Feature / source
MemtraceGitNexusChromaDBCGC Exact-symbol lookupmempalace, 1K queriesAcc@1 / precision@10fair/results_iso_1k_mempalace_*
Memtrace
96.6% / 0.967best precision; -0.4 pp Acc@1GitNexus97.0% / 0.702highest file-specific Acc@1
ChromaDB62.4% / 0.188chunk vectors trail exact symbols
CGC7.9% / 0.521shared-harness CGC row
Token economymempalace, same 1K queriesAcc@1 / 1k tokbenchmarks/README Bench #1
Memtrace
495.52correct hits per 1k response tokensGitNexus126.903.90x behind Memtrace
ChromaDB32.11large chunk payloads
CGC28.97low Acc@1 hurts token score
Exact-symbol lookupDjango, 1K queriesAcc@1 / precision@10fair/results_iso_1k_django_*
Memtrace
79.7% / 0.807best precision; -1.4 pp Acc@1GitNexus81.1% / 0.660highest file-specific Acc@1
ChromaDB34.2% / 0.128semantic chunks trail on symbols
CGCDNFDjango index did not finish
Query latencyDjango, 50K nodesAverage lookup latencyfair/results_iso_1k_django_*
Memtrace
0.19 mssub-millisecond on DjangoGitNexus20.47 ms108x slower
ChromaDB54.56 ms287x slower
CGCDNFDjango index did not finish
HEAD index speedDjango, 3.3K filesHEAD-only index timeBENCHMARKS-v0.3.22
Memtrace
13.6 squeryable HEAD graphGitNexus48.4 s3.6x slower
ChromaDB268.7 s19x slower
CGCDNF24+ min without progress
Working setDjango isolated processPeak RSSfair/results_iso_1k_django_*
Memtrace
26.2 MBsmallest resident setGitNexus30.9 MB1.18x larger
ChromaDB1133 MB43x larger
CGCDNFnot isolated on Django
Graph callersmempalace pyright GTFiltered callers recallbenchmarks/README Bench #3
Memtrace
0.851highest callers recallGitNexus0.013flow text rarely maps callers
ChromaDBN/Sgraph queries not supported
CGC0.584runner-up graph recall
Graph callersDjango pyright GTFiltered callers recallbenchmarks/README Bench #3
Memtrace
0.816generalizes to larger corpusGitNexus0.05315.4x behind Memtrace
ChromaDBN/Sgraph queries not supported
CGC0.000no Django caller recall
Incremental freshnessscratch fixture, 50 editsp95 time to queryablebenchmarks/README Bench #4
Memtrace
42.5 mssub-50 ms freshnessGitNexusN/Sbatch-only eval-server
ChromaDB5001 mstimeout p95
CGC613.7 ms14.4x slower
Semantic intentDjango PR-title NL queriesRecall@10benchmarks/README Bench #2
Memtrace
58.6%2nd on semantic workloadGitNexusN/Ano comparable recall row
ChromaDB66.8%semantic vector win
CGCN/Snot supported for NL retrieval
N/S means a benchmark method is not supported by that system. DNF means the published run did not finish under the documented benchmark conditions.