v0.3.0 per changelog

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Andrey Golovanov

.gitignore

@@ -126,6 +126,12 @@ exports/
 *.pickle
 *.pkl
 
+# Session handoff notes
+HANDOFF.md
+
+# uv lock file (not used for distribution)
+uv.lock
+
 # Temporary analysis & CLI output
 *_results.json
 results*.json
@@ -156,3 +162,11 @@ dev/perf_plots/
 # Experiment results (generated outputs, not source configs)
 experiments/*/results/
 experiments/**/results/
+
+# Research project outputs (simulation results, generated scenarios)
+# Config files (config.yml, objective.yml, etc.) ARE tracked.
+research_projects/*/results/
+research_projects/**/results/
+research_projects/**/phase*/
+research_projects/lm_*/
+research_projects/square_mesh_*/

CHANGELOG.md

@@ -5,6 +5,31 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.3.0] - 2026-03-26
+
+### Fixed
+
+- All metric modules now expand deduplicated flow_results by `occurrence_count` before statistical computation
+- Removed dead `iteration_metrics` extraction from iterops (field never existed in ngraph)
+
+### Added
+
+- Per-direction BAC (`per_flow` field on `BacResult`) for directional asymmetry analysis
+- Shared utilities in `metrics/common.py`: `expand_flow_results`, `canonical_dc`, `baseline_demand_map`
+- Mini DC-BB verification scenario with hand-calculated metric assertions
+- DC-BB autoresearch framework: scenario generator, structural analysis, parametric sweep, generation loop
+- CLI subcommands: `netlab autoresearch structural-analysis`, `sweep`, `cross-sweep`
+
+### Changed
+
+- Scenario generator produces per-mode TMP workflow steps (`tm_lh_path`, `tm_combined`, etc.) instead of single `tm_placement`
+
+## [0.2.2] - 2026-03-15
+
+### Changed
+
+- Relicensed from AGPL-3.0 to MIT
+
 ## [0.2.1] - 2026-02-02
 
 ### Added

netlab/autoresearch/backend.py

@@ -3,7 +3,7 @@
 Provides:
 - LLMBackend: ABC for single-shot text generation
 - MockBackend: scripted responses for testing
-- ClaudeCLIBackend: subprocess wrapper around `claude --print`
+- ClaudeCLIBackend: subprocess wrapper around `claude -p`
 - CodexCLIBackend: subprocess wrapper around `codex exec`
 - OpenAICompatibleBackend: HTTP client for OpenAI-compatible APIs
 """
@@ -53,7 +53,10 @@ def __init__(self, model: str = "opus") -> None:
         self.model = model
 
     def generate(self, prompt: str, system: str = "") -> str:
-        cmd = ["claude", "--print", "-p", prompt]
+        # Claude Code headless mode uses -p/--print with the prompt as its value.
+        cmd = ["claude", "-p", prompt]
+        if self.model:
+            cmd.extend(["--model", self.model])
         if system:
             cmd.extend(["--system-prompt", system])
         result = subprocess.run(