Aglet

Investigate query spelling variants with paired result traces

An investigation should establish whether a variant is a harmless spelling difference or a different request. Replay literal and corrected phrases under the same scope, compare results and accepted wording, and vary locale only when the evidence suggests it changes meaning.

Build a useful investigation brief

  1. Build the variant matrix

    List original phrase, proposed correction, locale, product family, result titles, filters, and first useful action. Include a valid literal term and a no-result typo. Preserve submitted and displayed queries so silent changes are visible.

  2. Probe false matches

    Hold the correction fixed while testing a neighboring product family and a brand term. Compare overlap, missing variants, and unrelated results. Record whether a correction helps only after a category or language cue; that condition should remain part of the finding.

  3. Package the policy brief

    State the equivalent cases, ambiguous cases, and confidence boundary. Recommend a suggestion, mapping, or literal fallback with a pass condition. Separate catalog naming, tokenizer behavior, and user choice so a vocabulary finding does not claim an implementation cause.

What to carry forward

The investigation is useful when paired variants have a reproducible overlap pattern and named false matches. Stop with a scoped correction policy or an explicit evidence gap. A rescued result does not prove that every similar edit should be corrected.

Technical background: GOV.UK developer documentation.

Keep the decision with the work.

Use a Work Item in Aglet to record the problem, the evidence you have, and the next decision. Add an owner and priority, then keep updates in the discussion so the next person can follow the reasoning.

Create an account See the product workflow