Merge pull request 'chore: bump version 0.5 → 0.6' (#138 ) from chore/bump-v0.6.0 into main

Reviewed-on: #138
chore: bump version 0.5 → 0.6
2026-05-10 15:01:45 +00:00 · 2026-05-10 23:56:51 +09:00 · 2026-05-10 14:53:21 +00:00 · 2026-05-10 23:47:47 +09:00 · 2026-05-10 23:45:55 +09:00 · 2026-05-10 23:28:48 +09:00
13 changed files with 214 additions and 69 deletions
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -3525,7 +3525,7 @@ dependencies = [

 [[package]]
 name = "kebab-app"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "base64 0.22.1",
@@ -3569,7 +3569,7 @@ dependencies = [

 [[package]]
 name = "kebab-chunk"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "blake3",
@@ -3584,7 +3584,7 @@ dependencies = [

 [[package]]
 name = "kebab-cli"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "clap",
@@ -3605,7 +3605,7 @@ dependencies = [

 [[package]]
 name = "kebab-config"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "dirs 5.0.1",
@@ -3620,7 +3620,7 @@ dependencies = [

 [[package]]
 name = "kebab-core"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "blake3",
@@ -3634,7 +3634,7 @@ dependencies = [

 [[package]]
 name = "kebab-embed"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "blake3",
@@ -3648,7 +3648,7 @@ dependencies = [

 [[package]]
 name = "kebab-embed-local"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "fastembed",
@@ -3661,7 +3661,7 @@ dependencies = [

 [[package]]
 name = "kebab-eval"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "kebab-app",
@@ -3680,7 +3680,7 @@ dependencies = [

 [[package]]
 name = "kebab-llm"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "kebab-core",
@@ -3689,7 +3689,7 @@ dependencies = [

 [[package]]
 name = "kebab-llm-local"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "kebab-config",
@@ -3706,7 +3706,7 @@ dependencies = [

 [[package]]
 name = "kebab-mcp"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "kebab-app",
@@ -3724,7 +3724,7 @@ dependencies = [

 [[package]]
 name = "kebab-normalize"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "kebab-core",
@@ -3739,7 +3739,7 @@ dependencies = [

 [[package]]
 name = "kebab-parse-image"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "ab_glyph",
 "anyhow",
@@ -3763,7 +3763,7 @@ dependencies = [

 [[package]]
 name = "kebab-parse-md"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "kebab-core",
@@ -3780,7 +3780,7 @@ dependencies = [

 [[package]]
 name = "kebab-parse-pdf"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "blake3",
@@ -3793,7 +3793,7 @@ dependencies = [

 [[package]]
 name = "kebab-parse-types"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "kebab-core",
 "serde",
@@ -3801,7 +3801,7 @@ dependencies = [

 [[package]]
 name = "kebab-rag"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "blake3",
@@ -3822,7 +3822,7 @@ dependencies = [

 [[package]]
 name = "kebab-search"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "globset",
@@ -3841,7 +3841,7 @@ dependencies = [

 [[package]]
 name = "kebab-source-fs"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "blake3",
@@ -3858,7 +3858,7 @@ dependencies = [

 [[package]]
 name = "kebab-store-sqlite"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "blake3",
@@ -3879,7 +3879,7 @@ dependencies = [

 [[package]]
 name = "kebab-store-vector"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "arrow",
@@ -3903,7 +3903,7 @@ dependencies = [

 [[package]]
 name = "kebab-tui"
-version = "0.5.0"
+version = "0.6.0"
 dependencies = [
 "anyhow",
 "crossterm",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -30,7 +30,7 @@ edition       = "2024"
 rust-version  = "1.85"
 license       = "MIT OR Apache-2.0"
 repository    = "https://github.com/altair823/kebab"
-version       = "0.5.0"
+version       = "0.6.0"

 [workspace.dependencies]
 anyhow       = "1"
--- a/README.md
+++ b/README.md
@@ -7,7 +7,7 @@
 - **Rust toolchain** ≥ 1.85 (workspace 가 edition 2024 + resolver 3 사용). [rustup](https://rustup.rs) 권장.
 - **Ollama** — `kebab ask` 와 이미지 OCR/caption 가 사용. `https://ollama.com/download` 에서 설치 후 `ollama serve` 실행. 기본 LLM 은 gemma4 계열 (`ollama pull gemma4:e4b`) — OCR / caption 도 같은 family 라 모델 하나만 pull 하면 됨. 더 큰 variant 원하면 `gemma4:26b` 등으로 config override. config 의 `[models.llm].endpoint` 에 host:port 명시.
 - **빌드 디스크** — 첫 빌드 시 `target/` 가 6–10 GB (Lance + DataFusion + fastembed). 여유 확인.
- **fastembed 모델** — 첫 `kebab ingest` 시 `multilingual-e5-small` (~470 MB) 자동 다운로드.
+- **fastembed 모델** — 첫 `kebab ingest` 시 `multilingual-e5-large` (~1.3 GB, fb-39b) 자동 다운로드. `config.toml` 에서 `model = "multilingual-e5-small"` 로 명시하면 이전 모델 사용.

 ## 설치

@@ -133,7 +133,7 @@ flowchart TB
    subgraph Pipeline["도메인 + 파이프라인"]
        parse["parse-md / parse-pdf / parse-image"]
        chunker["chunker (md-heading-v1, pdf-page-v1)"]
-        embedder["embedder (fastembed multilingual-e5-small)"]
+        embedder["embedder (fastembed multilingual-e5-large)"]
        retriever["retriever (lexical / vector / hybrid RRF)"]
        rag["RAG pipeline"]
    end
@@ -178,7 +178,12 @@ flowchart TB

 ## Configuration

- `~/.config/kebab/config.toml` — `kebab init` 가 XDG 경로에 생성. `[workspace]` (root, exclude — include 필드는 제거됨, 지원 형식은 자동 결정), `[storage]`, `[chunking]`, `[models.embedding]`, `[models.llm]`, `[image.ocr]`, `[image.caption]`, `[search]`, `[rag]`, `[ui]` 절. `[ui] theme = "dark" | "light"` 로 TUI 팔레트 선택 (default `"dark"`, 알 수 없는 값은 dark fallback). `[search] stale_threshold_days = 30` (p9-fb-32) — search hit / RAG citation 의 `stale` 플래그 기준 (default 30 일, `0` 으로 비활성화). 옛 config 의 `workspace.include = [...]` 은 silently 무시 + 단발 deprecation warning (p9-fb-25).
+- `~/.config/kebab/config.toml` — `kebab init` 가 XDG 경로에 생성. `[workspace]` (root, exclude — include 필드는 제거됨, 지원 형식은 자동 결정), `[storage]`, `[chunking]`, `[models.embedding]`, `[models.llm]`, `[image.ocr]`, `[image.caption]`, `[search]`, `[rag]`, `[ui]` 절. 
+  - `[models.embedding]` — 
+    - `model` (default `"multilingual-e5-large"`, fb-39b) — 다국어 sentence embedding 모델. 1024-dim. ONNX (~1.3 GB) 첫 실행 시 fastembed cache (`config.storage.model_dir/fastembed/`) 에 자동 다운로드. `"multilingual-e5-small"` (384 dim) 는 backwards-compat 으로 사용 가능 — TOML 에 명시.
+    - `dimensions` (default `1024`) — 모델의 embedding 차원. config 와 LanceDB stored dim 불일치 시 검색 결과 0 건 (orphan table). 모델 변경 시 `kebab reset --vector-only && kebab ingest` 로 vector index 재구축 권장.
+  - `[ui] theme = "dark" | "light"` 로 TUI 팔레트 선택 (default `"dark"`, 알 수 없는 값은 dark fallback). 
+  - `[search] stale_threshold_days = 30` (p9-fb-32) — search hit / RAG citation 의 `stale` 플래그 기준 (default 30 일, `0` 으로 비활성화). 옛 config 의 `workspace.include = [...]` 은 silently 무시 + 단발 deprecation warning (p9-fb-25).
 - `[rag] prompt_template_version` (default `"rag-v2"`) — RAG system prompt version. `"rag-v1"` 은 legacy backwards-compat (사용자 명시 시 유지). v2 강화 규칙: (1) fact 인용 시 [#번호] 앞에 chunk 속 원문 큰따옴표 표기, (2) 학습 지식 동원 금지, (3) 근거 모호 시 "확실하지 않다" 명시.
 - `--config <path>` flag — 임시 워크스페이스 / 격리 테스트 시 사용. CLI / TUI 모두 honor.
 - `KEBAB_*` env — 일부 키 override (`KEBAB_RAG_SCORE_GATE`, `KEBAB_EVAL_GOLDEN`, `KEBAB_COMMIT_HASH` 등).
--- a/crates/kebab-config/src/lib.rs
+++ b/crates/kebab-config/src/lib.rs
@@ -302,9 +302,9 @@ impl Config {
            models: ModelsCfg {
                embedding: EmbeddingModelCfg {
                    provider: "fastembed".to_string(),
-                    model: "multilingual-e5-small".to_string(),
+                    model: "multilingual-e5-large".to_string(),
                    version: "v1".to_string(),
-                    dimensions: 384,
+                    dimensions: 1024,
                    batch_size: 64,
                },
                llm: LlmCfg {
@@ -764,7 +764,8 @@ mod tests {
        let c = Config::defaults();
        assert_eq!(c.rag.score_gate, 0.30);
        assert_eq!(c.chunking.target_tokens, 500);
-        assert_eq!(c.models.embedding.dimensions, 384);
+        assert_eq!(c.models.embedding.model, "multilingual-e5-large");
+        assert_eq!(c.models.embedding.dimensions, 1024);
        assert_eq!(c.search.rrf_k, 60);
    }

@@ -947,9 +948,9 @@ chunker_version = "md-heading-v1"

 [models.embedding]
 provider = "fastembed"
-model = "multilingual-e5-small"
+model = "multilingual-e5-large"
 version = "v1"
-dimensions = 384
+dimensions = 1024
 batch_size = 64

 [models.llm]
--- a/crates/kebab-embed-local/Cargo.toml
+++ b/crates/kebab-embed-local/Cargo.toml
@@ -5,14 +5,14 @@ edition       = { workspace = true }
 rust-version  = { workspace = true }
 license       = { workspace = true }
 repository    = { workspace = true }
-description   = "Local fastembed-rs adapter implementing kb_core::Embedder (multilingual-e5-small default)"
+description   = "Local fastembed-rs adapter implementing kb_core::Embedder (multilingual-e5-large default, e5-small backwards-compat)"

 [dependencies]
 kebab-config = { path = "../kebab-config" }
 kebab-embed = { path = "../kebab-embed" }
 # Default features bring `ort-download-binaries` (bundled ONNX runtime)
 # and `hf-hub-native-tls` (first-run model download). No extra features
-# needed for the multilingual-e5-small path.
+# needed for the multilingual-e5-{small,large} paths.
 fastembed = { workspace = true }
 tracing   = { workspace = true }
 anyhow    = { workspace = true }
--- a/crates/kebab-embed-local/src/lib.rs
+++ b/crates/kebab-embed-local/src/lib.rs
@@ -1,8 +1,9 @@
 //! `kb-embed-local` — `FastembedEmbedder`, a local ONNX-backed
 //! [`Embedder`](kebab_embed::Embedder) implementation.
 //!
-//! Wraps [`fastembed::TextEmbedding`] for the default `multilingual-e5-small`
-//! (384-dim) model. Honors `config.models.embedding.batch_size` and applies
+//! Wraps [`fastembed::TextEmbedding`]. Default is `multilingual-e5-large`
+//! (1024-dim, p9-fb-39b); `multilingual-e5-small` (384-dim) is also supported
+//! for backwards-compat. Honors `config.models.embedding.batch_size` and applies
 //! the e5 prefix convention (§11.3 of the design report):
 //!
 //! * `EmbeddingKind::Document` → `"passage: "` prefix
@@ -69,9 +70,9 @@ impl FastembedEmbedder {
            .with_context(|| format!("create fastembed cache dir {}", cache_dir.display()))?;

        // 2. Resolve the fastembed enum variant from
-        //    `config.models.embedding.model`. Currently only the default
-        //    `multilingual-e5-small` is wired; other model names error
-        //    out with a clear message rather than silently misconfiguring.
+        //    `config.models.embedding.model`. Currently `multilingual-e5-large`
+        //    (default) and `multilingual-e5-small` are wired; other model names
+        //    error out with a clear message rather than silently misconfiguring.
        let model_name = resolve_model(&config.models.embedding.model)?;

        // 3. Verify dim match BEFORE loading the model — if the config
@@ -100,7 +101,7 @@ impl FastembedEmbedder {
            target: "kebab-embed-local",
            model = %config.models.embedding.model,
            cache_dir = %cache_dir.display(),
-            "loading embedding model (first run will download ~470MB)"
+            "loading embedding model (first run downloads model weights — ~470MB for e5-small, ~1.3GB for e5-large)"
        );
        let inner = TextEmbedding::try_new(opts)
            .context("fastembed: TextEmbedding::try_new")?;
@@ -193,17 +194,18 @@ fn prefix_input(input: &EmbeddingInput<'_>) -> String {
 }

 /// Resolve a `config.models.embedding.model` string to a fastembed
-/// `EmbeddingModel` enum variant. Only `multilingual-e5-small` is wired
-/// for p3-2; additional model names should be added (and their dims
-/// pinned in tests) as needed.
+/// `EmbeddingModel` enum variant. Currently supports `multilingual-e5-small`
+/// (384-dim) and `multilingual-e5-large` (1024-dim); additional model names
+/// should be added (and their dims pinned in tests) as needed.
 fn resolve_model(name: &str) -> Result<EmbeddingModel> {
    match name {
        "multilingual-e5-small" => Ok(EmbeddingModel::MultilingualE5Small),
+        "multilingual-e5-large" => Ok(EmbeddingModel::MultilingualE5Large),
        other => anyhow::bail!(
            "kb-embed-local: unsupported embedding model {other:?}; \
-             this adapter currently only ships `multilingual-e5-small`. \
-             Add a new arm to `resolve_model` (and a fastembed feature \
-             flag if needed) to support more."
+             this adapter currently ships `multilingual-e5-small` and \
+             `multilingual-e5-large`. Add a new arm to `resolve_model` \
+             (and a fastembed feature flag if needed) to support more."
        ),
    }
 }
@@ -294,6 +296,12 @@ mod tests {
        resolve_model("multilingual-e5-small").expect("default model resolves");
    }

+    #[test]
+    fn resolve_model_supports_e5_large() {
+        let m = resolve_model("multilingual-e5-large").expect("e5-large should resolve");
+        let _ = m;
+    }
+
    #[test]
    fn resolve_unknown_model_errors() {
        let err = resolve_model("not-a-real-model").expect_err("unknown model errors");
@@ -301,6 +309,21 @@ mod tests {
        assert!(msg.contains("unsupported embedding model"), "msg={msg}");
    }

+    // ── check_dim ────────────────────────────────────────────────────
+
+    #[test]
+    fn check_dim_passes_for_1024() {
+        check_dim(1024, 1024).expect("matching dims must pass");
+    }
+
+    #[test]
+    fn check_dim_rejects_384_vs_1024() {
+        let err = check_dim(384, 1024).expect_err("dim mismatch must error");
+        let msg = format!("{err}");
+        assert!(msg.contains("384") && msg.contains("1024"),
+            "error must mention both dims, got: {msg}");
+    }
+
    // expand_path tests live in `kb-config::paths`. The adapter imports
    // it and trusts the upstream coverage rather than duplicating it.
 }
--- a/crates/kebab-embed-local/tests/embed_model.rs
+++ b/crates/kebab-embed-local/tests/embed_model.rs
@@ -3,10 +3,11 @@
 //!
 //! ## Why every test in this file is `#[ignore]`
 //!
-//! The first call to `FastembedEmbedder::new` downloads ~470 MB of
-//! weights from Hugging Face into `data_dir/models/fastembed/`. Doing
-//! that on every `cargo test` invocation is wasteful, so the bare
-//! invocation skips this file entirely.
+//! The first call to `FastembedEmbedder::new` downloads ~1.3 GB of
+//! weights (multilingual-e5-large per p9-fb-39b default) from Hugging
+//! Face into `data_dir/models/fastembed/`. Doing that on every
+//! `cargo test` invocation is wasteful, so the bare invocation skips
+//! this file entirely.
 //!
 //! Run the full suite with:
 //! ```text
@@ -58,19 +59,20 @@ fn shared_embedder() -> &'static FastembedEmbedder {
 // ─── construction ─────────────────────────────────────────────────────

 #[test]
-#[ignore = "downloads ~470MB ONNX model on first run; CI-only"]
-fn default_config_constructs_with_dims_384() {
+#[ignore = "downloads ~1.3GB ONNX model on first run; CI-only"]
+fn default_config_constructs_with_dims_1024() {
+    // p9-fb-39b: default flipped to multilingual-e5-large (1024 dim).
    let emb = shared_embedder();
-    assert_eq!(emb.dimensions(), 384);
-    assert_eq!(emb.model_id().0, "multilingual-e5-small");
+    assert_eq!(emb.dimensions(), 1024);
+    assert_eq!(emb.model_id().0, "multilingual-e5-large");
    assert_eq!(emb.model_version().0, "v1");
 }

 #[test]
-#[ignore = "downloads ~470MB ONNX model on first run; CI-only"]
+#[ignore = "downloads ~1.3GB ONNX model on first run; CI-only"]
 fn mismatched_dims_in_config_errors_at_construction() {
    let (mut cfg, _tmp) = test_config();
-    cfg.models.embedding.dimensions = 512; // model is 384
+    cfg.models.embedding.dimensions = 512; // model is 1024 (e5-large default)
    // `FastembedEmbedder` deliberately does not implement `Debug`
    // (its inner ONNX session has no useful debug shape), so we
    // can't use `expect_err`; match the Result manually.
@@ -80,7 +82,7 @@ fn mismatched_dims_in_config_errors_at_construction() {
    };
    let msg = format!("{err}");
    assert!(msg.contains("dimension mismatch"), "msg={msg}");
-    assert!(msg.contains("384"), "msg={msg}");
+    assert!(msg.contains("1024"), "msg={msg}");
    assert!(msg.contains("512"), "msg={msg}");
 }

@@ -104,8 +106,8 @@ fn document_and_query_yield_different_vectors() {
        ])
        .expect("embed two inputs");
    assert_eq!(out.len(), 2);
-    assert_eq!(out[0].len(), 384);
-    assert_eq!(out[1].len(), 384);
+    assert_eq!(out[0].len(), 1024);
+    assert_eq!(out[1].len(), 1024);

    // Both vectors are L2-normalized → cosine similarity == dot product.
    let cos: f32 = out[0]
@@ -142,11 +144,11 @@ fn output_vectors_are_l2_normalized() {
    ];
    let out = emb.embed(&inputs).expect("embed");
    // Per `kebab_embed::assert_unit_norm` docs: `5e-4` is the safe bound at
-    // 384 dims (f32::EPSILON × √384 ≈ 2.3e-6, but ONNX kernels add
+    // 1024 dims (f32::EPSILON × √1024 ≈ 2.3e-6, but ONNX kernels add
    // their own per-component noise; 1e-3 is very generous and matches
    // the spec's `± 1e-3`).
    kebab_embed::assert_unit_norm(&out, 1e-3);
-    kebab_embed::assert_vector_shape(&out, 384);
+    kebab_embed::assert_vector_shape(&out, 1024);
 }

 // ─── determinism ──────────────────────────────────────────────────────
@@ -254,7 +256,7 @@ fn snapshot_aggregate_hash_is_stable() {
    // Round every component to 4 decimal places, hash deterministically.
    let mut hasher = DefaultHasher::new();
    for (i, v) in out.iter().enumerate() {
-        assert_eq!(v.len(), 384, "row {i} dim mismatch");
+        assert_eq!(v.len(), 1024, "row {i} dim mismatch");
        for x in v {
            let rounded: i32 = (*x * 1.0e4).round() as i32;
            rounded.hash(&mut hasher);
--- a/docs/SMOKE.md
+++ b/docs/SMOKE.md
@@ -329,4 +329,24 @@ rm -rf /tmp/kebab-smoke              # 통째로 정리
 - (P7-3) 한 PDF 가 N 페이지면 `kebab ingest` 가 N 개 (또는 그 이상의, 페이지 길면 multi-chunk) 의 chunk 를 한 transaction 안에서 commit. 500 페이지 책 → 500+ chunk 한 번에 → embedding throughput 가 bottleneck. 임베딩 활성 워크스페이스에서 큰 PDF 를 처음 ingest 하면 분-단위 시간 + WAL 크기 증가 가능 — P+ 스케일 hardening task 까지 정상 동작이지만 비용은 측정 가능.
 - (P7-3 + follow-up) 동일 path 에 byte 가 다른 PDF 를 두 번째 ingest 하면 `purge_vector_orphans_for_workspace_path` 가 옛 chunk_id 를 LanceDB 에서 먼저 삭제, 이어서 `purge_orphan_at_workspace_path` 가 옛 doc / chunks / embedding_records 를 SQLite 에서 sweep. 새 byte 가 새 `doc_id` 로 색인됨. `IngestReport` 에 그 자산만 `new+=1` (다른 자산은 `updated`). 두 store 모두 정합 — 옛 본문 검색 시 옛 chunks 가 더 이상 surface 되지 않음.

+### Embedding upgrade (fb-39b)
+
+`multilingual-e5-small` 에서 `multilingual-e5-large` 로 업그레이드 시퀀스:
+
+```bash
+# 기존 vector index 정리 (orphan table 회피)
+kebab --config /tmp/kebab-smoke/config.toml reset --vector-only
+
+# config.toml 의 [models.embedding] 갱신:
+#   model = "multilingual-e5-large"
+#   dimensions = 1024
+
+# 재-ingest — fastembed 가 첫 실행 시 e5-large ONNX (~1.3 GB) 자동 다운로드.
+# 다운로드 시간 + 모든 chunk re-embed 시간 (e5-small 대비 ~3-4×).
+kebab --config /tmp/kebab-smoke/config.toml ingest
+
+# fb-39 의 P@k metric 으로 small vs large 비교:
+kebab --config /tmp/kebab-smoke/config.toml eval run
+```
+
 자세한 history 와 발견된 버그는 [tasks/HOTFIXES.md](../tasks/HOTFIXES.md) 참조.
--- a/docs/superpowers/specs/2026-04-27-kebab-final-form-design.md
+++ b/docs/superpowers/specs/2026-04-27-kebab-final-form-design.md
@@ -93,7 +93,7 @@ retrieval trace

 grounded ✓  qwen2.5:14b-instruct  rag-v1  3 chunks
 prompt   1184 tokens  completion 312 tokens  latency 1842 ms
-embedding multilingual-e5-small  index v1.0
+embedding multilingual-e5-large  index v1.0
 ```

 ### 1.3 `kebab ask` (refusal — score gate)
@@ -212,7 +212,7 @@ variant 별 해당 키만 채움. `path` 와 `uri` 는 항상 채움 (`uri` 는
    "vector_rank": 2
  },
  "index_version": "v1.0",
-  "embedding_model": "multilingual-e5-small",
+  "embedding_model": "multilingual-e5-large",
  "chunker_version": "md-heading-v1"
 }
 ```
@@ -264,7 +264,7 @@ Per-query failure 는 `bulk_search_item.v1.error` (error.v1) 에 격리, 다른
  "grounded": true,
  "refusal_reason": null,
  "model": { "id": "qwen2.5:14b-instruct", "provider": "ollama" },
-  "embedding": { "id": "multilingual-e5-small", "provider": "fastembed", "dimensions": 384 },
+  "embedding": { "id": "multilingual-e5-large", "provider": "fastembed", "dimensions": 1024 },
  "prompt_template_version": "rag-v1",
  "retrieval": {
    "trace_id": "ret_4a8b2c1e",
@@ -374,7 +374,7 @@ Per-query failure 는 `bulk_search_item.v1.error` (error.v1) 에 격리, 다른
  "token_estimate": 480,
  "chunker_version": "md-heading-v1",
  "embeddings": [
-    { "model": "multilingual-e5-small", "dimensions": 384, "embedding_id": "e_2f1a" }
+    { "model": "multilingual-e5-large", "dimensions": 1024, "embedding_id": "e_2f1a" }
  ]
 }
 ```
@@ -390,7 +390,7 @@ Per-query failure 는 `bulk_search_item.v1.error` (error.v1) 에 격리, 다른
    { "name": "data_dir_writable", "ok": true, "detail": "~/.local/share/kebab" },
    { "name": "sqlite_open", "ok": true, "detail": "kebab.sqlite (schema v1)" },
    { "name": "lancedb_open", "ok": true, "detail": "lancedb/" },
-    { "name": "embedding_model", "ok": true, "detail": "multilingual-e5-small (384d)" },
+    { "name": "embedding_model", "ok": true, "detail": "multilingual-e5-large (1024d)" },
    { "name": "ollama_reachable", "ok": true, "detail": "http://127.0.0.1:11434" },
    { "name": "ollama_model_pulled", "ok": false, "detail": "qwen2.5:14b-instruct missing", "hint": "ollama pull qwen2.5:14b-instruct" }
  ]
@@ -1209,9 +1209,9 @@ chunker_version           = "md-heading-v1"

 [models.embedding]
 provider   = "fastembed"
-model      = "multilingual-e5-small"
+model      = "multilingual-e5-large"
 version    = "v1"
-dimensions = 384
+dimensions = 1024
 batch_size = 64

 [models.llm]
@@ -1474,7 +1474,7 @@ $ kebab doctor
 ✓ data_dir_writable     ~/.local/share/kebab
 ✓ sqlite_open           kebab.sqlite (schema v1)
 ✓ lancedb_open          lancedb/
-✓ embedding_model       multilingual-e5-small (384d)
+✓ embedding_model       multilingual-e5-large (1024d)
 ✓ ollama_reachable      http://127.0.0.1:11434
 ✗ ollama_model_pulled   qwen2.5:14b-instruct missing
                        hint: ollama pull qwen2.5:14b-instruct
--- a/tasks/HOTFIXES.md
+++ b/tasks/HOTFIXES.md
@@ -14,6 +14,22 @@ historical contract that was implemented; this file accumulates the
 deltas so phase 5+ readers can find the live behavior without diffing
 git history.

+## 2026-05-10 — p9-fb-39b: embedding upgrade UX
+
+**무엇이 바뀌었나**: default embedding 이 `multilingual-e5-small` (384 dim) 에서 `multilingual-e5-large` (1024 dim) 로 변경. LanceDB 테이블은 `(model, dim)` 으로 네임스페이스되어 새 모델은 fresh 테이블에 쓰고, 옛 `chunk_embeddings_multilingual-e5-small_384` 테이블은 orphan 상태 됨.
+
+**user TOML 에 small 명시한 경우**: backwards-compat 유지. 사용자가 `[models.embedding] model = "multilingual-e5-small"` 로 명시했으면 그대로 small 사용 (새 default 무시).
+
+**idempotent re-embed**: fb-23 incremental ingest 가 embedding_version mismatch 감지하면 자동으로 이전 chunk 를 새 모델로 re-embed. 다음 `kebab ingest` 호출 시 기존 chunk 의 embedding 을 새 테이블에 재작성.
+
+**disk 절약**: 이전 모델의 orphan 테이블을 먼저 정리하려면 `kebab reset --vector-only` 실행 (LanceDB + SQLite `embedding_records` 모두 wipe). 이후 `kebab ingest` 가 모든 chunk 를 새 모델로 re-embed 해 새 테이블 채움.
+
+**search/ask 결과**: re-embed 전까지는 empty hit (새 모델에 데이터가 없음). `kebab ingest` 후 정상 검색 가능.
+
+**Spec contract 와의 관계**: design §5 (storage) + §9 (versioning cascade) 의 embedding_model.id / dimensions 변경. wire 의 `embedding_version` 필드 (kebab-app schema.v1.models.embedding_version 가 config.models.embedding.model 값을 그대로 emit) 변경 — CLAUDE.md cascade rule 의 release 트리거. 본 PR 머지 후 `chore: bump version 0.5 → 0.6` + tag 필요.
+
+**Spec deviation**: design `2026-05-10-p9-fb-39b-embedding-upgrade-design.md` 의 §Migration policy + §Public surface delta 가 `LanceVectorStore::open` 안 신규 `error.v1.code = "embedding_dim_mismatch"` 명시했으나 구현 제외. 이유: LanceDB tables 가 `(model, dim)` namespaced — silent orphan + empty-hit 으로 surface (hard error 아님). 명시 error 필요 시 별도 startup health check 작업 필요 (fb-39c 후보 또는 doctor 확장).
+
 ## 2026-05-09 — p9-fb-34: search wire wrapped in search_response.v1

 **무엇이 바뀌었나**: `kebab search --json` stdout 이 기존 `search_hit.v1[]` 배열에서 신규 `search_response.v1` object 로 교체. wrapper 가 `hits`, `next_cursor`, `truncated` 세 필드를 가짐.
--- a/tasks/INDEX.md
+++ b/tasks/INDEX.md
@@ -130,6 +130,7 @@ P0~P5 는 직렬. P6~P9 는 P5 이후 병렬 가능.
    ### 🎯 0.5.0 — RAG quality (cascade 동반: V00X + reindex)
    - [p9-fb-38 score semantics](p9/p9-fb-38-score-semantics.md) — ✅ 머지 (2026-05-10)
    - [p9-fb-39 retrieval precision 튜닝](p9/p9-fb-39-retrieval-precision-tuning.md) — ✅ 머지 (2026-05-10) — eval foundation only, lever 적용 deferred
+    - [p9-fb-39b embedding upgrade](p9/p9-fb-39b-embedding-upgrade.md) — ✅ 머지 (2026-05-10) — multilingual-e5-large default
    - [p9-fb-40 fact-grounded answer](p9/p9-fb-40-fact-grounded-answer.md) — ✅ 머지 (2026-05-10)

    ### 🎯 0.6.0 또는 P+ — reasoning
--- a/tasks/p9/p9-fb-39-retrieval-precision-tuning.md
+++ b/tasks/p9/p9-fb-39-retrieval-precision-tuning.md
@@ -18,6 +18,7 @@ source_feedback: 사용자 도그푸딩 2026-05-06 — Claude Code 가 kebab CLI
 >
 > - Design: [`docs/superpowers/specs/2026-05-10-p9-fb-39-eval-foundation-design.md`](../../docs/superpowers/specs/2026-05-10-p9-fb-39-eval-foundation-design.md)
 > - Plan: [`docs/superpowers/plans/2026-05-10-p9-fb-39-eval-foundation.md`](../../docs/superpowers/plans/2026-05-10-p9-fb-39-eval-foundation.md)
+> - fb-39b (lever 적용 — embedding upgrade): [`tasks/p9/p9-fb-39b-embedding-upgrade.md`](./p9-fb-39b-embedding-upgrade.md) ✅

 ## 증상 / 동기

--- a/tasks/p9/p9-fb-39b-embedding-upgrade.md
+++ b/tasks/p9/p9-fb-39b-embedding-upgrade.md
@@ -0,0 +1,76 @@
+---
+phase: P9
+component: kebab-embed-local + kebab-config + kebab-store-vector + docs
+task_id: p9-fb-39b
+title: "Embedding model upgrade (multilingual-e5-large)"
+status: completed
+target_version: 0.6.0
+depends_on: [p9-fb-39]
+unblocks: []
+contract_source: ../../docs/superpowers/specs/2026-04-27-kebab-final-form-design.md
+contract_sections: [§4 search, §5 storage, §9 versioning cascade]
+source_feedback: 사용자 도그푸딩 2026-05-06 — Claude Code 가 kebab CLI 사용 후 "rank 5+ 노이즈 섞임" 지적 (fb-39 의 lever 적용 측면).
+---
+
+# p9-fb-39b — Embedding model upgrade
+
+> ✅ **구현 완료.** fb-39 의 lever 후보 4개 중 embedding model 업그레이드 lever 적용. P@k metric (fb-39) 으로 small vs large 비교 가능.
+>
+> - Design update: [`docs/superpowers/specs/2026-04-27-kebab-final-form-design.md`](../../docs/superpowers/specs/2026-04-27-kebab-final-form-design.md) §5 / §9
+> - Plan: [`docs/superpowers/plans/2026-05-10-p9-fb-39b-embedding-upgrade.md`](../../docs/superpowers/plans/2026-05-10-p9-fb-39b-embedding-upgrade.md)
+
+## 요약
+
+- `multilingual-e5-small` (384 dim) → `multilingual-e5-large` (1024 dim) default flip.
+- 기존 user TOML 이 small 명시 시 그대로 유지 (backwards-compat).
+- fb-23 incremental ingest 가 embedding_version mismatch 감지 → 자동 re-embed.
+- 0.5 → 0.6 minor bump 트리거 (design §9 cascade rule, current Cargo.toml = 0.5.0).
+
+## 구현 항목
+
+1. **config defaults flip** — `[models.embedding] model = "multilingual-e5-large"`, `dimensions = 1024`.
+2. **fastembed e5-large resolution** — `kebab-embed-local` 의 `resolve_model()` 에 e5-large arm 추가.
+3. **fixture sweep** — 모든 unit/integration 테스트의 default embedding 모델 확인. Config 에서 명시하지 않으면 새 default 따름 (`provider = "none"` 테스트 제외).
+4. **design contract update** — design §5 (storage example) + §9 (versioning table) 의 embedding_model.id + dimensions 갱신.
+5. **HOTFIXES entry** — 사용자 재 ingest 절차 + backwards-compat 동작 명시.
+6. **README update** — `[models.embedding]` 섹션의 기본값 + `dimensions` 필드 설명 갱신.
+7. **SMOKE.md append** — 스모크 테스트 중 embedding 업그레이드 검증 절차 (reset → config 갱신 → ingest → eval).
+8. **tasks/INDEX.md append** — p9-fb-39b row 추가 (p9-fb-39 sibling).
+
+## Allowed dependencies
+
+- `kebab-embed-local` — fastembed crate + `kebab-core`
+- `kebab-config` — toml crate
+- `kebab-store-vector` — lancedb crate (table naming 로직만 영향)
+- `kebab-app` — 와이어링만 (API 변경 없음)
+
+## Forbidden dependencies
+
+- parse-* crate (parser 무관)
+- llm-* crate (embedding 과 무관)
+- search crate (검색 로직은 adapter pattern 으로 이미 generic)
+
+## Test
+
+- `cargo test -p kebab-embed-local -- e5_large` (새 arm 테스트)
+- `cargo test -p kebab-config -- embedding_defaults` (config defaults)
+- `cargo test --workspace --no-fail-fast -j 1` (full regression)
+- Smoke: `kebab --config /tmp/smoke.toml doctor | grep embedding` → `multilingual-e5-large (1024d)`
+- Smoke: `kebab --config /tmp/smoke.toml ingest` → embedding 진행 표시 + dimension check
+- P@k eval: `kebab eval run` (fb-39 의 golden set) small vs large 비교
+
+## Backward compat notes
+
+- Pre-fb-39b user 가 config 에서 명시하지 않은 embedding → new default (large) 자동 적용. TOML 에 `model = "multilingual-e5-small"` 명시하면 유지.
+- `kebab-config` 의 `EmbeddingCfg.model` 은 String 필드 — TOML 에 명시한 값이 default 를 override (serde 기본 동작).
+- Orphan LanceDB table (`chunk_embeddings_multilingual-e5-small_384`) 은 다음 `kebab ingest` 실행 후 stale 취급 — 사용자가 수동 `kebab reset --vector-only` 로 정리 가능.
+
+## Binary version bump
+
+- 0.5.0 → 0.6.0 (current Cargo.toml = 0.5.0; embedding_version cascade triggers minor bump per design §9).
+- Release notes: `embedding default: multilingual-e5-small (384d) → multilingual-e5-large (1024d), P@k metric ↑`.
+
+## Post-merge deviation
+
+- **`embedding_dim_mismatch` ErrorV1 dropped**: design spec §Migration policy 가 `LanceVectorStore::open` 안 dim mismatch 감지 + 신규 `error.v1.code = "embedding_dim_mismatch"` 를 명시했으나 구현에서 제외. 이유: LanceDB tables 가 `(model, dim)` namespaced (`crates/kebab-store-vector/src/paths.rs:21`) — 신규 model 변경 시 새 table 자동 생성, 옛 table orphan. dim mismatch 가 hard error 되지 않고 검색 결과 0건 (silent precision loss) 으로 surface. HOTFIXES 항목이 documentation source. 명시 error 가 의미 있으려면 별도 startup health check 필요 — fb-39c 후보 또는 v0.7.0 의 doctor 확장.
+- 영향: 사용자가 model 변경 후 `kebab ingest` 안 하면 검색 결과 0건. README + SMOKE walkthrough 가 reset --vector-only && ingest 시퀀스 안내.
Author	SHA1	Message	Date
altair823	72798bd3ff	Merge pull request 'chore: bump version 0.5 → 0.6' (#138 ) from chore/bump-v0.6.0 into main Reviewed-on: #138	2026-05-10 15:01:45 +00:00
th-kim0823	c3177561b9	chore: bump version 0.5 → 0.6 v0.6.0 batches RAG quality batch: - fb-38 score semantics (search_hit.v1 score_kind) - fb-40 fact-grounded answer (rag-v2 prompt template) - fb-42 bulk multi-query (kebab search --bulk + mcp__kebab__bulk_search) - fb-39 eval foundation (precision_at_k_chunk metric) - fb-39b embedding upgrade (multilingual-e5-large default) embedding_version cascade triggers minor bump per design §9. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-10 23:56:51 +09:00
altair823	a465b71f99	Merge pull request 'feat(fb-39b): embedding upgrade — multilingual-e5-large default' (#137 ) from feat/fb-39b-embedding-upgrade into main Reviewed-on: #137	2026-05-10 14:53:21 +00:00
th-kim0823	787007172a	fix(fb-39b): address PR #137 round 2 review - target_version 0.7.0 → 0.6.0 (current Cargo.toml = 0.5.0; embedding_version cascade bumps to 0.6, not 0.7) - 요약 bullet "0.6 → 0.7" → "0.5 → 0.6" 정정 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-10 23:47:47 +09:00
th-kim0823	b954e9ce66	fix(fb-39b): address PR #137 round 1 review - CI-only embed_model.rs tests updated 384 → 1024 + e5-small → e5-large references (incl. file header download size, snapshot dim assert, L2 norm comment) - kebab-embed-local module docs + Cargo.toml description list both models (small + large) - Stale tracing message expanded with both model sizes - Task spec Post-merge deviation section: record dropped embedding_dim_mismatch ErrorV1 + reason (LanceDB (model, dim) namespacing makes hard-error redundant) - Task spec + HOTFIXES version bump 0.6→0.7 corrected to 0.5→0.6 (current Cargo.toml = 0.5.0; fb-42 0.6 cut deferred per user direction) - HOTFIXES "embedding_version bump 아님" line corrected — cascade rule DOES trigger release bump, plus deviation note for the dropped error Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-10 23:45:55 +09:00
th-kim0823	c62a8ff503	docs(fb-39b): design + HOTFIXES + new task spec + INDEX + README + SMOKE Tasks 4 + 5: comprehensive doc update for embedding upgrade (multilingual-e5-large). - design §5 + §9: update embedding_model / dimensions references (384 -> 1024) - HOTFIXES: add fb-39b entry with user re-ingest procedure + backwards-compat notes - tasks/p9-fb-39b-embedding-upgrade.md: new task spec (completed status) - INDEX.md: add fb-39b row under RAG quality phase - fb-39 task banner: append fb-39b link as lever implementation - README: update config defaults + fastembed model size + embedding field docs - SMOKE.md: append embedding upgrade verification section with e5-small -> e5-large sequence Wire schema: no change (additive at config level, new table created by existing code). Binary version: 0.6.0 -> 0.7.0 (cascade rule: embedding_model change = minor bump). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-10 23:28:48 +09:00
th-kim0823	69c94b6692	feat(embed,config): add multilingual-e5-large + flip default config (fb-39b) Task 1: Add multilingual-e5-large arm to kebab-embed-local::resolve_model with tests for 1024-dim variants and error cases. Task 2: Flip kebab-config defaults from e5-small (384-dim) to e5-large (1024-dim) across defaults(), test assertions, and TOML template. All tests pass; clippy clean. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-10 23:05:36 +09:00