{
 "title": "Do E5 and BGE embedding models need their query and passage prefixes?",
 "run": "2026-11-05",
 "script": "https://mixpeek.com/measurements/2026-11-05-embedding-query-prefixes.py",
 "guide": "https://mixpeek.com/guides/do-e5-and-bge-embeddings-need-query-and-passage-prefixes",
 "setup": "Exact cosine search over the full corpus of each BEIR test set; sentence-transformers 5.1.1 on Apple MPS; normalized embeddings; batch 64. CI: paired bootstrap over queries, 2,000 resamples, seed 0.",
 "metric_notes": "nDCG@10 with graded gains 2^rel-1; recall@10 over min(relevant,10); exact cosine over the full corpus",
 "results": [
  {
   "dataset": "scifact",
   "model": "intfloat/e5-small-v2",
   "arm": "documented (query: / passage:)",
   "query_prefix": "query: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.6708,
   "recall@10": 0.8046,
   "queries": 300,
   "docs": 5183,
   "s": 25.7
  },
  {
   "dataset": "scifact",
   "model": "intfloat/e5-small-v2",
   "arm": "no prefixes",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.677,
   "recall@10": 0.8112,
   "queries": 300,
   "docs": 5183,
   "s": 24.9,
   "ndcg_diff_vs_documented": 0.0061,
   "ndcg_diff_ci95": [
    -0.0151,
    0.0265
   ],
   "recall_diff_vs_documented": 0.0067
  },
  {
   "dataset": "scifact",
   "model": "intfloat/e5-small-v2",
   "arm": "query prefix only",
   "query_prefix": "query: ",
   "doc_prefix": "",
   "ndcg@10": 0.6813,
   "recall@10": 0.8112,
   "queries": 300,
   "docs": 5183,
   "s": 0.2,
   "ndcg_diff_vs_documented": 0.0105,
   "ndcg_diff_ci95": [
    0.0001,
    0.0221
   ],
   "recall_diff_vs_documented": 0.0067
  },
  {
   "dataset": "scifact",
   "model": "intfloat/e5-small-v2",
   "arm": "passage: on both sides",
   "query_prefix": "passage: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.6181,
   "recall@10": 0.7529,
   "queries": 300,
   "docs": 5183,
   "s": 0.2,
   "ndcg_diff_vs_documented": -0.0528,
   "ndcg_diff_ci95": [
    -0.0758,
    -0.0307
   ],
   "recall_diff_vs_documented": -0.0516
  },
  {
   "dataset": "scifact",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "documented (instruction on queries)",
   "query_prefix": "Represent this sentence for searching relevant passages: ",
   "doc_prefix": "",
   "ndcg@10": 0.7077,
   "recall@10": 0.8362,
   "queries": 300,
   "docs": 5183,
   "s": 20.8
  },
  {
   "dataset": "scifact",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "no prefixes",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.7243,
   "recall@10": 0.8412,
   "queries": 300,
   "docs": 5183,
   "s": 0.2,
   "ndcg_diff_vs_documented": 0.0165,
   "ndcg_diff_ci95": [
    0.0032,
    0.0304
   ],
   "recall_diff_vs_documented": 0.005
  },
  {
   "dataset": "scifact",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "instruction on both sides",
   "query_prefix": "Represent this sentence for searching relevant passages: ",
   "doc_prefix": "Represent this sentence for searching relevant passages: ",
   "ndcg@10": 0.7108,
   "recall@10": 0.8469,
   "queries": 300,
   "docs": 5183,
   "s": 25.1,
   "ndcg_diff_vs_documented": 0.003,
   "ndcg_diff_ci95": [
    -0.0059,
    0.012
   ],
   "recall_diff_vs_documented": 0.0107
  },
  {
   "dataset": "scifact",
   "model": "sentence-transformers/all-MiniLM-L6-v2",
   "arm": "documented (no prefixes)",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.6479,
   "recall@10": 0.79,
   "queries": 300,
   "docs": 5183,
   "s": 6.7
  },
  {
   "dataset": "scifact",
   "model": "sentence-transformers/all-MiniLM-L6-v2",
   "arm": "e5-style query: / passage:",
   "query_prefix": "query: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.6495,
   "recall@10": 0.7907,
   "queries": 300,
   "docs": 5183,
   "s": 6.6,
   "ndcg_diff_vs_documented": 0.0015,
   "ndcg_diff_ci95": [
    -0.011,
    0.0142
   ],
   "recall_diff_vs_documented": 0.0007
  },
  {
   "dataset": "nfcorpus",
   "model": "intfloat/e5-small-v2",
   "arm": "documented (query: / passage:)",
   "query_prefix": "query: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.3239,
   "recall@10": 0.2986,
   "queries": 323,
   "docs": 3633,
   "s": 22.0
  },
  {
   "dataset": "nfcorpus",
   "model": "intfloat/e5-small-v2",
   "arm": "no prefixes",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.3208,
   "recall@10": 0.3003,
   "queries": 323,
   "docs": 3633,
   "s": 20.6,
   "ndcg_diff_vs_documented": -0.0032,
   "ndcg_diff_ci95": [
    -0.0105,
    0.0038
   ],
   "recall_diff_vs_documented": 0.0018
  },
  {
   "dataset": "nfcorpus",
   "model": "intfloat/e5-small-v2",
   "arm": "query prefix only",
   "query_prefix": "query: ",
   "doc_prefix": "",
   "ndcg@10": 0.3227,
   "recall@10": 0.2994,
   "queries": 323,
   "docs": 3633,
   "s": 0.3,
   "ndcg_diff_vs_documented": -0.0012,
   "ndcg_diff_ci95": [
    -0.0068,
    0.0041
   ],
   "recall_diff_vs_documented": 0.0008
  },
  {
   "dataset": "nfcorpus",
   "model": "intfloat/e5-small-v2",
   "arm": "passage: on both sides",
   "query_prefix": "passage: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.2947,
   "recall@10": 0.2802,
   "queries": 323,
   "docs": 3633,
   "s": 0.3,
   "ndcg_diff_vs_documented": -0.0292,
   "ndcg_diff_ci95": [
    -0.0392,
    -0.0199
   ],
   "recall_diff_vs_documented": -0.0184
  },
  {
   "dataset": "nfcorpus",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "documented (instruction on queries)",
   "query_prefix": "Represent this sentence for searching relevant passages: ",
   "doc_prefix": "",
   "ndcg@10": 0.3423,
   "recall@10": 0.3148,
   "queries": 323,
   "docs": 3633,
   "s": 18.2
  },
  {
   "dataset": "nfcorpus",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "no prefixes",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.3429,
   "recall@10": 0.3159,
   "queries": 323,
   "docs": 3633,
   "s": 0.3,
   "ndcg_diff_vs_documented": 0.0006,
   "ndcg_diff_ci95": [
    -0.0068,
    0.0084
   ],
   "recall_diff_vs_documented": 0.0011
  },
  {
   "dataset": "nfcorpus",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "instruction on both sides",
   "query_prefix": "Represent this sentence for searching relevant passages: ",
   "doc_prefix": "Represent this sentence for searching relevant passages: ",
   "ndcg@10": 0.3478,
   "recall@10": 0.321,
   "queries": 323,
   "docs": 3633,
   "s": 20.2,
   "ndcg_diff_vs_documented": 0.0055,
   "ndcg_diff_ci95": [
    -0.0003,
    0.0115
   ],
   "recall_diff_vs_documented": 0.0062
  },
  {
   "dataset": "nfcorpus",
   "model": "sentence-transformers/all-MiniLM-L6-v2",
   "arm": "documented (no prefixes)",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.3187,
   "recall@10": 0.3057,
   "queries": 323,
   "docs": 3633,
   "s": 5.0
  },
  {
   "dataset": "nfcorpus",
   "model": "sentence-transformers/all-MiniLM-L6-v2",
   "arm": "e5-style query: / passage:",
   "query_prefix": "query: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.3068,
   "recall@10": 0.297,
   "queries": 323,
   "docs": 3633,
   "s": 6.3,
   "ndcg_diff_vs_documented": -0.0119,
   "ndcg_diff_ci95": [
    -0.0197,
    -0.0045
   ],
   "recall_diff_vs_documented": -0.0087
  },
  {
   "dataset": "fiqa",
   "model": "intfloat/e5-small-v2",
   "arm": "documented (query: / passage:)",
   "query_prefix": "query: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.3725,
   "recall@10": 0.4471,
   "queries": 648,
   "docs": 57638,
   "s": 244.1
  },
  {
   "dataset": "fiqa",
   "model": "intfloat/e5-small-v2",
   "arm": "no prefixes",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.3564,
   "recall@10": 0.4292,
   "queries": 648,
   "docs": 57638,
   "s": 216.8,
   "ndcg_diff_vs_documented": -0.0162,
   "ndcg_diff_ci95": [
    -0.0269,
    -0.0059
   ],
   "recall_diff_vs_documented": -0.0179
  },
  {
   "dataset": "fiqa",
   "model": "intfloat/e5-small-v2",
   "arm": "query prefix only",
   "query_prefix": "query: ",
   "doc_prefix": "",
   "ndcg@10": 0.364,
   "recall@10": 0.4411,
   "queries": 648,
   "docs": 57638,
   "s": 2.3,
   "ndcg_diff_vs_documented": -0.0085,
   "ndcg_diff_ci95": [
    -0.0167,
    -0.0006
   ],
   "recall_diff_vs_documented": -0.006
  },
  {
   "dataset": "fiqa",
   "model": "intfloat/e5-small-v2",
   "arm": "passage: on both sides",
   "query_prefix": "passage: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.1514,
   "recall@10": 0.1959,
   "queries": 648,
   "docs": 57638,
   "s": 1.8,
   "ndcg_diff_vs_documented": -0.2211,
   "ndcg_diff_ci95": [
    -0.2438,
    -0.1985
   ],
   "recall_diff_vs_documented": -0.2512
  },
  {
   "dataset": "fiqa",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "documented (instruction on queries)",
   "query_prefix": "Represent this sentence for searching relevant passages: ",
   "doc_prefix": "",
   "ndcg@10": 0.4006,
   "recall@10": 0.4646,
   "queries": 648,
   "docs": 57638,
   "s": 166.5
  },
  {
   "dataset": "fiqa",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "no prefixes",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.3837,
   "recall@10": 0.4397,
   "queries": 648,
   "docs": 57638,
   "s": 2.4,
   "ndcg_diff_vs_documented": -0.0169,
   "ndcg_diff_ci95": [
    -0.0261,
    -0.0069
   ],
   "recall_diff_vs_documented": -0.0249
  },
  {
   "dataset": "fiqa",
   "model": "BAAI/bge-small-en-v1.5",
   "arm": "instruction on both sides",
   "query_prefix": "Represent this sentence for searching relevant passages: ",
   "doc_prefix": "Represent this sentence for searching relevant passages: ",
   "ndcg@10": 0.3896,
   "recall@10": 0.4486,
   "queries": 648,
   "docs": 57638,
   "s": 201.4,
   "ndcg_diff_vs_documented": -0.0111,
   "ndcg_diff_ci95": [
    -0.0193,
    -0.0022
   ],
   "recall_diff_vs_documented": -0.016
  },
  {
   "dataset": "fiqa",
   "model": "sentence-transformers/all-MiniLM-L6-v2",
   "arm": "documented (no prefixes)",
   "query_prefix": "",
   "doc_prefix": "",
   "ndcg@10": 0.3687,
   "recall@10": 0.4395,
   "queries": 648,
   "docs": 57638,
   "s": 82.3
  },
  {
   "dataset": "fiqa",
   "model": "sentence-transformers/all-MiniLM-L6-v2",
   "arm": "e5-style query: / passage:",
   "query_prefix": "query: ",
   "doc_prefix": "passage: ",
   "ndcg@10": 0.354,
   "recall@10": 0.4211,
   "queries": 648,
   "docs": 57638,
   "s": 105.2,
   "ndcg_diff_vs_documented": -0.0148,
   "ndcg_diff_ci95": [
    -0.0235,
    -0.0059
   ],
   "recall_diff_vs_documented": -0.0184
  }
 ]
}