From a23ad342b117402a800e05872a6415656eb1e728 Mon Sep 17 00:00:00 2001
From: bokiko <bokiko@users.noreply.github.com>
Date: Wed, 18 Mar 2026 00:30:08 +0000
Subject: [PATCH] kyzn(deep): improve performance
 [run:20260318-002459-8eee8d4b]
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Health: 96 → 96 (→0)
Cost: $0.96201645
---
 .kyzn/.gitignore   |  3 +++
 .kyzn/config.yaml  | 35 +++++++++++++++++++++++++++++++++++
 scripts/indexer.py | 27 +++++++++++++++++----------
 3 files changed, 55 insertions(+), 10 deletions(-)
 create mode 100644 .kyzn/.gitignore
 create mode 100644 .kyzn/config.yaml

diff --git a/.kyzn/.gitignore b/.kyzn/.gitignore
new file mode 100644
index 0000000..101f1a8
--- /dev/null
+++ b/.kyzn/.gitignore
@@ -0,0 +1,3 @@
+# kyzn — gitignored local data
+history/
+reports/
diff --git a/.kyzn/config.yaml b/.kyzn/config.yaml
new file mode 100644
index 0000000..c48c9b9
--- /dev/null
+++ b/.kyzn/config.yaml
@@ -0,0 +1,35 @@
+# kyzn configuration — commit this file
+# Generated by: kyzn init
+# Date: 2026-03-18T00:24:58Z
+
+project:
+  name: bloxcue
+  type: generic
+
+preferences:
+  mode: 
+How aggressive should improvements be?
+  1) Deep — real improvements only (no cosmetic changes)
+  2) Clean — dead weight cleanup (remove unused code, fix naming)
+  3) Full — everything (maximum value per run)
+
+Choice [1]: 1
+  budget: Budget per run (USD) [2.50]: 2.50
+  max_turns: 30
+  diff_limit: 2000
+  trust: 
+Trust level for auto-merging?
+  1) Guardian — always create PR, always wait for approval (recommended)
+  2) Autopilot — auto-merge if build passes + tests pass + diff < threshold
+
+Choice [1]: 1
+  on_build_fail: 
+If the build breaks after improvements, what should kyzn do?
+  1) Write a report explaining what happened (recommended)
+  2) Silently discard the branch
+  3) Create a draft PR so you can see what was attempted
+
+Choice [1]: 1
+
+focus:
+  priorities: []
diff --git a/scripts/indexer.py b/scripts/indexer.py
index a803459..c9de44c 100755
--- a/scripts/indexer.py
+++ b/scripts/indexer.py
@@ -822,8 +822,10 @@ def load_index() -> Dict:
             _index_cache = json.loads(INDEX_FILE.read_text())
             _index_mtime = current_mtime
             return _index_cache
-        except Exception:
-            pass
+        except (OSError, json.JSONDecodeError) as e:
+            print(f"[bloxcue] Warning: failed to load index cache ({e}), rebuilding", file=sys.stderr)
+        except Exception as e:
+            print(f"[bloxcue] Warning: unexpected error loading index ({e}), rebuilding", file=sys.stderr)
 
     return build_index()
 
@@ -955,12 +957,15 @@ def search(query: str, limit: int = 5) -> List[Dict]:
 
         # Bigram bonus (phrase matching)
         entry_bigrams = entry.get("bigrams", [])
+        entry_bigrams_set = set(entry_bigrams)
         for qbigram in query_bigrams:
-            for ebigram in entry_bigrams:
-                if qbigram == ebigram:
-                    score += 5.0
-                elif qbigram in ebigram or ebigram in qbigram:
-                    score += 2.5
+            if qbigram in entry_bigrams_set:
+                score += 5.0
+            else:
+                for ebigram in entry_bigrams:
+                    if qbigram in ebigram or ebigram in qbigram:
+                        score += 2.5
+                        break
 
         # Fuzzy fallback: catch matches that BM25 missed (typos, partial matches)
         if score == 0:
@@ -1027,6 +1032,7 @@ def search(query: str, limit: int = 5) -> List[Dict]:
     # MMR diversity: reduce redundancy in top results
     if len(results) > 1:
         selected = [results[0]]
+        selected_term_sets = [set(results[0]["entry"].get("keywords", []))]
         remaining = results[1:]
 
         while remaining and len(selected) < limit:
@@ -1036,8 +1042,7 @@ def search(query: str, limit: int = 5) -> List[Dict]:
             for i, candidate in enumerate(remaining):
                 cand_terms = set(candidate["entry"].get("keywords", []))
                 max_overlap = 0.0
-                for sel in selected:
-                    sel_terms = set(sel["entry"].get("keywords", []))
+                for sel_terms in selected_term_sets:
                     if cand_terms and sel_terms:
                         union = len(cand_terms | sel_terms)
                         if union > 0:
@@ -1051,7 +1056,9 @@ def search(query: str, limit: int = 5) -> List[Dict]:
                     best_mmr = mmr
                     best_idx = i
 
-            selected.append(remaining.pop(best_idx))
+            chosen = remaining.pop(best_idx)
+            selected.append(chosen)
+            selected_term_sets.append(set(chosen["entry"].get("keywords", [])))
 
         results = selected