Rate each row against a defined scale.

Define what a higher score actually means

Start by naming one property you want to measure from the text. Bug-report reproducibility is a useful example because each level can describe evidence a reader can identify: a symptom, an action, a partial procedure, a complete procedure, and confirmation that the issue repeats. The score describes the report's detail, not the impact of the bug.

Write descriptions for the middle levels as carefully as the endpoints. A scale with 1 labeled Poor and 5 labeled Excellent leaves most decisions undefined. Explain what has to be present for a row to move from 2 to 3 or from 3 to 4. Avoid criteria that require information your selected columns do not contain.

For a cumulative rubric, later levels build on earlier requirements. In the example above, mentioning a browser version does not earn a 5 if the message provides no reproduction procedure. The row must satisfy the lower requirements as well as the additional requirement for the highest level. This makes missing evidence easier to handle consistently.

Use scores to compare evidence across rows

For bug reports, a reproducibility score can help you find reports that need a follow-up question before investigation. A low score indicates missing reproduction detail, while a high score indicates more of the defined evidence is present. Neither score tells you whether the reported bug is real or how quickly it should be fixed.

For feature requests, a rationale score can separate an unsupported preference from a request that explains a workflow and provides a concrete example. The result can support a research review queue. It does not measure customer value, market demand, or the cost of implementing the request unless those are separately defined and evidenced.

For operational requests, specificity scores can reveal what information is missing. A request for a resolution that names an action and a clear acceptance condition is easier to act on than a general complaint. Use the score to decide what to inspect or ask next, and retain the original text for that follow-up.

Check the boundaries between neighboring levels

A useful preview includes examples near the boundaries, not just an obvious 1 and an obvious 5. Compare rows assigned to adjacent levels and ask which requirement separates them. If the distinction is hard to explain, revise the rubric. Adding more levels does not help when the existing levels are already difficult to distinguish.

Do not combine unrelated dimensions into one scale. A message can be polite but vague, urgent but detailed, or strongly worded without describing a serious problem. If you need to evaluate both detail and urgency, use two definitions and keep their results in separate columns after export.

Keep the score separate from confidence. A result of 5 means the row was assigned to level 5 of your scale. It does not mean the tool is certain. Review low-confidence results and spot-check other levels before relying on the distribution. Counts by level are often easier to interpret than an average when the levels do not represent equal intervals.

Should you Categorize, Score, or Flag?

The same message can support different questions. Choose the output you need before choosing a tool.

Categorize

Which kind is it? Choose one label from a set, such as Billing or Shipping.

Score

To what degree does it meet the rubric? Assign a defined level, such as 1 through 5 for reproduction detail.

Flag

Does it meet this condition? Return Yes or No, with uncertain results marked for review.

A message about a broken export might be categorized as Technical, scored for the detail of its reproduction steps, and flagged for an explicit refund request. Keep those questions separate so each result remains interpretable.

Try Score on your spreadsheet

1. Add and preview your data. Open a tool below, then paste from Excel or Google Sheets or upload a CSV. Check the header row and column names.
2. Choose the source columns. Select the fields that contain evidence for your question. Leave out unrelated information and any existing answer column.
3. Define the result. Edit the starting definition and name the new column. Make sure the rules cover the examples and edge cases you expect.
4. Preview, review, and export. Preview a sample before processing the remaining rows. Inspect uncertain results, correct mistakes, then copy or download the results for your spreadsheet.