Two people can look at the same video and disagree about whether it contains a test. That’s a research problem worth fixing before the spreadsheet gets bigger. We’ll use a small pilot, record the disagreements and apply one rubric across a monitor-review sample. Claude Code keeps that method reusable.
Kabo, an AI agent for creator research, provides source evidence. Claude Code maintains the codebook, pilot disagreements and conclusions affected by a definition change.
Prepare the materials for this Claude Code task
Keep the team’s labels and disagreements beside the source rows. This article is about making a shared comparison consistent, not choosing the next series from scratch.
[ ] channels.csv — Public channel URLs and why each belongs in the comparison. [ ] codebook.md — Definitions for format, viewer problem and evidence type. [ ] scope.md — Same dates for every channel; separate Shorts, streams and long-form. [ ] review-notes.md — Pilot disagreements and the decisions that resolve them. [ ] our-test-plan.md — Equipment and conditions we can document ourselves.
Connect Kabo, then check the project context
Open the content project, install Kabo for Claude Code and complete sign-in. For a direct server connection, follow the setup guide.
Have reviewers code the same small pilot
Our monitor-review team uses “test” in different ways. Before a full competitor collection, select a few relevant sources with Kabo and ask contributors to review the same evidence. Record their labels and reasons without forcing agreement.
If nobody inspected the footage, “the creator says they tested it” must not become “we observed the test.” A precise spreadsheet cannot repair an undefined category.
Read codebook.md and scope.md. Use Kabo to collect a small relevant pilot from channels.csv. Show the source URL, available evidence and proposed labels for 3 rows. Add the reviewer labels I supply to review-notes.md; if a reviewer has not responded, mark that review pending rather than inventing agreement. Wait before expanding the sample.
Resolve the meaning of the label before recoding
Ask which definition caused each disagreement. A useful rule distinguishes a claim about a test from an observed procedure and states what evidence is required. The team approves that rule; Claude Code can then apply it to affected rows.
Claimed test: the creator states that testing happened; procedure not independently observed in our evidence. Observed procedure: checked visual evidence shows the relevant setup or action; record what was inspected. Unclear: available evidence cannot establish either label confidently. Never promote a title, caption or spoken claim into an observed procedure.
A definition change that affects a conclusion
An illustrative pilot entry:
- Initial label: observed test.
- Available evidence: a transcript says “we tested this,” with no inspected procedure.
- Corrected label: claimed test.
- Affected conclusion: remove this row from the count of observed procedures.
- Still allowed: note that the creator describes a test, with source attribution.
The correction concerns evidence, not whether the product review was good or bad.
Find the rows and findings that depend on the old rule
Keep the shared dates, format boundaries and coverage notes for the actual collection. Then apply the approved definition change selectively. Claude Code should identify which rows used the old meaning and which summaries depend on them.
A lower count after recoding is not a change in competitor strategy. It may simply be a more careful definition.
Compare the previous and approved test-evidence definitions. Identify rows in the collected sample affected by the change; retain source references and evidence type. Recode those rows, list unresolved cases and revise only findings that depend on them. Save the codebook decision and a before/after findings note. Do not reinterpret a coverage change as a channel strategy shift.
Write findings from the agreed table
After the team reviews the corrected rows, write the findings with links to the accepted evidence. A separate production note can suggest our own monitor test, but it should not use the competitor count as proof that our product claim is true.
For an individual creator choosing one practical pilot, the Codex comparison guide keeps the decision narrower.
Keep the definition when the next sample arrives
Use a new dated source snapshot and the agreed codebook on the next pass. Record coverage differences separately. If a future example exposes another edge case, add it to the review notes before changing the definition again.
The codebook confused a spoken product claim with a visible test. Update that definition, identify affected rows and revise only the conclusions that depend on them.
Keep the codebook beside the accepted comparison
Review the evidence labels and disagreements in Google Sheets. The deliverable is a consistent coded study with a record of why the definition changed. For collection and evidence-linking mechanics, use the MCP competitor recipe.
Worked examples are illustrative. Product screenshots show existing Kabo web results, not a run of the example in this guide. Use your own materials and the evidence returned in your task.