AI search engine and MCP server for bots and agents. One Rust binary indexes code, docs, logs, JSONL/CSV and PDFs for BM25 full-text, vector and hybrid search, code search without grep, agent memory and RAG. Elasticsearch-compatible API; Corpus Hub serves hundreds of reference memory corpora.
Rust agent-memoryai-agentsbm25claude-codeclicode-searchcursordatabaseelasticsearchfull-text-searchhybrid-searchjsonllocal-firstmcpmcp-servermodel-context-protocolragrustsearch-enginevector-search
bench: run 11 — 45W/44L (filtered-agg + mixed-fold fixes) — variance draw, real fixes landed
xerj-org5461aaf64dbc
diff --git a/demo/playbooks/BENCHMARK_VS_ES.md b/demo/playbooks/BENCHMARK_VS_ES.mdindex 412acc11a..c3dcb75bd 100644--- a/demo/playbooks/BENCHMARK_VS_ES.md+++ b/demo/playbooks/BENCHMARK_VS_ES.md@@ -6,97 +6,97 @@ Verdict is from XERJ's POV: WIN = XERJ better (lower latency / higher docs·s /| dimension | XERJ | ES | ratio | verdict ||---|--:|--:|--:|:--:|-| ingest 100k × c1 (docs/s) | 105,919 | 70,980 | 1.49× | WIN |-| ingest 100k × c8 (docs/s) | 306,345 | 275,136 | 1.11× | WIN |-| ingest 1m × c1 (docs/s) | 111,135 | 70,065 | 1.59× | WIN |-| ingest 1m × c8 (docs/s) | 291,802 | 382,198 | 0.76× | LOSE |-| read q: match_all (p50 ms) | 2.11 | 2.92 | 1.38× | WIN |-| read q: match_none (p50 ms) | 1.80 | 2.18 | 1.21× | WIN |-| read q: match(model) (p50 ms) | 1.93 | 2.49 | 1.29× | WIN |-| read q: match_phrase(top_doc) (p50 ms) | 1.16 | 2.56 | 2.20× | WIN |-| read q: match_phrase_prefix (p50 ms) | 1.17 | unsupported (400) | — | N/A |-| read q: match_bool_prefix (p50 ms) | 2.11 | 6.23 | 2.95× | WIN |-| read q: multi_match (p50 ms) | 1.69 | 6.90 | 4.09× | WIN |-| read q: combined_fields (p50 ms) | 1.27 | unsupported (400) | — | N/A |-| read q: query_string (p50 ms) | 1.33 | 10.71 | 8.07× | WIN |-| read q: simple_query_string (p50 ms) | 2.04 | 2.86 | 1.40× | WIN |-| read q: more_like_this (p50 ms) | 2.16 | 2.62 | 1.22× | WIN |-| read q: term(status) (p50 ms) | 2.18 | 2.79 | 1.28× | WIN |-| read q: terms(model) (p50 ms) | 1.99 | 5.54 | 2.78× | WIN |-| read q: range(latency_ms) (p50 ms) | 2.18 | 6.08 | 2.79× | WIN |-| read q: range(@timestamp) (p50 ms) | 1.97 | 2.33 | 1.19× | WIN |-| read q: range(cost_usd) (p50 ms) | 2.16 | 7.63 | 3.53× | WIN |-| read q: prefix(model) (p50 ms) | 2.60 | 10.80 | 4.15× | WIN |-| read q: wildcard(model) (p50 ms) | 2.41 | 11.07 | 4.59× | WIN |-| read q: regexp(model) (p50 ms) | 2.29 | 11.22 | 4.89× | WIN |-| read q: fuzzy(model) (p50 ms) | 2.52 | 2.74 | 1.09× | WIN |-| read q: exists(cost_usd) (p50 ms) | 2.04 | 1.59 | 0.78× | LOSE |-| read q: ids (p50 ms) | 2.45 | 1.18 | 0.48× | LOSE |-| read q: term(cache_hit) (p50 ms) | 1.82 | 2.08 | 1.14× | WIN |-| read q: bool must+filter+should+must_not (p50 ms) | 2.50 | 17.61 | 7.06× | WIN |-| read q: constant_score (p50 ms) | 1.85 | 2.15 | 1.16× | WIN |-| read q: boosting (p50 ms) | 2.37 | 35.73 | 15.10× | WIN |-| read q: dis_max (p50 ms) | 2.33 | 6.34 | 2.72× | WIN |-| read q: function_score (p50 ms) | 2.23 | 38.48 | 17.22× | WIN |-| read q: pinned (p50 ms) | 1.78 | 15.31 | 8.60× | WIN |-| read agg: avg (p50 ms) | 1.68 | 1.42 | 0.84× | LOSE |-| read agg: sum (p50 ms) | 2.61 | 1.61 | 0.62× | LOSE |-| read agg: min (p50 ms) | 1.71 | 2.45 | 1.43× | WIN |-| read agg: max (p50 ms) | 2.11 | 1.98 | 0.94× | LOSE |-| read agg: stats (p50 ms) | 1.97 | 1.87 | 0.94× | LOSE |-| read agg: extended_stats (p50 ms) | 2.06 | 1.58 | 0.77× | LOSE |-| read agg: value_count (p50 ms) | 2.26 | 1.92 | 0.85× | LOSE |-| read agg: cardinality (p50 ms) | 2.28 | 1.73 | 0.76× | LOSE |-| read agg: percentiles (p50 ms) | 2.46 | 1.56 | 0.64× | LOSE |-| read agg: percentile_ranks (p50 ms) | 2.64 | 1.84 | 0.70× | LOSE |-| read agg: median_absolute_deviation (p50 ms) | 2.55 | 1.82 | 0.71× | LOSE |-| read agg: matrix_stats (p50 ms) | 2.09 | 1.29 | 0.62× | LOSE |-| read agg: scripted_metric (p50 ms) | 2.45 | 1.42 | 0.58× | LOSE |-| read agg: top_hits (sub) (p50 ms) | 1.84 | 2.27 | 1.23× | WIN |-| read agg: terms (p50 ms) | 2.03 | 1.74 | 0.86× | LOSE |-| read agg: rare_terms (p50 ms) | 1.68 | 2.07 | 1.24× | WIN |-| read agg: significant_terms (p50 ms) | 1.56 | 1.24 | 0.80× | LOSE |-| read agg: histogram (p50 ms) | 1.22 | 1.49 | 1.22× | WIN |-| read agg: date_histogram (p50 ms) | 1.58 | 1.45 | 0.91× | LOSE |-| read agg: auto_date_histogram (p50 ms) | 1.69 | 1.21 | 0.71× | LOSE |-| read agg: variable_width_histogram (p50 ms) | 1.31 | 2.86 | 2.18× | WIN |-| read agg: range (p50 ms) | 2.36 | 2.49 | 1.05× | WIN |-| read agg: date_range (p50 ms) | 0.94 | 1.99 | 2.12× | WIN |-| read agg: filter (p50 ms) | 1.90 | 1.99 | 1.05× | WIN |-| read agg: filters (p50 ms) | 2.37 | 2.03 | 0.86× | LOSE |-| read agg: missing (p50 ms) | 2.68 | 1.84 | 0.69× | LOSE |-| read agg: global (p50 ms) | 2.38 | 1.93 | 0.81× | LOSE |-| read agg: adjacency_matrix (p50 ms) | 2.49 | 2.05 | 0.82× | LOSE |-| read agg: composite (p50 ms) | 1.73 | 2.23 | 1.29× | WIN |-| read agg: random_sampler (p50 ms) | 2.35 | 5.29 | 2.25× | WIN |-| read agg: terms+avg(cost) (p50 ms) | 2.08 | 2.27 | 1.09× | WIN |-| read pipe: sum_bucket (p50 ms) | 1.51 | 1.94 | 1.28× | WIN |-| read pipe: avg_bucket (p50 ms) | 1.62 | 1.88 | 1.16× | WIN |-| read pipe: max_bucket (p50 ms) | 2.19 | 2.35 | 1.08× | WIN |-| read pipe: stats_bucket (p50 ms) | 1.93 | 2.17 | 1.13× | WIN |-| read pipe: percentiles_bucket (p50 ms) | 1.86 | 2.44 | 1.31× | WIN |-| read pipe: derivative (p50 ms) | 2.24 | 3.16 | 1.41× | WIN |-| read pipe: cumulative_sum (p50 ms) | 2.69 | 2.18 | 0.81× | LOSE |-| read pipe: moving_fn (p50 ms) | 2.59 | 1.72 | 0.66× | LOSE |-| read pipe: serial_diff (p50 ms) | 1.79 | 1.98 | 1.11× | WIN |-| read pipe: bucket_script (p50 ms) | 2.13 | 1.63 | 0.77× | LOSE |-| read pipe: bucket_selector (p50 ms) | 2.15 | 1.61 | 0.75× | LOSE |-| read pipe: bucket_sort (p50 ms) | 2.44 | 1.62 | 0.66× | LOSE |-| read feat: sort-heavy (p50 ms) | 2.50 | 17.65 | 7.06× | WIN |-| read feat: deep from+size (from 500) (p50 ms) | 2.08 | 1.80 | 0.87× | LOSE |-| read feat: search_after (p50 ms) | 4.20 | 19.28 | 4.59× | WIN |-| read feat: highlight (p50 ms) | 2.18 | 2.87 | 1.32× | WIN |-| read feat: _count (p50 ms) | 2.04 | 2.18 | 1.07× | WIN |-| read feat: _msearch (p50 ms) | 2.53 | 1.84 | 0.73× | LOSE |-| read feat: _mget (p50 ms) | 2.18 | 1.13 | 0.52× | LOSE |-| mixed match_all (p99 ms, under write) | 98.16 | 4.55 | 0.05× | LOSE |-| mixed bool (p99 ms, under write) | 83.52 | 12.91 | 0.15× | LOSE |-| mixed range (p99 ms, under write) | 161.32 | 8.68 | 0.05× | LOSE |-| mixed terms (p99 ms, under write) | 75.80 | 4.54 | 0.06× | LOSE |-| mixed cardinality (p99 ms, under write) | 88.10 | 20.62 | 0.23× | LOSE |-| kNN k=10 (p50 ms) | 1.38 | 4.47 | 3.23× | WIN |+| ingest 100k × c1 (docs/s) | 108,203 | 72,740 | 1.49× | WIN |+| ingest 100k × c8 (docs/s) | 334,234 | 286,662 | 1.17× | WIN |+| ingest 1m × c1 (docs/s) | 109,683 | 70,452 | 1.56× | WIN |+| ingest 1m × c8 (docs/s) | 298,765 | 393,446 | 0.76× | LOSE |+| read q: match_all (p50 ms) | 1.91 | 2.95 | 1.54× | WIN |+| read q: match_none (p50 ms) | 1.12 | 2.22 | 1.98× | WIN |+| read q: match(model) (p50 ms) | 1.47 | 1.98 | 1.35× | WIN |+| read q: match_phrase(top_doc) (p50 ms) | 1.06 | 2.36 | 2.24× | WIN |+| read q: match_phrase_prefix (p50 ms) | 1.29 | unsupported (400) | — | N/A |+| read q: match_bool_prefix (p50 ms) | 2.27 | 5.06 | 2.23× | WIN |+| read q: multi_match (p50 ms) | 2.11 | 6.27 | 2.98× | WIN |+| read q: combined_fields (p50 ms) | 1.88 | unsupported (400) | — | N/A |+| read q: query_string (p50 ms) | 1.99 | 11.60 | 5.82× | WIN |+| read q: simple_query_string (p50 ms) | 2.34 | 2.76 | 1.18× | WIN |+| read q: more_like_this (p50 ms) | 1.99 | 2.51 | 1.26× | WIN |+| read q: term(status) (p50 ms) | 2.64 | 2.66 | 1.01× | WIN |+| read q: terms(model) (p50 ms) | 2.52 | 6.33 | 2.51× | WIN |+| read q: range(latency_ms) (p50 ms) | 2.50 | 6.52 | 2.61× | WIN |+| read q: range(@timestamp) (p50 ms) | 2.16 | 2.09 | 0.96× | LOSE |+| read q: range(cost_usd) (p50 ms) | 2.15 | 7.84 | 3.64× | WIN |+| read q: prefix(model) (p50 ms) | 2.49 | 10.15 | 4.07× | WIN |+| read q: wildcard(model) (p50 ms) | 2.15 | 10.33 | 4.80× | WIN |+| read q: regexp(model) (p50 ms) | 1.85 | 9.96 | 5.37× | WIN |+| read q: fuzzy(model) (p50 ms) | 2.95 | 2.39 | 0.81× | LOSE |+| read q: exists(cost_usd) (p50 ms) | 2.20 | 1.51 | 0.69× | LOSE |+| read q: ids (p50 ms) | 2.42 | 1.65 | 0.68× | LOSE |+| read q: term(cache_hit) (p50 ms) | 2.33 | 1.93 | 0.83× | LOSE |+| read q: bool must+filter+should+must_not (p50 ms) | 1.73 | 16.58 | 9.58× | WIN |+| read q: constant_score (p50 ms) | 2.50 | 2.05 | 0.82× | LOSE |+| read q: boosting (p50 ms) | 1.91 | 34.93 | 18.29× | WIN |+| read q: dis_max (p50 ms) | 2.22 | 5.92 | 2.67× | WIN |+| read q: function_score (p50 ms) | 2.29 | 39.11 | 17.09× | WIN |+| read q: pinned (p50 ms) | 1.76 | 15.56 | 8.83× | WIN |+| read agg: avg (p50 ms) | 1.70 | 2.41 | 1.42× | WIN |+| read agg: sum (p50 ms) | 2.65 | 1.65 | 0.62× | LOSE |+| read agg: min (p50 ms) | 1.62 | 1.44 | 0.89× | LOSE |+| read agg: max (p50 ms) | 2.38 | 2.12 | 0.89× | LOSE |+| read agg: stats (p50 ms) | 2.04 | 1.63 | 0.80× | LOSE |+| read agg: extended_stats (p50 ms) | 1.87 | 1.93 | 1.03× | WIN |+| read agg: value_count (p50 ms) | 1.92 | 1.87 | 0.97× | LOSE |+| read agg: cardinality (p50 ms) | 2.05 | 2.37 | 1.15× | WIN |+| read agg: percentiles (p50 ms) | 2.09 | 2.15 | 1.03× | WIN |+| read agg: percentile_ranks (p50 ms) | 2.61 | 2.36 | 0.90× | LOSE |+| read agg: median_absolute_deviation (p50 ms) | 2.27 | 1.40 | 0.62× | LOSE |+| read agg: matrix_stats (p50 ms) | 2.76 | 1.40 | 0.51× | LOSE |+| read agg: scripted_metric (p50 ms) | 2.14 | 1.73 | 0.81× | LOSE |+| read agg: top_hits (sub) (p50 ms) | 1.84 | 2.19 | 1.20× | WIN |+| read agg: terms (p50 ms) | 1.87 | 2.03 | 1.09× | WIN |+| read agg: rare_terms (p50 ms) | 1.85 | 1.40 | 0.75× | LOSE |+| read agg: significant_terms (p50 ms) | 1.99 | 1.51 | 0.76× | LOSE |+| read agg: histogram (p50 ms) | 2.22 | 1.87 | 0.84× | LOSE |+| read agg: date_histogram (p50 ms) | 2.42 | 2.04 | 0.84× | LOSE |+| read agg: auto_date_histogram (p50 ms) | 2.29 | 1.40 | 0.61× | LOSE |+| read agg: variable_width_histogram (p50 ms) | 2.55 | 1.62 | 0.63× | LOSE |+| read agg: range (p50 ms) | 1.97 | 2.03 | 1.03× | WIN |+| read agg: date_range (p50 ms) | 1.60 | 1.87 | 1.17× | WIN |+| read agg: filter (p50 ms) | 2.10 | 1.41 | 0.67× | LOSE |+| read agg: filters (p50 ms) | 2.08 | 1.99 | 0.96× | LOSE |+| read agg: missing (p50 ms) | 1.83 | 1.93 | 1.06× | WIN |+| read agg: global (p50 ms) | 2.55 | 1.74 | 0.68× | LOSE |+| read agg: adjacency_matrix (p50 ms) | 1.83 | 2.16 | 1.18× | WIN |+| read agg: composite (p50 ms) | 1.89 | 1.79 | 0.94× | LOSE |+| read agg: random_sampler (p50 ms) | 1.91 | 5.85 | 3.06× | WIN |+| read agg: terms+avg(cost) (p50 ms) | 2.42 | 1.83 | 0.76× | LOSE |+| read pipe: sum_bucket (p50 ms) | 2.74 | 1.76 | 0.64× | LOSE |+| read pipe: avg_bucket (p50 ms) | 2.46 | 1.89 | 0.77× | LOSE |+| read pipe: max_bucket (p50 ms) | 1.92 | 1.76 | 0.92× | LOSE |+| read pipe: stats_bucket (p50 ms) | 2.19 | 2.11 | 0.96× | LOSE |+| read pipe: percentiles_bucket (p50 ms) | 2.15 | 2.23 | 1.04× | WIN |+| read pipe: derivative (p50 ms) | 2.38 | 1.73 | 0.73× | LOSE |+| read pipe: cumulative_sum (p50 ms) | 1.99 | 2.01 | 1.01× | WIN |+| read pipe: moving_fn (p50 ms) | 2.46 | 2.17 | 0.88× | LOSE |+| read pipe: serial_diff (p50 ms) | 2.35 | 2.01 | 0.85× | LOSE |+| read pipe: bucket_script (p50 ms) | 2.13 | 1.55 | 0.73× | LOSE |+| read pipe: bucket_selector (p50 ms) | 2.27 | 2.13 | 0.94× | LOSE |+| read pipe: bucket_sort (p50 ms) | 2.13 | 1.76 | 0.83× | LOSE |+| read feat: sort-heavy (p50 ms) | 1.89 | 18.03 | 9.53× | WIN |+| read feat: deep from+size (from 500) (p50 ms) | 1.83 | 2.81 | 1.53× | WIN |+| read feat: search_after (p50 ms) | 3.89 | 18.91 | 4.86× | WIN |+| read feat: highlight (p50 ms) | 1.86 | 1.91 | 1.03× | WIN |+| read feat: _count (p50 ms) | 1.59 | 2.03 | 1.28× | WIN |+| read feat: _msearch (p50 ms) | 2.42 | 1.56 | 0.64× | LOSE |+| read feat: _mget (p50 ms) | 2.23 | 2.04 | 0.92× | LOSE |+| mixed match_all (p99 ms, under write) | 83.22 | 4.15 | 0.05× | LOSE |+| mixed bool (p99 ms, under write) | 64.08 | 12.01 | 0.19× | LOSE |+| mixed range (p99 ms, under write) | 72.37 | 8.07 | 0.11× | LOSE |+| mixed terms (p99 ms, under write) | 79.80 | 5.29 | 0.07× | LOSE |+| mixed cardinality (p99 ms, under write) | 99.06 | 17.29 | 0.17× | LOSE |+| kNN k=10 (p50 ms) | 1.13 | 4.25 | 3.76× | WIN || kNN recall@10 | 100.0% | 100.0% | 1.00× | WIN |-| index on-disk size | 634.4 MB | 806.8 MB | 1.27× | WIN |+| index on-disk size | 634.6 MB | 816.2 MB | 1.29× | WIN |## Skipped families (need a purpose-built index the flat corpus lacks)@@ -110,4 +110,4 @@ Verdict is from XERJ's POV: WIN = XERJ better (lower latency / higher docs·s /- `knn (as query/top-level)` — covered separately by --knn on a purpose-built dense_vector index- `percolate` — parses but no-ops to match_none — not benchmarkable for correctness-_Summary: 53 WIN, 36 LOSE, 2 N/A._+_Summary: 45 WIN, 44 LOSE, 2 N/A._diff --git a/demo/playbooks/SCORECARD.md b/demo/playbooks/SCORECARD.mdindex 412acc11a..c3dcb75bd 100644--- a/demo/playbooks/SCORECARD.md+++ b/demo/playbooks/SCORECARD.md@@ -6,97 +6,97 @@ Verdict is from XERJ's POV: WIN = XERJ better (lower latency / higher docs·s /| dimension | XERJ | ES | ratio | verdict ||---|--:|--:|--:|:--:|-| ingest 100k × c1 (docs/s) | 105,919 | 70,980 | 1.49× | WIN |-| ingest 100k × c8 (docs/s) | 306,345 | 275,136 | 1.11× | WIN |-| ingest 1m × c1 (docs/s) | 111,135 | 70,065 | 1.59× | WIN |-| ingest 1m × c8 (docs/s) | 291,802 | 382,198 | 0.76× | LOSE |-| read q: match_all (p50 ms) | 2.11 | 2.92 | 1.38× | WIN |-| read q: match_none (p50 ms) | 1.80 | 2.18 | 1.21× | WIN |-| read q: match(model) (p50 ms) | 1.93 | 2.49 | 1.29× | WIN |-| read q: match_phrase(top_doc) (p50 ms) | 1.16 | 2.56 | 2.20× | WIN |-| read q: match_phrase_prefix (p50 ms) | 1.17 | unsupported (400) | — | N/A |-| read q: match_bool_prefix (p50 ms) | 2.11 | 6.23 | 2.95× | WIN |-| read q: multi_match (p50 ms) | 1.69 | 6.90 | 4.09× | WIN |-| read q: combined_fields (p50 ms) | 1.27 | unsupported (400) | — | N/A |-| read q: query_string (p50 ms) | 1.33 | 10.71 | 8.07× | WIN |-| read q: simple_query_string (p50 ms) | 2.04 | 2.86 | 1.40× | WIN |-| read q: more_like_this (p50 ms) | 2.16 | 2.62 | 1.22× | WIN |-| read q: term(status) (p50 ms) | 2.18 | 2.79 | 1.28× | WIN |-| read q: terms(model) (p50 ms) | 1.99 | 5.54 | 2.78× | WIN |-| read q: range(latency_ms) (p50 ms) | 2.18 | 6.08 | 2.79× | WIN |-| read q: range(@timestamp) (p50 ms) | 1.97 | 2.33 | 1.19× | WIN |-| read q: range(cost_usd) (p50 ms) | 2.16 | 7.63 | 3.53× | WIN |-| read q: prefix(model) (p50 ms) | 2.60 | 10.80 | 4.15× | WIN |-| read q: wildcard(model) (p50 ms) | 2.41 | 11.07 | 4.59× | WIN |-| read q: regexp(model) (p50 ms) | 2.29 | 11.22 | 4.89× | WIN |-| read q: fuzzy(model) (p50 ms) | 2.52 | 2.74 | 1.09× | WIN |-| read q: exists(cost_usd) (p50 ms) | 2.04 | 1.59 | 0.78× | LOSE |-| read q: ids (p50 ms) | 2.45 | 1.18 | 0.48× | LOSE |-| read q: term(cache_hit) (p50 ms) | 1.82 | 2.08 | 1.14× | WIN |-| read q: bool must+filter+should+must_not (p50 ms) | 2.50 | 17.61 | 7.06× | WIN |-| read q: constant_score (p50 ms) | 1.85 | 2.15 | 1.16× | WIN |-| read q: boosting (p50 ms) | 2.37 | 35.73 | 15.10× | WIN |-| read q: dis_max (p50 ms) | 2.33 | 6.34 | 2.72× | WIN |-| read q: function_score (p50 ms) | 2.23 | 38.48 | 17.22× | WIN |-| read q: pinned (p50 ms) | 1.78 | 15.31 | 8.60× | WIN |-| read agg: avg (p50 ms) | 1.68 | 1.42 | 0.84× | LOSE |-| read agg: sum (p50 ms) | 2.61 | 1.61 | 0.62× | LOSE |-| read agg: min (p50 ms) | 1.71 | 2.45 | 1.43× | WIN |-| read agg: max (p50 ms) | 2.11 | 1.98 | 0.94× | LOSE |-| read agg: stats (p50 ms) | 1.97 | 1.87 | 0.94× | LOSE |-| read agg: extended_stats (p50 ms) | 2.06 | 1.58 | 0.77× | LOSE |-| read agg: value_count (p50 ms) | 2.26 | 1.92 | 0.85× | LOSE |-| read agg: cardinality (p50 ms) | 2.28 | 1.73 | 0.76× | LOSE |-| read agg: percentiles (p50 ms) | 2.46 | 1.56 | 0.64× | LOSE |-| read agg: percentile_ranks (p50 ms) | 2.64 | 1.84 | 0.70× | LOSE |-| read agg: median_absolute_deviation (p50 ms) | 2.55 | 1.82 | 0.71× | LOSE |-| read agg: matrix_stats (p50 ms) | 2.09 | 1.29 | 0.62× | LOSE |-| read agg: scripted_metric (p50 ms) | 2.45 | 1.42 | 0.58× | LOSE |-| read agg: top_hits (sub) (p50 ms) | 1.84 | 2.27 | 1.23× | WIN |-| read agg: terms (p50 ms) | 2.03 | 1.74 | 0.86× | LOSE |-| read agg: rare_terms (p50 ms) | 1.68 | 2.07 | 1.24× | WIN |-| read agg: significant_terms (p50 ms) | 1.56 | 1.24 | 0.80× | LOSE |-| read agg: histogram (p50 ms) | 1.22 | 1.49 | 1.22× | WIN |-| read agg: date_histogram (p50 ms) | 1.58 | 1.45 | 0.91× | LOSE |-| read agg: auto_date_histogram (p50 ms) | 1.69 | 1.21 | 0.71× | LOSE |-| read agg: variable_width_histogram (p50 ms) | 1.31 | 2.86 | 2.18× | WIN |-| read agg: range (p50 ms) | 2.36 | 2.49 | 1.05× | WIN |-| read agg: date_range (p50 ms) | 0.94 | 1.99 | 2.12× | WIN |-| read agg: filter (p50 ms) | 1.90 | 1.99 | 1.05× | WIN |-| read agg: filters (p50 ms) | 2.37 | 2.03 | 0.86× | LOSE |-| read agg: missing (p50 ms) | 2.68 | 1.84 | 0.69× | LOSE |-| read agg: global (p50 ms) | 2.38 | 1.93 | 0.81× | LOSE |-| read agg: adjacency_matrix (p50 ms) | 2.49 | 2.05 | 0.82× | LOSE |-| read agg: composite (p50 ms) | 1.73 | 2.23 | 1.29× | WIN |-| read agg: random_sampler (p50 ms) | 2.35 | 5.29 | 2.25× | WIN |-| read agg: terms+avg(cost) (p50 ms) | 2.08 | 2.27 | 1.09× | WIN |-| read pipe: sum_bucket (p50 ms) | 1.51 | 1.94 | 1.28× | WIN |-| read pipe: avg_bucket (p50 ms) | 1.62 | 1.88 | 1.16× | WIN |-| read pipe: max_bucket (p50 ms) | 2.19 | 2.35 | 1.08× | WIN |-| read pipe: stats_bucket (p50 ms) | 1.93 | 2.17 | 1.13× | WIN |-| read pipe: percentiles_bucket (p50 ms) | 1.86 | 2.44 | 1.31× | WIN |-| read pipe: derivative (p50 ms) | 2.24 | 3.16 | 1.41× | WIN |-| read pipe: cumulative_sum (p50 ms) | 2.69 | 2.18 | 0.81× | LOSE |-| read pipe: moving_fn (p50 ms) | 2.59 | 1.72 | 0.66× | LOSE |-| read pipe: serial_diff (p50 ms) | 1.79 | 1.98 | 1.11× | WIN |-| read pipe: bucket_script (p50 ms) | 2.13 | 1.63 | 0.77× | LOSE |-| read pipe: bucket_selector (p50 ms) | 2.15 | 1.61 | 0.75× | LOSE |-| read pipe: bucket_sort (p50 ms) | 2.44 | 1.62 | 0.66× | LOSE |-| read feat: sort-heavy (p50 ms) | 2.50 | 17.65 | 7.06× | WIN |-| read feat: deep from+size (from 500) (p50 ms) | 2.08 | 1.80 | 0.87× | LOSE |-| read feat: search_after (p50 ms) | 4.20 | 19.28 | 4.59× | WIN |-| read feat: highlight (p50 ms) | 2.18 | 2.87 | 1.32× | WIN |-| read feat: _count (p50 ms) | 2.04 | 2.18 | 1.07× | WIN |-| read feat: _msearch (p50 ms) | 2.53 | 1.84 | 0.73× | LOSE |-| read feat: _mget (p50 ms) | 2.18 | 1.13 | 0.52× | LOSE |-| mixed match_all (p99 ms, under write) | 98.16 | 4.55 | 0.05× | LOSE |-| mixed bool (p99 ms, under write) | 83.52 | 12.91 | 0.15× | LOSE |-| mixed range (p99 ms, under write) | 161.32 | 8.68 | 0.05× | LOSE |-| mixed terms (p99 ms, under write) | 75.80 | 4.54 | 0.06× | LOSE |-| mixed cardinality (p99 ms, under write) | 88.10 | 20.62 | 0.23× | LOSE |-| kNN k=10 (p50 ms) | 1.38 | 4.47 | 3.23× | WIN |+| ingest 100k × c1 (docs/s) | 108,203 | 72,740 | 1.49× | WIN |+| ingest 100k × c8 (docs/s) | 334,234 | 286,662 | 1.17× | WIN |+| ingest 1m × c1 (docs/s) | 109,683 | 70,452 | 1.56× | WIN |+| ingest 1m × c8 (docs/s) | 298,765 | 393,446 | 0.76× | LOSE |+| read q: match_all (p50 ms) | 1.91 | 2.95 | 1.54× | WIN |+| read q: match_none (p50 ms) | 1.12 | 2.22 | 1.98× | WIN |+| read q: match(model) (p50 ms) | 1.47 | 1.98 | 1.35× | WIN |+| read q: match_phrase(top_doc) (p50 ms) | 1.06 | 2.36 | 2.24× | WIN |+| read q: match_phrase_prefix (p50 ms) | 1.29 | unsupported (400) | — | N/A |+| read q: match_bool_prefix (p50 ms) | 2.27 | 5.06 | 2.23× | WIN |+| read q: multi_match (p50 ms) | 2.11 | 6.27 | 2.98× | WIN |+| read q: combined_fields (p50 ms) | 1.88 | unsupported (400) | — | N/A |+| read q: query_string (p50 ms) | 1.99 | 11.60 | 5.82× | WIN |+| read q: simple_query_string (p50 ms) | 2.34 | 2.76 | 1.18× | WIN |+| read q: more_like_this (p50 ms) | 1.99 | 2.51 | 1.26× | WIN |+| read q: term(status) (p50 ms) | 2.64 | 2.66 | 1.01× | WIN |+| read q: terms(model) (p50 ms) | 2.52 | 6.33 | 2.51× | WIN |+| read q: range(latency_ms) (p50 ms) | 2.50 | 6.52 | 2.61× | WIN |+| read q: range(@timestamp) (p50 ms) | 2.16 | 2.09 | 0.96× | LOSE |+| read q: range(cost_usd) (p50 ms) | 2.15 | 7.84 | 3.64× | WIN |+| read q: prefix(model) (p50 ms) | 2.49 | 10.15 | 4.07× | WIN |+| read q: wildcard(model) (p50 ms) | 2.15 | 10.33 | 4.80× | WIN |+| read q: regexp(model) (p50 ms) | 1.85 | 9.96 | 5.37× | WIN |+| read q: fuzzy(model) (p50 ms) | 2.95 | 2.39 | 0.81× | LOSE |+| read q: exists(cost_usd) (p50 ms) | 2.20 | 1.51 | 0.69× | LOSE |+| read q: ids (p50 ms) | 2.42 | 1.65 | 0.68× | LOSE |+| read q: term(cache_hit) (p50 ms) | 2.33 | 1.93 | 0.83× | LOSE |+| read q: bool must+filter+should+must_not (p50 ms) | 1.73 | 16.58 | 9.58× | WIN |+| read q: constant_score (p50 ms) | 2.50 | 2.05 | 0.82× | LOSE |+| read q: boosting (p50 ms) | 1.91 | 34.93 | 18.29× | WIN |+| read q: dis_max (p50 ms) | 2.22 | 5.92 | 2.67× | WIN |+| read q: function_score (p50 ms) | 2.29 | 39.11 | 17.09× | WIN |+| read q: pinned (p50 ms) | 1.76 | 15.56 | 8.83× | WIN |+| read agg: avg (p50 ms) | 1.70 | 2.41 | 1.42× | WIN |+| read agg: sum (p50 ms) | 2.65 | 1.65 | 0.62× | LOSE |+| read agg: min (p50 ms) | 1.62 | 1.44 | 0.89× | LOSE |+| read agg: max (p50 ms) | 2.38 | 2.12 | 0.89× | LOSE |+| read agg: stats (p50 ms) | 2.04 | 1.63 | 0.80× | LOSE |+| read agg: extended_stats (p50 ms) | 1.87 | 1.93 | 1.03× | WIN |+| read agg: value_count (p50 ms) | 1.92 | 1.87 | 0.97× | LOSE |+| read agg: cardinality (p50 ms) | 2.05 | 2.37 | 1.15× | WIN |+| read agg: percentiles (p50 ms) | 2.09 | 2.15 | 1.03× | WIN |+| read agg: percentile_ranks (p50 ms) | 2.61 | 2.36 | 0.90× | LOSE |+| read agg: median_absolute_deviation (p50 ms) | 2.27 | 1.40 | 0.62× | LOSE |+| read agg: matrix_stats (p50 ms) | 2.76 | 1.40 | 0.51× | LOSE |+| read agg: scripted_metric (p50 ms) | 2.14 | 1.73 | 0.81× | LOSE |+| read agg: top_hits (sub) (p50 ms) | 1.84 | 2.19 | 1.20× | WIN |+| read agg: terms (p50 ms) | 1.87 | 2.03 | 1.09× | WIN |+| read agg: rare_terms (p50 ms) | 1.85 | 1.40 | 0.75× | LOSE |+| read agg: significant_terms (p50 ms) | 1.99 | 1.51 | 0.76× | LOSE |+| read agg: histogram (p50 ms) | 2.22 | 1.87 | 0.84× | LOSE |+| read agg: date_histogram (p50 ms) | 2.42 | 2.04 | 0.84× | LOSE |+| read agg: auto_date_histogram (p50 ms) | 2.29 | 1.40 | 0.61× | LOSE |+| read agg: variable_width_histogram (p50 ms) | 2.55 | 1.62 | 0.63× | LOSE |+| read agg: range (p50 ms) | 1.97 | 2.03 | 1.03× | WIN |+| read agg: date_range (p50 ms) | 1.60 | 1.87 | 1.17× | WIN |+| read agg: filter (p50 ms) | 2.10 | 1.41 | 0.67× | LOSE |+| read agg: filters (p50 ms) | 2.08 | 1.99 | 0.96× | LOSE |+| read agg: missing (p50 ms) | 1.83 | 1.93 | 1.06× | WIN |+| read agg: global (p50 ms) | 2.55 | 1.74 | 0.68× | LOSE |+| read agg: adjacency_matrix (p50 ms) | 1.83 | 2.16 | 1.18× | WIN |+| read agg: composite (p50 ms) | 1.89 | 1.79 | 0.94× | LOSE |+| read agg: random_sampler (p50 ms) | 1.91 | 5.85 | 3.06× | WIN |+| read agg: terms+avg(cost) (p50 ms) | 2.42 | 1.83 | 0.76× | LOSE |+| read pipe: sum_bucket (p50 ms) | 2.74 | 1.76 | 0.64× | LOSE |+| read pipe: avg_bucket (p50 ms) | 2.46 | 1.89 | 0.77× | LOSE |+| read pipe: max_bucket (p50 ms) | 1.92 | 1.76 | 0.92× | LOSE |+| read pipe: stats_bucket (p50 ms) | 2.19 | 2.11 | 0.96× | LOSE |+| read pipe: percentiles_bucket (p50 ms) | 2.15 | 2.23 | 1.04× | WIN |+| read pipe: derivative (p50 ms) | 2.38 | 1.73 | 0.73× | LOSE |+| read pipe: cumulative_sum (p50 ms) | 1.99 | 2.01 | 1.01× | WIN |+| read pipe: moving_fn (p50 ms) | 2.46 | 2.17 | 0.88× | LOSE |+| read pipe: serial_diff (p50 ms) | 2.35 | 2.01 | 0.85× | LOSE |+| read pipe: bucket_script (p50 ms) | 2.13 | 1.55 | 0.73× | LOSE |+| read pipe: bucket_selector (p50 ms) | 2.27 | 2.13 | 0.94× | LOSE |+| read pipe: bucket_sort (p50 ms) | 2.13 | 1.76 | 0.83× | LOSE |+| read feat: sort-heavy (p50 ms) | 1.89 | 18.03 | 9.53× | WIN |+| read feat: deep from+size (from 500) (p50 ms) | 1.83 | 2.81 | 1.53× | WIN |+| read feat: search_after (p50 ms) | 3.89 | 18.91 | 4.86× | WIN |+| read feat: highlight (p50 ms) | 1.86 | 1.91 | 1.03× | WIN |+| read feat: _count (p50 ms) | 1.59 | 2.03 | 1.28× | WIN |+| read feat: _msearch (p50 ms) | 2.42 | 1.56 | 0.64× | LOSE |+| read feat: _mget (p50 ms) | 2.23 | 2.04 | 0.92× | LOSE |+| mixed match_all (p99 ms, under write) | 83.22 | 4.15 | 0.05× | LOSE |+| mixed bool (p99 ms, under write) | 64.08 | 12.01 | 0.19× | LOSE |+| mixed range (p99 ms, under write) | 72.37 | 8.07 | 0.11× | LOSE |+| mixed terms (p99 ms, under write) | 79.80 | 5.29 | 0.07× | LOSE |+| mixed cardinality (p99 ms, under write) | 99.06 | 17.29 | 0.17× | LOSE |+| kNN k=10 (p50 ms) | 1.13 | 4.25 | 3.76× | WIN || kNN recall@10 | 100.0% | 100.0% | 1.00× | WIN |-| index on-disk size | 634.4 MB | 806.8 MB | 1.27× | WIN |+| index on-disk size | 634.6 MB | 816.2 MB | 1.29× | WIN |## Skipped families (need a purpose-built index the flat corpus lacks)@@ -110,4 +110,4 @@ Verdict is from XERJ's POV: WIN = XERJ better (lower latency / higher docs·s /- `knn (as query/top-level)` — covered separately by --knn on a purpose-built dense_vector index- `percolate` — parses but no-ops to match_none — not benchmarkable for correctness-_Summary: 53 WIN, 36 LOSE, 2 N/A._+_Summary: 45 WIN, 44 LOSE, 2 N/A._