T3: Wire pathHints/languageHints into retrieval layer with metadata filtering

doc/configuration.md

@@ -117,7 +117,7 @@ Controls how queries are matched against the index.
 {
   "retrieval": {
     "topK": 10,
-    "minScore": 0.5,
+    "minScore": 0.35,
     "hybridSearch": {
       "enabled": true,
       "keywordWeight": 0.4
@@ -136,9 +136,9 @@ Controls how queries are matched against the index.
 | Option | Default | Description |
 |---|---|---|
 | `topK` | `10` | Default number of chunks fetched per query |
-| `minScore` | `0.5` | Minimum relevance score (0–1) |
+| `minScore` | `0.35` | Minimum relevance score (0–1) |
 | `hybridSearch.enabled` | `true` | Enable combined TF×IDF + vector search |
-| `hybridSearch.keywordWeight` | `0.4` | Weight for keyword score in fusion: `(1-kw)*vScore + kw*kScore` |
+| `hybridSearch.keywordWeight` | `0.4` | Keyword weight in RRF fusion: `vContrib = (1-kw)×(K+1)/(K+vRank+1)` |
 | `contextOptimization.enabled` | `true` | Enable post-retrieval optimization pipeline |
 | `contextOptimization.maxPerFile` | `3` | Max chunks per file in final result (0 = unlimited) |
 | `contextOptimization.mergeAdjacent` | `true` | Merge consecutive same-file chunks separated by ≤ gap |

doc/development.md

@@ -3,10 +3,10 @@
 ## Project Setup
 
 ```bash
-git clone https://github.com/your-org/OpenCodeRAG.git
+git clone https://github.com/MrDoe/OpenCodeRAG.git
 cd OpenCodeRAG
 npm install --legacy-peer-deps
-opencode-rag setup --force
+opencode-rag init --force
 ```
 
 > LanceDB and other dependencies have peer dependency conflicts — `--legacy-peer-deps` is required. Run `opencode-rag setup --force` after building to sync the development copy into the runtime.

doc/eval-ranking-report.md

@@ -0,0 +1,68 @@
+# Ranking Order Comparison
+
+**main (734fc60)** vs **t1-cosine-l2 (95fd596)**
+**Generated:** 2026-07-06T11:12:41.259Z
+
+## Config
+
+| Setting | `main` | `t1-cosine-l2` |
+|---|---|---|
+| Embedding | qwen3-embedding:0.6b | qwen3-embedding:0.6b |
+| topK | 20 | 20 |
+| minScore | 0.5 | 0.35 |
+| Hybrid | true | true |
+| keywordWeight | 0.4 | 0.4 |
+| Index chunks | 542 | 542 |
+
+## Ranking Agreement
+
+| Metric | Value |
+|---|---|
+| Top-1 match | 13/35 |
+| Top-3 identical | 1/35 |
+| Top-5 identical | 0/35 |
+| Full top-K identical | 0/35 |
+| Avg overlap in top-5 | 2.9 |
+| Avg overlap in top-K | 2.9 |
+| Kendall's τ (avg) | 0.3619 |
+| Queries with keyword contribution | 35/35 |
+
+## Per-Query Detail
+
+| # | Query | Top-1 same | Top-5 same | Full same | τ | main score | branch score |
+|---|---|:---:|:---:|:---:|:---:|:---:|:---:|
+| 1 | How does the retrieval pipeline work end-to-end? | ✗ | ✗ | ✗ | -0.333 | 0.896 | 0.922 |
+| 2 | How does the plugin interact with chat messages? | ✓ | ✗ | ✗ | 0.333 | 0.988 | 0.981 |
+| 3 | How does the keyword index combine with vecto... | ✗ | ✗ | ✗ | -1.000 | 0.897 | 0.949 |
+| 4 | Where is the embedder factory defined? | ✗ | ✗ | ✗ | 0.333 | 0.900 | 0.877 |
+| 5 | Where is the LanceDB store implementation? | ✗ | ✗ | ✗ | 1.000 | 0.900 | 0.968 |
+| 6 | Find all usages of the retrieve function | ✓ | ✗ | ✗ | 1.000 | 1.000 | 1.000 |
+| 7 | Find all usages of SearchResult type | ✓ | ✗ | ✗ | 1.000 | 0.992 | 0.990 |
+| 8 | How does the chunker factory register new lan... | ✗ | ✗ | ✗ | 1.000 | 0.900 | 0.981 |
+| 9 | What is the default minScore configuration? | ✗ | ✗ | ✗ | -1.000 | 0.914 | 0.876 |
+| 10 | How does the session logger capture token usage? | ✗ | ✗ | ✗ | -0.333 | 0.919 | 0.951 |
+| 11 | How does L2 normalization affect vector searc... | ✗ | ✗ | ✗ | 0.333 | 0.900 | 0.925 |
+| 12 | What is the MetadataFilter interface used for? | ✗ | ✗ | ✗ | 1.000 | 0.921 | 0.846 |
+| 13 | How does the background indexer handle file c... | ✓ | ✗ | ✗ | 1.000 | 0.995 | 0.987 |
+| 14 | Where is the config validation logic? | ✗ | ✗ | ✗ | -1.000 | 0.900 | 0.789 |
+| 15 | How are PDF documents chunked and indexed? | ✗ | ✗ | ✗ | -1.000 | 0.900 | 0.864 |
+| 16 | What embedding providers are supported? | ✗ | ✗ | ✗ | 0.333 | 0.882 | 0.908 |
+| 17 | How does the CLI parse and dispatch commands? | ✗ | ✗ | ✗ | 0.333 | 0.938 | 0.959 |
+| 18 | How does the session logger persist events? | ✗ | ✗ | ✗ | 1.000 | 0.822 | 0.959 |
+| 19 | What is the manifest schema version used for? | ✗ | ✗ | ✗ | -0.333 | 0.962 | 0.852 |
+| 20 | How does the OpenCode plugin register tools? | ✗ | ✗ | ✗ | 0.333 | 0.932 | 0.957 |
+| 21 | Where is the globMatch function defined? | ✓ | ✗ | ✗ | 0.000 | 0.963 | 0.915 |
+| 22 | How does the proxy-aware HTTP client work? | ✗ | ✗ | ✗ | 1.000 | 0.900 | 0.944 |
+| 23 | What is the FETCH_OVERFETCH_FACTOR constant? | ✓ | ✗ | ✗ | 1.000 | 0.955 | 0.816 |
+| 24 | How does the TUI settings menu work? | ✓ | ✗ | ✗ | 1.000 | 0.954 | 0.984 |
+| 25 | How are image descriptions generated? | ✗ | ✗ | ✗ | 1.000 | 0.900 | 0.931 |
+| 26 | retrieve function | ✓ | ✗ | ✗ | 0.667 | 1.000 | 0.939 |
+| 27 | KeywordIndex class | ✗ | ✗ | ✗ | 1.000 | 0.914 | 0.944 |
+| 28 | embedder factory | ✗ | ✗ | ✗ | -0.333 | 0.900 | 0.949 |
+| 29 | vector store LanceDB | ✓ | ✗ | ✗ | 1.000 | 1.000 | 0.994 |
+| 30 | find usages SearchResult | ✓ | ✗ | ✗ | 0.667 | 1.000 | 1.000 |
+| 31 | chunkFile method | ✓ | ✗ | ✗ | 0.333 | 0.950 | 0.901 |
+| 32 | embedBatch function | ✓ | ✗ | ✗ | 1.000 | 1.000 | 1.000 |
+| 33 | retriever retrieve vector search | ✗ | ✗ | ✗ | 1.000 | 0.872 | 0.959 |
+| 34 | session logger token | ✓ | ✗ | ✗ | 0.333 | 0.990 | 0.975 |
+| 35 | config validation | ✗ | ✗ | ✗ | -1.000 | 0.900 | 0.790 |