refactor(memory-search): drop explanatory comments from hybrid fusion

yetval · yetval · commit c02d11dc7fdc · 2026-07-13T12:44:17.000-04:00
The dropMediaTextSignal name and the regression coverage in hybrid.test.ts
already describe each gating condition, so the block comments added no
information the code did not carry. No logic change.
diff --git a/extensions/memory-core/src/memory/hybrid.ts b/extensions/memory-core/src/memory/hybrid.ts
@@ -182,23 +182,11 @@ export async function mergeHybridResults(params: {
         : entry.exactPathSpecificity > 0
           ? 0
           : entry.pathScore;
-    // Make fusion modality-aware: non-text media (image/audio) only carries a
-    // synthetic label as text, so its keyword signal is structurally near zero.
-    // For such candidates drop the text weight and renormalize by the remaining
-    // vector weight so the score collapses to the vector signal on the same
-    // [0,1] scale as text candidates. Only drop the signal when the candidate
-    // has a vector signal but no keyword signal, so a keyword-only or both-signal
-    // media hit keeps its text-weighted score. Gate the drop on a positive
-    // configured vector weight so a valid keyword match is never removed when
-    // vectorWeight is 0.
     const dropMediaTextSignal =
       entry.hasVector &&
       !entry.hasKeyword &&
       params.vectorWeight > 0 &&
       params.isNonTextMediaPath?.(entry.path) === true;
-    // Renormalize only for the vector-only media case. Every other candidate
-    // keeps the established unnormalized weighted formula, so custom weights
-    // that do not sum to one preserve their score scale and result ordering.
     const contentScore = dropMediaTextSignal
       ? entry.vectorScore
       : params.vectorWeight * entry.vectorScore + params.textWeight * keywordScore;
