feat(swe-dataset-factory): vendor DeepSWE real_pr factory and N=20 product

Ship the swe-factory package, unit tests, small fixtures, and certified
DeepSWE Real-PR N=20 product packs under swe-dataset-factory/ as a
clean subproject (no clash with src/swe_forge). Document the path and
eval-deepswe entry in the root README; unignore product jsonl/json.

Author: echobt

.gitignore

@@ -92,3 +92,16 @@ tasks-*/
 *.jsonl
 test-output/
 *.log
+
+# Vendored deepswe product under swe-dataset-factory/ (keep product evidence)
+!swe-dataset-factory/
+!swe-dataset-factory/**
+!swe-dataset-factory/**/*.jsonl
+!swe-dataset-factory/**/*.json
+!swe-dataset-factory/**/*.md
+# still never commit subproject secrets or caches
+swe-dataset-factory/.env
+swe-dataset-factory/.venv/
+swe-dataset-factory/**/__pycache__/
+swe-dataset-factory/**/node_modules/
+swe-dataset-factory/fixtures/harbor_motors/.ts_deps/

README.md

@@ -4,7 +4,7 @@
 
 **Real-code software engineering benchmarks for Platform agents**
 
-[![License](https://img.shields.io/github/license/PlatformNetwork/Agent-SWE)](https://github.com/PlatformNetwork/Agent-SWE/blob/main/LICENSE)
+[![License](https://img.shields.io/github/license/BaseIntelligence/Agent-SWE)](https://github.com/BaseIntelligence/Agent-SWE/blob/main/LICENSE)
 [![Platform SDK](https://img.shields.io/badge/Platform-SDK-black)](https://github.com/PlatformNetwork/platform)
 [![Python](https://img.shields.io/badge/python-3.12+-blue.svg)](https://www.python.org/)
 [![SWE-Forge](https://img.shields.io/badge/SWE--Forge-CortexLM-blue)](https://huggingface.co/datasets/CortexLM/swe-forge)
@@ -64,7 +64,7 @@ flowchart LR
 ## Install
 
 ```bash
-git clone https://github.com/PlatformNetwork/Agent-SWE.git
+git clone https://github.com/BaseIntelligence/Agent-SWE.git
 cd Agent-SWE
 pip install -e ".[dev]"
 ```
@@ -208,9 +208,37 @@ Agent-SWE/
 │   ├── export/
 │   ├── swe/
 │   └── synthetic/
+├── swe-dataset-factory/   # DeepSWE Real-PR product + factory (subproject)
 └── tests/
 ```
 
+## DeepSWE Real-PR factory (`swe-dataset-factory/`)
+
+Agent-SWE also vendors the **SWE Dataset Factory** under [`swe-dataset-factory/`](swe-dataset-factory/): a separate package that ships Docker-verifiable DeepSWE / Harbor packs from live-mined public PRs.
+
+| Surface | Path | Role |
+|---|---|---|
+| Product N=20 | `swe-dataset-factory/datasets/deepswe_v1/` | Certified Real-PR packs (`source_track=real_pr`) |
+| Seed archive | `swe-dataset-factory/datasets/deepswe_v1_seed5_archive/` | Historical seed5 only |
+| Soft panel | `swe-dataset-factory/datasets/panel_deepswe_5pack/` | Report JSON evidence |
+| CLI package | `swe-dataset-factory/src/swe_factory/` | `swe-factory` entrypoint |
+
+Install and use from the subproject (Python >= 3.12):
+
+```bash
+cd swe-dataset-factory
+pip install -e ".[dev]"
+cp .env.example .env   # never commit .env
+
+# DeepSWE-grade Pier + mini-swe serial eval (product root N=20)
+swe-factory eval-deepswe --product-root datasets/deepswe_v1 --help
+```
+
+Details, ship commands, and honesty boundaries live in
+[`swe-dataset-factory/README.md`](swe-dataset-factory/README.md).
+Forge mining (`swe-forge` under `src/`) remains the existing Agent-SWE path;
+the factory is a clean subproject so it does not clash with `src/swe_forge`.
+
 ## Platform Integration
 
 Agent-SWE is designed to feed Platform challenge validators with deterministic repository-repair tasks. Validators can sample tasks, run agent patches in isolated workspaces, and turn task completion rates into raw challenge scores for Platform.

swe-dataset-factory/.env.example

@@ -0,0 +1,19 @@
+# OpenRouter (required for live teacher/panel stages)
+OPENROUTER_API_KEY=sk-or-v1-replace-me
+OPENROUTER_BASE_URL=https://openrouter.ai/api/v1
+
+# Models — OpenRouter ids only
+FACTORY_TEACHER_MODEL=anthropic/claude-opus-4.8
+# Real-PR wave panel pair (VAL-RPANEL-001); Opus optional / not required.
+FACTORY_PANEL_MODELS=x-ai/grok-4.5,moonshotai/kimi-k2.6
+
+# Hard spend cap (exact + unresolved reserved must stay under this)
+FACTORY_BUDGET_USD=600
+
+# Optional GitHub token for PR mining rate limits
+# GITHUB_TOKEN=
+
+# Oxylabs Web Scraper API (required for live GitHub HTTP page/raw fetches)
+# Always uses source=universal only; never amazon/marketplace templates.
+# OXYLABS_USERNAME=
+# OXYLABS_PASSWORD=

swe-dataset-factory/.gitignore

@@ -0,0 +1,52 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.egg-info/
+.eggs/
+dist/
+build/
+.mypy_cache/
+.ruff_cache/
+.pytest_cache/
+.coverage
+htmlcov/
+.venv/
+venv/
+*.egg
+
+# Secrets and local env
+.env
+.env.local
+*.pem
+
+# Local documentation / validation evidence (never publish)
+.docs-evidence/
+
+# Generated work caches and bulk non-product datasets (not vendored here)
+datasets/_clone_cache/
+datasets/_work*/
+datasets/sdf-ship-*/
+datasets/mine_allowlist*/
+datasets/real_pr_pool_live_*/
+datasets/_div_probe/
+datasets/_ship_work*/
+datasets/live_materials/
+datasets/micro_keep_*/
+datasets/ut_*/
+
+# Harbor motor dual-run jest node_modules cache (local only)
+fixtures/harbor_motors/.ts_deps/
+**/node_modules/
+
+# OS / editor
+.DS_Store
+.idea/
+.vscode/
+*.swp
+*~
+
+# Docker / runtime
+*.log
+tmp/
+.cache/