You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Exact duplicates are stored again although the README says duplicates are merged #707
The README says a near-duplicate is merged into the memory it restates instead of being filed beside it. An exact duplicate is stored again, and an identical automatic capture replayed seven times was stored seven times.
Evidence
Measured by an independent audit on 2026-10-10 in an isolated SQLite profile (main at b4a62b2) and read in the README.
README.md lines 140 and 141: "Novel content is written; a near-duplicate is merged into the memory it restates rather than filed beside it."
The audit wrote an exact duplicate of an existing memory: action: stored with a new id. A near-duplicate was stored too. The same Bash capture replayed 7 times produced 7 rows, and the duplicates pushed the target out of the top 3 for the test queries. consolidate did not merge the exact duplicates.
Related, separate from this issue: a contradicting memory written without supersedes_id is stored and ranked next to the original, with nothing detecting the contradiction.
Expected
Exact duplicates are detected by a content hash at write time (automatic capture included) and merged or refused with a visible reason. Near-duplicates follow what the README says or the README says what the write gate really does for deliberate writes. consolidate merges exact duplicates already in the store. A test that writes the same text seven times and expects one row.
Activity
fix(memory): refuse exact replays and consolidate legacy duplicates (#…
fix(memory): deduplicate exact replays and consolidate existing copies (
Merge origin/main into fix/706-auto-recall-provenance