces, or alternate labels change the category. The proposed worksheet requires the exact agreed labels. Any [normalization](</articles/unicode-normalization-business-record-matching/>) needs its own documented mapping. Do not silently treat a new label as an existing category because the words seem related.

Give page count a unit and counting convention. A file's byte size does not answer a page-count question. If the submitted document changes, [preserve which version supplied the count](</articles/spreadsheet-version-control-business-decisions/>).

Treat missing input separately from a permitted value. Unknown category is not reference. Unknown length is not zero. The [null, blank, and zero guide](</articles/null-blank-zero-business-data/>) explains why those distinctions belong in the [data contract](</articles/csv-data-contract-business-handoffs/>).

These are proposed business definitions. They make the example inspectable without claiming how every decision engine handles absent or invalid input.

<a id="put-a-deliberately-defective-table-on-paper"></a>

## Put a Deliberately Defective Table on Paper

Suppose the documentation owner provides this first draft. The review threshold is twenty pages. The threshold is invented for this example.

| Rule | Category | Page count condition | Output |
| --- | --- | --- | --- |
| R1 | Guide | Less than 20 | Standard review |
| R2 | Guide | Greater than 20 | Extended review |
| R3 | Reference | At least 1 | Standard review |
| R4 | Reference | At least 20 | Extended review |

Read every row as a complete condition. Both the category and page-count condition must hold in this paper exercise. Matching one cell does not match the row.

Keep stable rule identifiers. Row numbers change when someone sorts the sheet. A worksheet that names R4 remains understandable after R4 moves above R3.

The guide rows leave twenty uncovered. The reference rows overlap from twenty upward. Those are separate observations, even though both involve the same threshold.

Do not repair the table while tracing it. [Preserve the defective version](</articles/late-arriving-data-report-revisions/>) and record its behavior first. Otherwise the team loses the evidence explaining why the revision exists.

A broad default row can conceal these defects. It may make the result look complete while leaving the intended classification unresolved. First determine what the existing conditions say.

<a id="trace-below-at-and-above-the-boundary"></a>

## Trace Below, At, and Above the Boundary

Use nineteen, twenty, and twenty-one pages for each category. These cases sit immediately beside the threshold within the positive-integer domain.

| Case | Category | Pages | Matched rules | Observation |
| --- | --- | --- | --- | --- |
| G19 | Guide | 19 | R1 | One match: standard review |
| G20 | Guide | 20 | None | Gap: no rule supplies an output |
| G21 | Guide | 21 | R2 | One match: extended review |
| R19 | Reference | 19 | R3 | One match: standard review |
| R20 | Reference | 20 | R3, R4 | Overlap: standard and extended review |
| R21 | Reference | 21 | R3, R4 | Overlap: standard and extended review |

For G20, twenty is neither less than twenty nor greater than twenty. For R20, twenty satisfies both at least one and at least twenty. The same overlap remains at twenty-one.

Write the matched identifiers before choosing an output. A reviewer who writes only extended review for R20 has already introduced [a selection policy](</articles/decision-table-hit-policy-business-rules/>) absent from this draft.

[Camunda 8.10's hit-policy documentation](<https://docs.camunda.io/docs/components/modeler/dmn/decision-table-hit-policy/>) makes result selection a separate part of table evaluation. This worksheet records matches before that selection. Keep that distinction when moving from paper to an engine.

Boundary cases are useful evidence. They are not proof that every input combination is covered. Add category checks, minimum values, [invalid records](</articles/crm-data-quality-automation/>), and any additional boundary introduced by a revision.

<a id="keep-no-match-separate-from-manual-review"></a>

## Keep No Match Separate From Manual Review

No match means the current rows supplied no answer for the tested input. Manual review means the approved policy deliberately selected a human review path.

Those statements have different evidence. G20 currently has no matching rule. Writing manual review into its observed-output cell would invent a rule that the table does not contain.

A workflow can stop for investigation after an evaluation fails to produce an answer. Record that stop as workflow behavior. Do not use it to rewrite the underlying decision result.

If the owner wants every valid input to receive one classification, the owner must resolve G20. If the owner wants a separate exception path, describe its eligibility and responsibility explicitly.

Missing input deserves another trace. For a guide with unknown page count, record the missing value as unknown. Do not evaluate it as nineteen, zero, or false in this paper worksheet.

Ask the implementation owner to demonstrate actual engine behavior for missing inputs before release. Ask the policy owner what the workflow should do with the record. These are related questions with different owners.

For example, the proposed intake policy might hold an unreadable document for correction. That proposal becomes a rule only after the owner approves its scope and expected behavior.

<a id="use-a-coverage-worksheet-with-open-questions"></a>

## Use a Coverage Worksheet With Open Questions

Keep one worksheet per table version. Preserve expected output independently from the output observed during the trace. Where policy is unresolved, write unresolved.

| Case | Inputs | Matched rules | Expected output | Observed output | Unresolved policy question | Owner |
| --- | --- | --- | --- | --- | --- | --- |
| G20 | Guide, 20 pages | None | Unresolved | No rule output | [Which side includes 20?](</articles/histogram-bins-business-reports/>) | Documentation owner |
| R20 | Reference, 20 pages | R3, R4 | Unresolved | Two candidate outputs | Should only one review path apply? | Documentation owner |
| G19 | Guide, 19 pages | R1 | Standard review | Standard review | None for this case | Workflow maintainer |
| M1 | Guide, page count unknown | Not evaluated in paper trace | Unresolved | Missing required input | Hold, reject, or seek correction? | Documentation owner |
| T1 | Template, 19 pages | None under listed categories | Unresolved | Category outside domain | Add category or return for correction? | Documentation owner |

For M1, the worksheet deliberately makes no engine prediction. Preserve that qualification when copying the case into an implementation ticket.

Record evidence beside each resolved question. [A dated policy decision](</articles/decision-journal-business-operators/>) can authorize a revised boundary. A developer's successful run can demonstrate implemented behavior. One record cannot substitute for the other.

Use the [dispatch-system guide](</articles/dispatch-system-client-management/>) when the accepted classification needs a [work queue](</articles/work-item-aging-small-service-teams/>) and an owner. Keep queue assignment separate from the evidence that established the classification.

<a id="repair-the-policy-before-translating-its-syntax"></a>

## Repair the Policy Before Translating Its Syntax

Suppose the fictional owner decides that documents below twenty pages receive standard review. Documents of twenty pages or more receive extended review, for both permitted categories.

Version DOC-REVIEW-2 replaces R2's greater-than condition with at-least-twenty. It replaces R3's at-least-one condition with less-than-twenty. R1 and R4 retain their conditions.

The revised table has one matching row for each of the six boundary cases. G19 and R19 receive standard review. G20, G21, R20, and R21 receive extended review.

That repair addresses the positive-integer domain. It does not resolve unknown counts or template documents. Keep M1 and T1 open until the owner makes those separate decisions.

Under the revised domain, coverage can also be reasoned about directly. Every positive integer is either below twenty or at least twenty. No positive integer satisfies both comparisons. Each allowed category has both regions represented.

That reasoning supports this small table. It does not justify a blanket [completeness](</articles/data-quality-dimensions-business-reports/>) claim for a future table with more categories, dates, or conditions.

The implementation owner can now translate the approved comparisons into the chosen engine's syntax. Preserve the approved plain-language rule beside the implementation reference so reviewers can compare intent with behavior.

<a id="review-changes-as-new-decisions"></a>

## Review Changes as New Decisions

A future threshold change needs a new version and new boundary cases. If the owner changes twenty to thirty, test twenty-nine, thirty, and thirty-one for each permitted category.

Retain the earlier cases as well. They show whether a change affects records away from the new boundary. State which version applies to documents already waiting for review.

A new category also expands the input domain. Adding template to the dropdown does not create template rules. Confirm its coverage before the intake form can submit it to the classifier.

The handoff is ready for implementation review when the policy owner has resolved its questions and the maintainer can reproduce its cases. Keep unresolved items visible through that review.

Count the matches. Name the gap. Give the decision an owner.

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

## Sources

- [Camunda 8.10: Decision-table overview](<https://docs.camunda.io/docs/components/modeler/dmn/decision-table/>). Table structure. Checked 2026.10.28.
- [Camunda 8.10: Decision-table hit policy](<https://docs.camunda.io/docs/components/modeler/dmn/decision-table-hit-policy/>). Evaluation semantics boundary. Checked 2026.10.28.
