Failure search

Search indexed mistake sections when reviewer wording does not map cleanly to known labels.

Use indexed search when a reviewer phrase does not map cleanly to a known label. This example searches only mistakes sections, then reads the exact section for one candidate guideline.

QUERY=${QUERY:-"uncertainty shown as exact values confidence interval misleading"}

Search mistake sections

uvx --from 'chartcoach[index]@latest' chartcoach catalog find \
  --mode fts \
  --where "role = 'section.mistakes'" \
  --limit 3 \
  --format table \
  "${QUERY}"

Output, shortened for the page:

Rank	Guideline id	Matched role	Matched text
1	`use-standard-uncertainty-display-with-text-explanation`	`section.mistakes`	Showing a confidence interval without any textual explanation.
2	`do-not-use-caption-only-fixes-for-ci-charts`	`section.mistakes`	Add definitions or extra text about prediction intervals while leaving the confidence-interval marks unchanged.
3	`show-prediction-intervals-for-treatment-effect-judgments`	`section.mistakes`	Keep 95% confidence-interval bars and assume readers will mentally convert them into outcome variability.

The index reports roles as section.<role>. Use the manifest role name when reading the exact record.

Read the selected section

GUIDELINE_ID=use-standard-uncertainty-display-with-text-explanation

uvx chartcoach@latest catalog read "${GUIDELINE_ID}" \
  --section mistakes \
  --source-detail minimal \
  --format markdown

Use JSONL with jq when a script or agent should consume search candidates:

uvx --from 'chartcoach[index]@latest' chartcoach catalog find \
  --mode fts \
  --where "role = 'section.mistakes'" \
  --limit 3 \
  --format jsonl \
  "${QUERY}" |
  jq -s 'map({id, title, role: .matched_role, text: (.matched_text | split("\n")[0])})'

Run catalog read or catalog cite before using a search result as evidence in an answer, critique, or generated page.

Search mistake sections

Read the selected section

On this page