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: scorecard run 3 (post search_after fix) — 48W/37L/6NA — DIAGNOSTIC
xerj-org6b7282be16ad
diff --git a/demo/playbooks/SCORECARD.md b/demo/playbooks/SCORECARD.mdindex 3889b9315..ccdc96230 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) | 64,451 | 48,979 | 1.32× | WIN |-| ingest 100k × c8 (docs/s) | 233,602 | 199,589 | 1.17× | WIN |-| ingest 1m × c1 (docs/s) | 65,225 | 69,328 | 0.94× | LOSE |-| ingest 1m × c8 (docs/s) | 205,273 | 374,490 | 0.55× | LOSE |-| read q: match_all (p50 ms) | 2.14 | 5.07 | 2.37× | WIN |-| read q: match_none (p50 ms) | 2.71 | 3.10 | 1.14× | WIN |-| read q: match(model) (p50 ms) | 1.93 | 4.58 | 2.37× | WIN |-| read q: match_phrase(top_doc) (p50 ms) | 1.87 | 4.39 | 2.35× | WIN |-| read q: match_phrase_prefix (p50 ms) | 2.80 | unsupported (400) | — | N/A |-| read q: match_bool_prefix (p50 ms) | 1.93 | 6.34 | 3.29× | WIN |-| read q: multi_match (p50 ms) [result mismatch: hits 0 vs 277449] | 2.04 | 7.32 | mismatch | N/A |-| read q: combined_fields (p50 ms) | 2.05 | unsupported (400) | — | N/A |-| read q: query_string (p50 ms) [result mismatch: hits 0 vs 273706] | 2.30 | 11.89 | mismatch | N/A |-| read q: simple_query_string (p50 ms) [result mismatch: hits 0 vs 987529] | 2.11 | 3.15 | mismatch | N/A |-| read q: more_like_this (p50 ms) | 1.78 | 3.17 | 1.78× | WIN |-| read q: term(status) (p50 ms) | 2.51 | 2.65 | 1.06× | WIN |-| read q: terms(model) (p50 ms) | 2.23 | 6.42 | 2.88× | WIN |-| read q: range(latency_ms) (p50 ms) | 1.95 | 6.12 | 3.14× | WIN |-| read q: range(@timestamp) (p50 ms) | 2.24 | 2.63 | 1.17× | WIN |-| read q: range(cost_usd) (p50 ms) | 1.14 | 6.54 | 5.75× | WIN |-| read q: prefix(model) (p50 ms) | 1.71 | 11.00 | 6.43× | WIN |-| read q: wildcard(model) (p50 ms) | 1.40 | 10.42 | 7.43× | WIN |-| read q: regexp(model) (p50 ms) | collapsed | 11.38 | — | LOSE |-| read q: fuzzy(model) (p50 ms) [result mismatch: hits 0 vs 277449] | 1.62 | 3.11 | mismatch | N/A |-| read q: exists(cost_usd) (p50 ms) | 1.41 | 2.50 | 1.77× | WIN |-| read q: ids (p50 ms) | 1.10 | 2.24 | 2.03× | WIN |-| read q: term(cache_hit) (p50 ms) | 1.25 | 2.37 | 1.90× | WIN |-| read q: bool must+filter+should+must_not (p50 ms) | 0.82 | 20.54 | 25.10× | WIN |-| read q: constant_score (p50 ms) | 1.36 | 2.27 | 1.67× | WIN |-| read q: boosting (p50 ms) | 0.41 | 35.77 | 87.78× | WIN |-| read q: dis_max (p50 ms) | 1.49 | 6.27 | 4.20× | WIN |-| read q: function_score (p50 ms) | 1.15 | 35.40 | 30.84× | WIN |-| read q: pinned (p50 ms) | 0.98 | 15.67 | 15.99× | WIN |-| read agg: avg (p50 ms) | 0.89 | 2.66 | 2.97× | WIN |-| read agg: sum (p50 ms) | 1.21 | 2.57 | 2.12× | WIN |-| read agg: min (p50 ms) | 1.11 | 2.24 | 2.01× | WIN |-| read agg: max (p50 ms) | 0.62 | 2.75 | 4.47× | WIN |-| read agg: stats (p50 ms) | 1.01 | 2.53 | 2.51× | WIN |-| read agg: extended_stats (p50 ms) | 0.75 | 2.49 | 3.33× | WIN |-| read agg: value_count (p50 ms) | 1.21 | 2.18 | 1.81× | WIN |-| read agg: cardinality (p50 ms) | 0.98 | 2.52 | 2.58× | WIN |-| read agg: percentiles (p50 ms) | 1.16 | 2.05 | 1.77× | WIN |-| read agg: percentile_ranks (p50 ms) | 1.19 | 2.33 | 1.97× | WIN |-| read agg: median_absolute_deviation (p50 ms) | 2.46 | 2.79 | 1.13× | WIN |-| read agg: matrix_stats (p50 ms) | 2.09 | 2.61 | 1.25× | WIN |-| read agg: scripted_metric (p50 ms) | 2.05 | 2.13 | 1.04× | WIN |-| read agg: top_hits (sub) (p50 ms) | collapsed | 2.96 | — | LOSE |-| read agg: terms (p50 ms) | collapsed | 2.12 | — | LOSE |-| read agg: rare_terms (p50 ms) | 2.01 | 1.96 | 0.97× | LOSE |-| read agg: significant_terms (p50 ms) | 1.55 | 2.03 | 1.31× | WIN |-| read agg: histogram (p50 ms) | 2.66 | 2.71 | 1.02× | WIN |-| read agg: date_histogram (p50 ms) | 2.31 | 2.53 | 1.10× | WIN |-| read agg: auto_date_histogram (p50 ms) | 2.16 | 2.10 | 0.97× | LOSE |-| read agg: variable_width_histogram (p50 ms) | collapsed | 3.56 | — | LOSE |-| read agg: range (p50 ms) | collapsed | 2.30 | — | LOSE |-| read agg: date_range (p50 ms) | collapsed | 3.18 | — | LOSE |-| read agg: filter (p50 ms) | collapsed | 2.59 | — | LOSE |-| read agg: filters (p50 ms) | collapsed | 3.25 | — | LOSE |-| read agg: missing (p50 ms) | 1.21 | 2.89 | 2.39× | WIN |-| read agg: global (p50 ms) [result mismatch: hits 257 vs 12471] | 1.20 | 2.77 | mismatch | N/A |-| read agg: adjacency_matrix (p50 ms) | collapsed | 3.10 | — | LOSE |-| read agg: composite (p50 ms) | collapsed | 2.94 | — | LOSE |-| read agg: random_sampler (p50 ms) | collapsed | 6.48 | — | LOSE |-| read agg: terms+avg(cost) (p50 ms) | collapsed | 2.62 | — | LOSE |-| read pipe: sum_bucket (p50 ms) | collapsed | 3.08 | — | LOSE |-| read pipe: avg_bucket (p50 ms) | collapsed | 3.00 | — | LOSE |-| read pipe: max_bucket (p50 ms) | collapsed | 3.34 | — | LOSE |-| read pipe: stats_bucket (p50 ms) | collapsed | 2.68 | — | LOSE |-| read pipe: percentiles_bucket (p50 ms) | collapsed | 2.87 | — | LOSE |-| read pipe: derivative (p50 ms) | collapsed | 3.56 | — | LOSE |-| read pipe: cumulative_sum (p50 ms) | collapsed | 2.53 | — | LOSE |-| read pipe: moving_fn (p50 ms) | collapsed | 3.84 | — | LOSE |-| read pipe: serial_diff (p50 ms) | collapsed | 3.25 | — | LOSE |-| read pipe: bucket_script (p50 ms) | collapsed | 3.26 | — | LOSE |-| read pipe: bucket_selector (p50 ms) | collapsed | 3.35 | — | LOSE |-| read pipe: bucket_sort (p50 ms) | collapsed | 3.59 | — | LOSE |-| read feat: sort-heavy (p50 ms) | 2.19 | 20.04 | 9.13× | WIN |-| read feat: deep from+size (from 500) (p50 ms) | 2.36 | 3.43 | 1.45× | WIN |-| read feat: search_after (p50 ms) | 9924.66 | 18.76 | 0.00× | LOSE |-| read feat: highlight (p50 ms) | collapsed | 2.94 | — | LOSE |-| read feat: _count (p50 ms) | collapsed | 1.74 | — | LOSE |-| read feat: _msearch (p50 ms) | collapsed | 2.34 | — | LOSE |-| read feat: _mget (p50 ms) | collapsed | 2.27 | — | LOSE |-| mixed match_all (p99 ms, under write) | unsupported | 20.22 | — | N/A |-| mixed bool (p99 ms, under write) | unsupported | 17.36 | — | N/A |-| mixed range (p99 ms, under write) | unsupported | 37.85 | — | N/A |-| mixed terms (p99 ms, under write) | unsupported | 6.15 | — | N/A |-| mixed cardinality (p99 ms, under write) | unsupported | 21.44 | — | N/A |-| kNN k=10 (p50 ms) | error | 4.84 | — | N/A |-| kNN recall@10 | error | 100.0% | — | N/A |-| index on-disk size | unsupported | 819.4 MB | — | N/A |+| ingest 100k × c1 (docs/s) | 20,597 | 34,593 | 0.60× | LOSE |+| ingest 100k × c8 (docs/s) | 82,320 | 211,905 | 0.39× | LOSE |+| ingest 1m × c1 (docs/s) | 59,537 | 60,433 | 0.99× | LOSE |+| ingest 1m × c8 (docs/s) | 144,187 | 240,270 | 0.60× | LOSE |+| read q: match_all (p50 ms) | 2.25 | 3.32 | 1.47× | WIN |+| read q: match_none (p50 ms) | 1.46 | 1.81 | 1.24× | WIN |+| read q: match(model) (p50 ms) | 1.59 | 2.64 | 1.66× | WIN |+| read q: match_phrase(top_doc) (p50 ms) | 0.63 | 2.23 | 3.55× | WIN |+| read q: match_phrase_prefix (p50 ms) | 1.57 | unsupported (400) | — | N/A |+| read q: match_bool_prefix (p50 ms) | 0.80 | 6.27 | 7.83× | WIN |+| read q: multi_match (p50 ms) [result mismatch: hits 0 vs 277449] | 1.27 | 6.62 | mismatch | N/A |+| read q: combined_fields (p50 ms) | 0.35 | unsupported (400) | — | N/A |+| read q: query_string (p50 ms) [result mismatch: hits 0 vs 273706] | 0.82 | 11.40 | mismatch | N/A |+| read q: simple_query_string (p50 ms) [result mismatch: hits 0 vs 987529] | 0.96 | 2.57 | mismatch | N/A |+| read q: more_like_this (p50 ms) | 0.61 | 2.64 | 4.34× | WIN |+| read q: term(status) (p50 ms) | 0.88 | 2.16 | 2.46× | WIN |+| read q: terms(model) (p50 ms) | 0.94 | 5.35 | 5.71× | WIN |+| read q: range(latency_ms) (p50 ms) | 0.64 | 7.09 | 11.02× | WIN |+| read q: range(@timestamp) (p50 ms) | 0.81 | 1.94 | 2.40× | WIN |+| read q: range(cost_usd) (p50 ms) | 0.74 | 7.50 | 10.11× | WIN |+| read q: prefix(model) (p50 ms) | 0.95 | 11.20 | 11.83× | WIN |+| read q: wildcard(model) (p50 ms) | 0.80 | 10.80 | 13.53× | WIN |+| read q: regexp(model) (p50 ms) | collapsed | 11.91 | — | LOSE |+| read q: fuzzy(model) (p50 ms) | 1.27 | 3.28 | 2.59× | WIN |+| read q: exists(cost_usd) (p50 ms) | 1.21 | 2.25 | 1.85× | WIN |+| read q: ids (p50 ms) | 1.16 | 2.11 | 1.82× | WIN |+| read q: term(cache_hit) (p50 ms) | 1.26 | 1.92 | 1.52× | WIN |+| read q: bool must+filter+should+must_not (p50 ms) | 1.82 | 19.54 | 10.76× | WIN |+| read q: constant_score (p50 ms) | 1.82 | 2.50 | 1.37× | WIN |+| read q: boosting (p50 ms) | 1.37 | 35.18 | 25.73× | WIN |+| read q: dis_max (p50 ms) | 1.16 | 7.30 | 6.30× | WIN |+| read q: function_score (p50 ms) | 1.19 | 39.28 | 33.06× | WIN |+| read q: pinned (p50 ms) | 1.33 | 14.96 | 11.25× | WIN |+| read agg: avg (p50 ms) | 0.92 | 2.80 | 3.03× | WIN |+| read agg: sum (p50 ms) | 1.18 | 2.00 | 1.69× | WIN |+| read agg: min (p50 ms) | 1.13 | 1.77 | 1.57× | WIN |+| read agg: max (p50 ms) | 0.82 | 2.04 | 2.50× | WIN |+| read agg: stats (p50 ms) | 0.96 | 1.71 | 1.78× | WIN |+| read agg: extended_stats (p50 ms) | 1.00 | 2.50 | 2.49× | WIN |+| read agg: value_count (p50 ms) | 1.16 | 2.36 | 2.04× | WIN |+| read agg: cardinality (p50 ms) | 0.74 | 1.85 | 2.50× | WIN |+| read agg: percentiles (p50 ms) | 0.70 | 2.79 | 3.99× | WIN |+| read agg: percentile_ranks (p50 ms) | 0.99 | 2.30 | 2.33× | WIN |+| read agg: median_absolute_deviation (p50 ms) | 2.09 | 2.07 | 0.99× | LOSE |+| read agg: matrix_stats (p50 ms) | 1.40 | 2.11 | 1.50× | WIN |+| read agg: scripted_metric (p50 ms) | 0.46 | 1.47 | 3.18× | WIN |+| read agg: top_hits (sub) (p50 ms) | collapsed | 2.16 | — | LOSE |+| read agg: terms (p50 ms) | collapsed | 2.18 | — | LOSE |+| read agg: rare_terms (p50 ms) | 1.41 | 1.78 | 1.26× | WIN |+| read agg: significant_terms (p50 ms) | 0.81 | 2.15 | 2.67× | WIN |+| read agg: histogram (p50 ms) | collapsed | 2.23 | — | LOSE |+| read agg: date_histogram (p50 ms) | 1.23 | 1.95 | 1.59× | WIN |+| read agg: auto_date_histogram (p50 ms) | 0.96 | 1.86 | 1.94× | WIN |+| read agg: variable_width_histogram (p50 ms) | collapsed | 3.57 | — | LOSE |+| read agg: range (p50 ms) | collapsed | 3.19 | — | LOSE |+| read agg: date_range (p50 ms) | collapsed | 2.75 | — | LOSE |+| read agg: filter (p50 ms) | collapsed | 2.49 | — | LOSE |+| read agg: filters (p50 ms) | collapsed | 2.60 | — | LOSE |+| read agg: missing (p50 ms) | 1.35 | 2.65 | 1.96× | WIN |+| read agg: global (p50 ms) [result mismatch: hits 257 vs 12471] | 1.70 | 2.93 | mismatch | N/A |+| read agg: adjacency_matrix (p50 ms) | collapsed | 3.15 | — | LOSE |+| read agg: composite (p50 ms) | collapsed | 3.27 | — | LOSE |+| read agg: random_sampler (p50 ms) | collapsed | 6.83 | — | LOSE |+| read agg: terms+avg(cost) (p50 ms) | collapsed | 2.95 | — | LOSE |+| read pipe: sum_bucket (p50 ms) | collapsed | 3.21 | — | LOSE |+| read pipe: avg_bucket (p50 ms) | collapsed | 3.37 | — | LOSE |+| read pipe: max_bucket (p50 ms) | collapsed | 2.60 | — | LOSE |+| read pipe: stats_bucket (p50 ms) | collapsed | 3.04 | — | LOSE |+| read pipe: percentiles_bucket (p50 ms) | collapsed | 3.29 | — | LOSE |+| read pipe: derivative (p50 ms) | collapsed | 3.15 | — | LOSE |+| read pipe: cumulative_sum (p50 ms) | collapsed | 2.69 | — | LOSE |+| read pipe: moving_fn (p50 ms) | collapsed | 3.01 | — | LOSE |+| read pipe: serial_diff (p50 ms) | collapsed | 3.14 | — | LOSE |+| read pipe: bucket_script (p50 ms) | collapsed | 3.38 | — | LOSE |+| read pipe: bucket_selector (p50 ms) | collapsed | 2.93 | — | LOSE |+| read pipe: bucket_sort (p50 ms) | collapsed | 2.82 | — | LOSE |+| read feat: sort-heavy (p50 ms) | 2.95 | 20.24 | 6.85× | WIN |+| read feat: deep from+size (from 500) (p50 ms) | 3.03 | 3.92 | 1.29× | WIN |+| read feat: search_after (p50 ms) | 73.47 | 18.59 | 0.25× | LOSE |+| read feat: highlight (p50 ms) | 1.55 | 4.10 | 2.65× | WIN |+| read feat: _count (p50 ms) | 1.16 | 3.17 | 2.73× | WIN |+| read feat: _msearch (p50 ms) | 1.25 | 3.38 | 2.69× | WIN |+| read feat: _mget (p50 ms) | 0.71 | 2.25 | 3.15× | WIN |+| mixed match_all (p99 ms, under write) | 15490.81 | 103.38 | 0.01× | LOSE |+| mixed bool (p99 ms, under write) | collapsed | 147.63 | — | LOSE |+| mixed range (p99 ms, under write) | collapsed | 132.79 | — | LOSE |+| mixed terms (p99 ms, under write) | collapsed | 158.86 | — | LOSE |+| mixed cardinality (p99 ms, under write) | collapsed | 90.57 | — | LOSE |+| kNN k=10 (p50 ms) | 4.67 | 22.43 | 4.80× | WIN |+| kNN recall@10 | 100.0% | 80.0% | 1.25× | WIN |+| index on-disk size | 689.8 MB | 143.2 MB | 0.21× | LOSE |## 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: 43 WIN, 33 LOSE, 15 N/A._+_Summary: 48 WIN, 37 LOSE, 6 N/A._