Analysis evidence and limitations

On this page

An analysis combines source observations, calculated summaries, and generated explanations. Use the source and the measure’s definition to understand what a statement establishes, then record the business conclusion it supports.

Source observations

A supplied script can establish that a transformation selects a field, calls a routine, or applies an expression. Its context matters: an expression can occur inside a conditional branch, and an apparent input can be prepared by another step.

Follow the relevant flow before interpreting a single operation as the behavior of the whole system. In the monthly-revenue example, finding a conversion call establishes where conversion occurs; the called routine explains how the rate is chosen.

Calculated summaries

Counts, effort indicators, and confidence summaries describe defined populations. A count of transformations differs from a count of files. A mean confidence value differs from the proportion of fully detailed assets.

Read the explanation attached to the measure and retain its population when using the number in an assessment. Analysis indicators provides a worked calculation and explains missing values.

Generated explanations

Descriptions marked ✦ AI (announced as AI-inferred to screen readers) provide an interpretation of the available material. Use them to orient the review and formulate a question. Confirm business meaning against the source and with the people who maintain it.

For example, a description may call a distinct-selection operation “customer deduplication.” The source can establish that identical combinations of selected values are removed. Establishing one row per customer also requires the business key and the surrounding processing. Record that stronger conclusion only after checking them.

The ✦ AI? marker means the explanation’s origin was not recorded.

The absence of an inference marker is not, by itself, evidence that a statement was copied verbatim from source. Use the available source context and method explanation to determine its basis.

Behavioral contracts

A behavioral contract describes expected behavior associated with an analyzed asset. It helps frame what a later implementation would need to preserve or what a test should examine. Where the analysis provides a contract grade or coverage measure, read the associated evidence and eligible population.

A graded contract and an executed acceptance test are different artifacts. To accept the target’s behavior, the team still needs test cases, expected results, and an execution outcome. An asset without a grade has no graded evidence to include in that comparison.

Partial and missing information

An asset can be present with only part of its behavior described. A dependency can be recognized while its implementation remains outside the project. Treat these as specific gaps to investigate rather than discarding the entire analysis.

In the reporting example, the supplied aggregation may fully describe monthly grouping while currency-rate selection remains unresolved. The assessment can record the grouping rule and assign a separate action to obtain the routine.

Write a finding another reviewer can follow

Use a short record that connects the observation to the decision. The following is a synthetic example of an engagement note, not a product-generated result:

Item	Example
Question	Can monthly revenue move without changing currency behavior?
Observation	Revenue preparation calls a shared currency routine
Supporting material	The supplied reporting script and its routine call
Established conclusion	Conversion is a dependency of the proposed scope
Open question	Which rate date and fallback does the routine select?
Next action	Source owner supplies the routine; consultant reviews those rules

Keep this note with the scope and source revision. After resolving the question, update the conclusion and retain the evidence that changed it.