Turn Research Notes into an Evidence Matrix

Research notes rarely arrive in one tidy format. A team may have interview excerpts, observation notes, meeting recaps, screenshots, and unanswered questions. When these materials are merged into a single narrative too early, an observation can become an explanation and a repeated phrase can begin to look like a conclusion.

An evidence matrix slows that jump down. It places each research question beside its source, the observation actually recorded, the interpretation being considered, evidence that does not fit, and the gap that remains. The matrix is not a scoring system. It is a review surface that helps a team see how far each statement is from its source.

Start with one decision, not every note

Define the decision the matrix should support. A broad request such as “summarize the research” encourages the team to include everything. A narrower decision creates useful boundaries.

Decision:
Which part of the fictional neighborhood library's evening-event information should the team test first?

Permitted source set:
Three fictional interviews, one lobby observation, and one planning-meeting recap.

The decision does not ask whether the entire service is good or bad. It asks which information problem deserves a test. Notes about unrelated services can remain outside this matrix without being discarded.

Give every source a stable key

Create a key before copying an excerpt or paraphrase into the matrix. The key should identify the source and location well enough for an authorized reviewer to reopen it.

I01-L12: interview 1, line 12;
I02-L08: interview 2, line 8;
O01-18:42: observation 1 at 18:42;
M01-A3: meeting recap 1, action item 3.

A source key is not proof by itself. It is a route back to the material. If the source cannot be reopened, mark that limitation instead of presenting the note as fully verified.

Separate observation from interpretation

Use different columns for what the source records and what the team thinks it may mean. Consider this fictional observation:

Observation: O01-18:42
A visitor photographed the event poster, then asked the desk whether registration was required.

Interpretation under review:
The poster may not make the registration requirement clear.

The observation does not establish why the visitor took a photograph. It also does not prove that all visitors misunderstand the poster. Keeping the interpretation in its own column makes that distance visible.

If your material comes from a permitted site visit, first build an observation brief that separates what happened from what it might mean. The matrix can then combine those bounded observations with other sources.

Build the smallest useful matrix

A working matrix can begin with three rows. Add columns only when they help reviewers check the reasoning.

Question	Source and observation	Interpretation to examine	Counterevidence	Gap and next check
Is registration required?	O01-18:42: visitor asked the desk after reading the poster	Registration wording may be hard to find	I03-L06: participant registered without help	Test the poster with people who have not seen the event page
When does the event end?	I01-L12: participant asked whether “7 PM” was the start or finish	The time label may be ambiguous	No second source addresses the label	Ask three readers to state the start and end time from the current poster
Is the room accessible?	I02-L08: participant requested room-access information	Access details may need a consistent location	M01-A3 says staff planned to add details, but completion is unverified	Check the current page and poster; record what is present before revising

This fictional matrix does not rank the three questions by importance. It shows the evidence available for each one and the smallest next check that could reduce uncertainty.

Record counterevidence in the same row

Evidence that does not fit should not be moved into a footnote or removed from the summary. I03-L06 describes a participant who registered without help. That does not cancel O01-18:42, but it limits a claim that the requirement is universally unclear.

Counterevidence can reveal useful conditions. Perhaps the successful participant arrived through a detailed event page while the visitor saw only the poster. That is a plausible question for the next test, not a confirmed explanation.

Keep source types visible

An interview statement, an observed action, and a planning commitment answer different questions. Avoid treating them as interchangeable.

An interview can show what a participant said or recalled.
An observation can show a recorded action in a specific setting.
A meeting recap can show what the team planned or believed at that time.
A current page or artifact can show what is visible now.

M01-A3 says staff planned to add access information. It does not prove the information was added. The matrix should preserve that as a planned action until the current artifact is checked.

Do not turn filled cells into confidence scores

A row with four sources is not automatically stronger than a row with one source. Four notes may repeat the same unverified statement, while one direct observation may answer a narrow question clearly. Evaluate relevance, source quality, context, independence, and recency instead of counting filled cells.

If a team needs a status label, use evidence states that describe the work:

observed: present in the permitted source;
corroborated: supported by more than one relevant, independent source;
contradicted: credible evidence points in different directions;
unknown: the source set does not answer the question;
outdated: the evidence may no longer describe the current state.

These labels describe the evidence record. They do not convert a small qualitative set into a population estimate.

Preserve disagreements between people and sources

If two interviews offer different accounts, keep both. The difference may reflect different contexts, expectations, or experiences. A useful matrix can display the disagreement without forcing an average.

When synthesizing a discussion with multiple speakers, a viewpoint map can preserve who said what and where the views diverged. Bring the relevant, source-linked statements into the matrix only after those differences remain visible.

Turn each gap into a bounded next check

“More research needed” is too vague. A gap should name the missing evidence and a small way to obtain or verify it.

Gap:
The current poster has not been checked against the meeting commitment.

Next check:
Open the current poster version, record the registration and accessibility wording, and compare it with M01-A3.

Owner:
Content coordinator

Completion evidence:
Dated screenshot and source link

This makes the matrix operational without pretending that the check has already happened.

Write a conclusion that stays within the matrix

A bounded conclusion for the fictional example could read:

The reviewed notes contain one observed registration question, one direct question about the time label, and one request for accessibility information. A separate participant registered without help, and a planned accessibility update has not been verified against the current materials. The next step is to test the poster's registration and time wording and confirm whether the promised access details are present.

The conclusion names the evidence, the counterexample, the unverified commitment, and the next check. It does not claim that most library visitors share the same difficulty.

Copy this research evidence matrix template

Decision the matrix should support:
Permitted source set:
Date range:

Research question:
Source key and location:
Source type: interview / observation / meeting record / current artifact
Observation or direct statement:
Interpretation under review:
Counterevidence or conflicting source:
Evidence state: observed / corroborated / contradicted / unknown / outdated
Gap:
Smallest next check:
Owner:
Completion evidence:

Bounded conclusion:
What the sources support:
What they do not support:
Next decision:

Run the final traceability check

Before sharing the matrix, reopen each source key. Confirm that the observation is faithful to the source, the interpretation remains labeled, conflicting evidence is visible, and every claimed completed action has current proof. Remove any row that combines unrelated questions merely because the wording sounds similar.

Choose three research questions from material you are authorized to review. Build one row for each question with a source, an observation, an interpretation, a counterexample or limit, and a specific gap. Then write one bounded conclusion that another reviewer can trace back to the same evidence.