[ajianaz]

What

Second step of #1515 (option B, staged). uteke_core::RecallSettings owns the nine [recall] keys, their defaults, conversions to the engine types (graph_rerank(), recall_config()), and the UTEKE_RECALL_* / UTEKE_GRAPH_* overrides with their range checks, moved from the CLI and tested with an injected lookup. The CLI config uses it as its [recall] type (alias) and one call replaces ~80 lines of env handling; the server drops RecallFileSection and its Option chains and keeps its startup sanitizing of default_strategy.

Why

The same section was written twice (CLI struct with defaults, server Option struct with the same defaults copied) and the env overrides existed only in the CLI. Part 3 adds one apply step for the CLI, the server and MCP (closes #1355) and is where it is decided whether the server honours the UTEKE_RECALL_* variables; this part keeps it as it is. Refs #1515.

Testing

• clippy -D warnings, fmt, cargo test --workspace pass; cargo run -p docgen changes nothing
• 6 new core tests (defaults, equal to the built-in request defaults, partial section, conversions, valid overrides, invalid/out-of-range/NaN overrides)
• side by side, with the real ONNX model: the CLI and the server built from develop and from this branch, same [recall] section and environment, give identical recall results for default, min_score that filters, default_strategy = vector, invalid strategy (falls back to hybrid in the server), jaccard/graph settings, and env overrides (valid, invalid, beating the file)
• reviewed locally with the Cora MCP agent reviewer before pushing: no blocking finding; one minor finding of mine (two near-identical helpers) was fixed before this push

[github-actions]

🔍 Cora AI Code Review

✅ No issues found. Code looks good!

---

Review powered by cora-code · BYOK · MIT