Deduplication / Consensus

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

Two reviewers report a cache-miss null dereference in fetchItem; another reports an unbounded remote timeout there. Teaching null reports share one path, while timeout has distinct trigger/remedy despite the same function.

Mechanism

Record symbol, trigger, mechanism and consequence, then compare evidence. Merge the same issue while retaining sources; preserve distinct conditions/remedies. Explain severity/disagreement from impact rather than averages/votes. Map every source finding to the final inventory.

Bad example

Merge everything in fetchItem, choose severity by majority and discard independent timeout trigger/remedy.

Good example

Compare teaching (symbol,trigger,failure) and actual paths. Merge duplicate cache-miss null findings with both references; retain remote-timeout bounds separately. Record impact disagreements without severity averaging or file-based deduplication. Verify sufficient final counterparts for every source obligation.

Why the change matters

Nearby locations establish adjacency, not shared cause. Semantic mapping makes reports concise without dropping independent failure/remedy or treating consensus as truth.

Observable expectation

The illustrative list has one null and one timeout issue, retaining two null sources and the timeout bound. A different null trigger needs fresh comparison, not function-name merging.

Inspect retained remedies/impacts/disagreements.

Limits

Keys assist but require code evidence; identical titles/keywords do not prove cause. Partial overlap can retain separate obligations. Provenance is unconfirmed and model agreement does not replace verification.