Skip to content

armgit / xerj

Public Imported from GitHub
0

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.

docs(field-report): semantic verification diagnostics on macOS (#2017)

Jarvisa8b0deede431
diff --git a/user-feedback/16-agent-field-reports/2026-10-21-semantic-verify-diagnostics.md b/user-feedback/16-agent-field-reports/2026-10-21-semantic-verify-diagnostics.md
new file mode 100644
index 000000000..f68c9bb60
--- /dev/null
+++ b/user-feedback/16-agent-field-reports/2026-10-21-semantic-verify-diagnostics.md
@@ -0,0 +1,13 @@
+# Validating semantic-verify diagnostics on macOS (2026-10-21)
+
+**Agent:** OpenAI Codex · **XERJ:** xerj v1.0.0-rc.97 (source build) · **Platform:** macOS arm64
+
+**Pointed at:** A Rust-source fixture in the stateful HTTP reconciliation harness, then the repository's synthetic ES-YAML indices on an isolated temporary XERJ node.
+
+**Used it for:** Investigating #1971's missing diagnostic context; forcing a semantic read-back mismatch, checking refusal and resume, and running the protocol conformance gate.
+
+**Verdict:** The HTTP regression could exercise post-publication refusal and successful resume without changing published documents. Reporting live semantic-record counts, both sealed expectations, and dataset slugs makes that failure more useful. Whole-crate testing was noisy on this Mac: the same 20 tests failed on unmodified upstream and the patch. I did not reproduce the original Windows/.NET discrepancy or evaluate retrieval quality.
+
+**Numbers:** `cargo test -p xerj-autoindex --lib incremental_reconcile_http_tests` → 74 passed; ES-YAML runner → 1436 passed, 0 failed, 10 skipped (75 catch-lenient cases under the existing suite policy).
+
+**Filed alongside:** https://github.com/xerj-org/xerj/pull/2016