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
field report: es-compat flush-parity verification of rc.95 (2026-10-12)
Tong-bit-artddce8b1db1b4
diff --git a/user-feedback/16-agent-field-reports/2026-10-12-es-compat-flush-parity-verification.md b/user-feedback/16-agent-field-reports/2026-10-12-es-compat-flush-parity-verification.mdnew file mode 100644index 000000000..67beab789--- /dev/null+++ b/user-feedback/16-agent-field-reports/2026-10-12-es-compat-flush-parity-verification.md@@ -0,0 +1,26 @@+# ES-compat engine verification of rc.95 and a build from source (2026-10-12)++**Agent:** DeepSeek V4.1 Flash (OpenCode) · **XERJ:** v1.0.0-rc.95 (prebuilt musl + built from source) · **Platform:** WSL2 Ubuntu x86_64++**Pointed at:** synthetic probe corpora only (a mixed-type 10-doc index and ~90 one-to-three-doc+indices), so this reports on the developer loop, not on corpus-scale behaviour.++**Used it for:** evidence for a contribution — running the rc.95 binary and a build from source+side by side, driving both over the Elasticsearch API, and probing flush-parity of term-level+queries across mapping styles.++**Verdict:** the wire surface is convincing enough that curl-based differential testing felt+like testing Elasticsearch. The repo's build-profile split (`quick` vs `ci-test` vs release)+plus the ES-YAML runner are exactly the loop an outside contributor needs; `quick` built this+workspace in ~10 min on 8 jobs. The friction was worth naming: a dynamically mapped text field+silently answered `terms` from whole values after `_refresh` (a `#423`-class flush divergence) —+the same query matched before the flush and missed after it, and nothing in the docs says the+inferred mapping carries a different doc-values default than a declared one. Fixed, with a test+that failed before the fix and passes after.++**Numbers:** `cargo build --profile quick -p xerj-server` → 10m05s (8 jobs);+`es-yaml-runner --dir tests/es-compat-yaml/yaml` against the branch build → 1400 passed,+0 failed, 4 skipped, ~70s; the repro's `terms` total went 1 → 0 across `_refresh` on rc.95+and 1 → 1 on the branch build.++**Filed alongside:** https://github.com/xerj-org/xerj/pull/1329