code-context-control 2.63.1__tar.gz → 2.63.2__tar.gz

{code_context_control-2.63.1 → code_context_control-2.63.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-context-control
-Version: 2.63.1
+Version: 2.63.2
 Summary: Local MCP code-intelligence for AI coding tools: surgical search/read/edit, agent-config version history, path-level access + masking guards, and a multi-project hub.
 Author-email: Dimitri Tselenchuk <dtselenc@gmail.com>
 License-Expression: Apache-2.0
@@ -310,7 +310,7 @@ c3 bench aider                    # Aider Polyglot suite (external; burns real A
 c3 bench swe                      # SWE-bench Lite (external)
 ```
 
-The session benchmark's baseline models a *competent* agent working without C3 — one targeted search, each file read once — and scores answer quality alongside tokens. A run against C3's own repository (v2.43.0, 2026-07-02) measured **51.8% token savings (2.07×)** at quality parity. Your numbers will differ with your project's shape; that's why the harness ships with the tool.
+The session benchmark's baseline models a *competent* agent working without C3 — one targeted search, each file read once — and scores answer quality alongside tokens. Runs against C3's own repository land around **~50% token savings (2×)** at quality parity — 51.8% at v2.43.0, 49.9% at v2.63.2, with C3 scoring 98.8% on answer quality against the baseline's 96.5% in both. File sampling is deterministic (largest files first, no RNG), so a given commit reproduces the same figure. Your numbers will differ with your project's shape; that's why the harness ships with the tool.
 
 C3 also records real per-tool usage to `.c3/tool_telemetry.jsonl`, so estimates can be checked against what actually happened.
 

{code_context_control-2.63.1 → code_context_control-2.63.2}/README.md

@@ -247,7 +247,7 @@ c3 bench aider                    # Aider Polyglot suite (external; burns real A
 c3 bench swe                      # SWE-bench Lite (external)
 ```
 
-The session benchmark's baseline models a *competent* agent working without C3 — one targeted search, each file read once — and scores answer quality alongside tokens. A run against C3's own repository (v2.43.0, 2026-07-02) measured **51.8% token savings (2.07×)** at quality parity. Your numbers will differ with your project's shape; that's why the harness ships with the tool.
+The session benchmark's baseline models a *competent* agent working without C3 — one targeted search, each file read once — and scores answer quality alongside tokens. Runs against C3's own repository land around **~50% token savings (2×)** at quality parity — 51.8% at v2.43.0, 49.9% at v2.63.2, with C3 scoring 98.8% on answer quality against the baseline's 96.5% in both. File sampling is deterministic (largest files first, no RNG), so a given commit reproduces the same figure. Your numbers will differ with your project's shape; that's why the harness ships with the tool.
 
 C3 also records real per-tool usage to `.c3/tool_telemetry.jsonl`, so estimates can be checked against what actually happened.
 

{code_context_control-2.63.1 → code_context_control-2.63.2}/cli/c3.py

@@ -92,7 +92,7 @@ console = Console() if HAS_RICH else None
 # Config
 CONFIG_DIR = ".c3"
 CONFIG_FILE = ".c3/config.json"
-__version__ = "2.63.1"
+__version__ = "2.63.2"
 
 
 def _compress_file_cli(compressor, path, mode="smart", **kw):

{code_context_control-2.63.1 → code_context_control-2.63.2}/code_context_control.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-context-control
-Version: 2.63.1
+Version: 2.63.2
 Summary: Local MCP code-intelligence for AI coding tools: surgical search/read/edit, agent-config version history, path-level access + masking guards, and a multi-project hub.
 Author-email: Dimitri Tselenchuk <dtselenc@gmail.com>
 License-Expression: Apache-2.0
@@ -310,7 +310,7 @@ c3 bench aider                    # Aider Polyglot suite (external; burns real A
 c3 bench swe                      # SWE-bench Lite (external)
 ```
 
-The session benchmark's baseline models a *competent* agent working without C3 — one targeted search, each file read once — and scores answer quality alongside tokens. A run against C3's own repository (v2.43.0, 2026-07-02) measured **51.8% token savings (2.07×)** at quality parity. Your numbers will differ with your project's shape; that's why the harness ships with the tool.
+The session benchmark's baseline models a *competent* agent working without C3 — one targeted search, each file read once — and scores answer quality alongside tokens. Runs against C3's own repository land around **~50% token savings (2×)** at quality parity — 51.8% at v2.43.0, 49.9% at v2.63.2, with C3 scoring 98.8% on answer quality against the baseline's 96.5% in both. File sampling is deterministic (largest files first, no RNG), so a given commit reproduces the same figure. Your numbers will differ with your project's shape; that's why the harness ships with the tool.
 
 C3 also records real per-tool usage to `.c3/tool_telemetry.jsonl`, so estimates can be checked against what actually happened.
 

{code_context_control-2.63.1 → code_context_control-2.63.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "code-context-control"
-version = "2.63.1"
+version = "2.63.2"
 description = "Local MCP code-intelligence for AI coding tools: surgical search/read/edit, agent-config version history, path-level access + masking guards, and a multi-project hub."
 readme = "README.md"
 requires-python = ">=3.10"