Clarification on segments: total points exceed inserted count, partial indexing inside

[sfmqrb]

I’m running an insertion-only workload (no deletes/updates). I’ve inserted ~9M vectors, but when I sum num_points across all segments, I see ~12M vectors. My assumption is that I’m counting redundant vectors but they don't seem to be old segments still logged during/after merge/index opt.

Questions:

1. Are all these “extra” vectors actually being queried?
   In other words, is there any read amplification where the same vector is searched twice?

2. Why do some segments appear partially indexed?
   In my benchmarks, for example, in the snapshot below, some segments show num_points = X but number_of_indexed_vectors = Y < X. Is this expected? I thought a segment only could be either indexed or plain.

Example snapshot:

2025-11-21T21:30:46  Segment 38  num_vectors=3,230,715  indexed=0
2025-11-21T21:30:46  Segment 39  num_vectors=2,561,133  indexed=675,154
2025-11-21T21:30:46  Segment 41  num_vectors=1,595,285  indexed=0
2025-11-21T21:30:46  Segment 43  num_vectors=2,076,912  indexed=0
2025-11-21T21:30:46  Segment 42  num_vectors=1,351,322  indexed=1,351,322

[sfmqrb]

@timvisee @generall
Hey,
Will you please take a look at this?