fix: apply score_threshold filtering after fusion queries in local mode

[cbcoutinho]

Summary

• Apply score_threshold filtering after fusion queries in local mode to match server behavior
• Simplify and improve score threshold tests
• Add formula threshold test

This PR reimplements the fix from the main branch onto the dev branch.

Related: Closes original PR targeting main branch

---

This PR was generated with the help of AI, and reviewed by a Human

[netlify]

✅ Deploy Preview for poetic-froyo-8baba7 ready!

Name	Link
🔨 Latest commit	fb72e6d
🔍 Latest deploy log	https://app.netlify.com/projects/poetic-froyo-8baba7/deploys/69414dba19704500084186f7
😎 Deploy Preview	https://deploy-preview-1138--poetic-froyo-8baba7.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[coderabbitai]

📝 Walkthrough

Walkthrough

This pull request implements score_threshold filtering for fusion queries (RRF and DBSF) in the Qdrant client. The primary change adds a score_threshold filter in the _merge_sources method of local_collection.py, which prunes fused results before payload and vector retrieval. Supporting changes include two new test helper methods in the congruence test suite that validate score-thresholded fusion query variants, and two new integration tests that verify score_threshold behavior for both RRF and DBSF fusion types.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

• Local collection logic: The core change in _merge_sources introduces conditional score_threshold filtering after fusion computation. Requires verification that pruning occurs at the correct stage and that subsequent offset/pagination handling accounts for filtered results.
• Test coverage: New helper methods add test variants; new integration tests cover both RRF and DBSF with explicit threshold values. These additions are straightforward but multiple files are touched.

Possibly related issues

• [Local Mode] score_threshold not applied to FusionQuery results #1136: Directly addresses the bug where score_threshold was not being applied to FusionQuery/RrfQuery results in the _merge_sources method—this PR implements the fix for that exact issue.

Suggested reviewers

• coszio

Pre-merge checks and finishing touches

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 20.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'fix: apply score_threshold filtering after fusion queries in local mode' directly and clearly describes the main change: implementing score_threshold filtering after fusion queries in local mode.
Description check	✅ Passed	The description is directly related to the changeset, explaining the purpose of applying score_threshold filtering after fusion queries, simplifying tests, and adding formula threshold tests.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 0

🧹 Nitpick comments (1)

tests/congruence_tests/test_query.py (1)

431-444: Clarify the TODO comment for future maintenance.

The TODO indicates that score_threshold is not applied in formula queries in core. This test verifies that local mode matches this behavior (i.e., both should return the same unfiltered results). Consider rewording to make the intent clearer:

-            score_threshold=1.0,  # todo: score threshold is not applied in formula queries in core
+            score_threshold=1.0,  # Note: score_threshold is intentionally NOT applied in formula queries (matches core behavior)

This distinguishes it from a "to-do" item and clarifies it's documenting expected behavior rather than a bug to fix.

📜 Review details

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 144f4ea and fb72e6d.

📒 Files selected for processing (3)

• qdrant_client/local/local_collection.py (1 hunks)
• tests/congruence_tests/test_query.py (2 hunks)
• tests/test_in_memory.py (1 hunks)

🧰 Additional context used

🧬 Code graph analysis (2)

tests/test_in_memory.py (1)

qdrant_client/http/models/models.py (2)

• FusionQuery (962-963)
• Fusion (950-959)

tests/congruence_tests/test_query.py (3)

qdrant_client/http/models/models.py (6)

• ScoredPoint (2510-2521)
• Prefetch (2023-2047)
• RrfQuery (2466-2467)
• Rrf (2458-2463)
• FormulaQuery (945-947)
• MultExpression (1679-1680)

qdrant_client/async_qdrant_remote.py (1)

• query_points (364-469)

qdrant_client/http/api/search_api.py (2)

• query_points (550-565)
• query_points (773-788)

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (3)

• GitHub Check: Redirect rules - poetic-froyo-8baba7
• GitHub Check: Header rules - poetic-froyo-8baba7
• GitHub Check: Pages changed - poetic-froyo-8baba7

🔇 Additional comments (5)

qdrant_client/local/local_collection.py (1)

826-839: LGTM! Score threshold filtering correctly applied after fusion.

The implementation correctly:

1. Filters fused results by score >= score_threshold before payload/vector retrieval (efficient)
2. Applies offset after filtering (correct order of operations)
3. Excludes FormulaQuery from threshold filtering, matching server behavior

tests/test_in_memory.py (2)

123-208: Well-structured test for RRF score threshold filtering.

The test properly validates:

• Baseline behavior without threshold (all 5 points returned)
• Filtered behavior with threshold (only qualifying points)
• Per-point score assertions for correctness

The docstring clearly documents expected RRF scores for the test data.

---

211-296: Good parallel test for DBSF fusion with score threshold.

Follows the same pattern as the RRF test with appropriate threshold value (1.0) for DBSF score distribution. This ensures both fusion algorithms are covered.

tests/congruence_tests/test_query.py (2)

416-429: Good congruence test for RRF fusion with score threshold.

The test validates that local mode matches remote behavior for RRF queries with score_threshold. The inline comment documenting expected results (3 results: 1.0, 0.5, 0.3̅) is helpful for debugging failures.

---

1333-1338: LGTM! Congruence tests added for both fusion score threshold variants.

The new test methods are properly integrated into test_dense_query_fusion, ensuring local-remote consistency for score-thresholded fusion queries.

[joein]

@cbcoutinho thanks for addressing my comment and your contribution!