TM-10
Duplicate or Near-Duplicate Summary Detected
Type Misclassification
Default severity: medium
Product BacklogPrioritizationQualityBusiness Value
What it detects
Two or more active issues (not linked as duplicates) have summaries that are identical or highly similar — indicating either an unlinked duplicate (the same work is being tracked twice, inflating issue count and velocity) or a copy-paste template story that was never differentiated (see CH-15 for clone detection, this rule targets semantic similarity without clone links). summary_pairs = compute_pairwise_similarity(active_issues.summaries) // similarity metric: cosine similarity on TF-IDF vectors OR Levenshtein distance ratio
Detection formula
FOR each_pair (I1, I2) WHERE similarity > config.tm.duplicate_similarity_threshold // default: 0.85 AND I1 NOT linked_as_duplicate_of I2 THEN FLAG probable_duplicate Report: I1.key, I2.key, similarity_score, both summaries side by side IF both issues in same sprint THEN CRITICAL (double-counting velocity)Examples in practice
- A team shows two or more active issues (not linked as duplicates) have summaries that are identical or highly similar while duplicate similarity threshold is set to 0.85.
- Example signal: Two or more active issues (not linked as duplicates) have summaries that are identical or highly similar — indicating either an unlinked duplicate (the same work is being tracked twice, inflating issue count and velocity) or a copy-paste template story that was never differentiated (see CH-15 for clone detection, this rule targets semantic similarity without clone links).
Suggested response
Retype or rewrite summaries so issue types match the actual work described.
Coaching playbook
Symptom
Two or more active issues (not linked as duplicates) have summaries that are identical or highly similar — indicating either an unlinked duplicate (the same work is being tracked twice, inflating issue count and velocity) or a copy-paste template story that was never differentiated (see CH-15 for clone detection, this rule targets semantic similarity without clone links). summary_pairs = compute_pairwise_similarity(active_issues.summaries) // similarity metric: cosine similarity on TF-IDF vectors OR Levenshtein distance ratio
Why it matters
When "Duplicate or Near-Duplicate Summary Detected" keeps appearing, the team is signalling a repeatable process gap. Left unexamined, the pattern hides where work really stalls and makes improvement metrics harder to trust.
What you can achieve
Retype or rewrite summaries so issue types match the actual work described.
Facilitation questions
- What system change would stop "Duplicate or Near-Duplicate Summary Detected" from firing again?
- What do the cited issues have in common — same root cause or same workaround?
- Who owns the two-week experiment and how will we verify on the next import?
Run this rule against your own tracker data with Flow Analyzer.