Document scoring and ranking behavior across major versions

Author: parveensaini

lucene/CHANGES.txt

@@ -159,6 +159,8 @@ Other
 * GITHUB#14607: Index open performs version check on each segment, ignores indexCreatedVersionMajor (Rahul Goswami)
 * GITHUB#15454: Restore binary readability for 8.x indexes like previously and throw IndexFormatTooOldException for an older default codec not found on classpath (Rahul Goswami)
 
+* GITHUB#15586: Document that scoring and ranking may change across major Lucene versions, and that applications requiring stable ranking should explicitly configure Similarity. (Parveen Saini)
+
 ======================= Lucene 10.4.0 =======================
 
 API Changes

lucene/MIGRATE.md

@@ -122,6 +122,17 @@ As part of GITHUB#13820, GITHUB#13825, GITHUB#13830, this issue corrects DataInp
 to be public and not-final, allowing subclasses to override it. This change also removes the protected
 DataInput.readGroupVInt method: subclasses should delegate or reimplement it entirely.
 
+### Scoring and ranking across major versions
+
+Lucene does not guarantee identical scoring or document ranking behavior across
+major version upgrades. Changes may occur due to improvements or modifications
+in Similarity implementations, query execution, or related internals.
+
+Applications that require stable ranking behavior across upgrades should
+explicitly configure the Similarity used for indexing and searching rather than
+relying on defaults. Reviewing `CHANGES.txt` for intervening major versions is
+recommended when upgrading.
+
 ### OpenNLP dependency upgrade
 
 [Apache OpenNLP](https://opennlp.apache.org) 2.x opens the door to accessing various models via the ONNX runtime.  To migrate you will need to update any deprecated OpenNLP methods that you may be using.