Man1Lab v1.3.0

Man1Lab v1.3.0

Execution Runtime · Materialization · Durable Local Runs

Release date: 2026-07-14

---

What's New

Execution Runtime

Man1Lab can now run a materialized reproduction plan on a durable local execution platform.

The Runtime layer now manages:

• Execution Engine lifecycle and sequential scheduling
• Runtime-owned ExecutionStore with atomic snapshots
• Append-only traces, artifact manifests, and reports
• Cross-process interruption, reload, and conservative reconciliation
• Resume of reusable successful tasks after artifact validation

---

Planning-to-Execution Materialization

Abstract Planning graphs are converted into immutable local task specifications before an ExecutionRun is created.

ExecutionStrategy + ExecutionGraph
        │
        ▼
Execution Materialization
        │
        ▼
READY gate
        │
        ▼
LocalExecutor run + durable report

Incomplete, conflicting, or unsupported cases are rejected safely as BLOCKED, UNSUPPORTED, or RECONCILIATION_REQUIRED — before execution begins.

---

Execution Evidence & Preparation

Discovery can now supply typed, pinned execution evidence for bounded reproduction workflows.

Supported evidence types include:

• Repository (workspace or pinned Git)
• Dataset and checkpoint sources
• Configuration (existing file or deterministic render)

Bounded preparation operations produce manifests, checksums, safe extraction, and receipts for inspectable runs.

---

One-Command Reproduction Path

When Discovery evidence is complete and conflict-free, Man1Lab can orchestrate the full path in one application flow:

Analysis → Discovery → Planning → Materialization → Execution → Report

Facade and console execution commands are integrated with LocalExecutor for local subprocess runs with durable logs and artifacts.

---

Improvements

• Stronger readiness gate between Planning and Execution
• More inspectable preparation and workload artifacts
• Better crash recovery for completed tasks
• Safer handling of missing or unsupported evidence
• Provider-driven DeiT release acceptance covering pinned evidence and real process interruption
• Expanded regression and release acceptance testing

---

Compatibility

No breaking changes to existing Planning, Runtime, Facade, or console boundaries.

CLI commands, SDK APIs, configuration files, and canonical planning artifacts remain compatible with previous releases.

Reproduction executes only when Discovery supplies complete, conflict-free, pinned execution evidence. This release does not claim arbitrary-paper reproduction. Authenticated downloads, Docker, remote GPU, cluster, and robot backends remain out of scope.

---

Current Workflow

Paper
   │
   ▼
Analysis
   │
   ▼
Research Resource Discovery
   │
   ▼
Execution Planning
   │
   ▼
Materialization
   │
   ▼
Local Execution
   │
   ▼
Report

Execution is available for bounded, evidence-backed local runs. Broader backend support remains under active development.

---

Statistics

• 997 tests collected
• Execution Engine
• Runtime ExecutionStore
• Planning-to-Execution Materialization
• Typed Execution Evidence
• LocalExecutor Platform Integration
• One-Command Orchestration
• DeiT Release Acceptance

---

Thank you for using Man1Lab.

Man1Lab v1.2.4

Man1Lab v1.2.4

Interactive Console · Workspace Persistence · Decision Foundation

Release date: 2026-07-09

---

What's New

Interactive Console

Man1Lab now provides a guided interactive console for AI paper reproduction.

man1lab

New console capabilities include:

• Guided workflow with next-step hints
• plan-all one-command planning pipeline
• Command history and line editing
• Workspace resume
• Improved startup banner and status display

---

Persistent Workspace

Every stage now produces reusable engineering artifacts.

workspace/
├── analysis/
├── discovery/
├── planning/
└── decision/

Artifacts are automatically persisted, allowing interrupted workflows to resume without repeating completed stages.

---

Better Decision Quality

Discovery and Execution Planning have been significantly improved.

New capabilities include:

• Research Asset pipeline
• Explainable Confidence
• Decision Trace
• Execution Graph
• Golden Benchmark framework

Planning decisions are now more explainable, consistent, and reproducible.

---

Platform Runtime

The Runtime layer now manages:

• Resource lifecycle
• Runtime sessions
• Lazy initialization
• Startup profiling
• Multi-provider model management

---

Improvements

• Improved console guidance
• Better workspace diagnostics
• Resume support across planning stages
• More robust LLM connection handling
• Better prompt resource resolution
• Improved model management
• Expanded regression testing

---

Compatibility

No breaking changes.

Existing CLI commands, SDK APIs, configuration files, and canonical artifacts remain compatible with previous releases.

---

Current Workflow

Paper
   │
   ▼
Analysis
   │
   ▼
Research Resource Discovery
   │
   ▼
Execution Planning
   │
   ▼
Execution (Coming Soon)

Execution remains under active development.

---

Statistics

• 826 tests passing
• Interactive Console
• Multi-provider LLM Platform
• Runtime Architecture
• Workspace Persistence
• Decision Foundation
• Golden Benchmarks

---

Thank you for using Man1Lab.

Man1Lab v1.2.3

Man1Lab v1.2.3 — Platform Runtime & Interactive Console

Release date: 2026-07-08
Repository: https://github.com/maniac1um/Man1Lab
Status: Release Candidate

---

Overview

v1.2.3 introduces Platform Runtime as first-class process infrastructure and ships the Interactive Console — a REPL-style interface via man1lab with no arguments.

v1.2.2 delivered the LLM Platform and first-run experience. v1.2.3 delivers runtime ownership, infrastructure integration, and interactive console.

Milestone: Platform Capability
Previous: v1.2.2 (LLM Platform & First-run Experience)
Next: Repository Understanding (v1.3)

---

Highlights

• Platform Runtime — lifecycle, context, lazy initialization, resource manager, session
• Runtime profiling — man1lab profile for startup observation
• Runtime integration — infrastructure resolved through RuntimeInfrastructure (8.5.1)
• Interactive Console — man1lab (no args) with built-in commands (help, doctor, analyze, reproduce, …)
• 765 unit tests passing
• MIT License — unchanged

---

Architecture Changes

New cross-cutting layer between Facade and business workflows:

CLI · Console · SDK
        ↓
Platform Facade (Man1Lab)
        ↓
PlatformRuntime
    ├── RuntimeContext → RuntimeResourceManager
    ├── RuntimeSession → SessionWorkspace
    └── RuntimeProfiler
        ↓
Business Workflows (Analysis → Discovery → Execution Planning → …)

No changes to canonical artifact pipeline or reproduction workflow semantics.

Full specification: architecture/RUNTIME.md

---

Interactive Console

man1lab          # enter interactive console
man1lab --help   # traditional CLI subcommands unchanged

Console commands delegate to the Platform Facade. Session opens on console start.

---

Execution Planning & LLM Platform

Unchanged from v1.2.2 — six embedded Execution Planning providers; LLM Model Registry and man1lab model CLI.

---

Compatibility

None. All v1.2.2 CLI subcommands, SDK methods, and workflow behavior remain compatible. Console is additive.

---

Migration Notes

From v1.2.2

No migration required.

pip install --upgrade man1lab
man1lab --version   # 1.2.3
man1lab doctor

Optional: run man1lab with no args to explore the interactive console.

---

Known Limitations

Inherited from v1.2.2 — see CURRENT_STATUS.md:

• Session workspace is in-memory only (no persistence)
• MCP / REST interfaces not implemented
• SDK does not expose model management methods
• No daemon mode

---

Tests

765 unit tests passing (pixi run test).

---

Upgrade from v1.2.2

pip install --upgrade man1lab
man1lab --version
man1lab doctor

No configuration migration required.

---

Phase Audits

reviews/8.1_runtime_performance_audit/ through reviews/8.6_man1lab_console/, reviews/8.5.1_runtime_integration/

Man1Lab v1.2.2

Man1Lab v1.2.2 — LLM Platform & First-run Experience

Release date: 2026-07-08
Repository: https://github.com/maniac1um/Man1Lab
Status: Release Candidate

---

Overview

v1.2.2 completes the LLM Platform layer and delivers a polished first-run experience for external users installing Man1Lab via pip. New users can configure and switch LLM providers entirely through the CLI without editing configuration files.

v1.2.1 delivered complete Execution Planning. v1.2.2 delivers production-ready model management and release hardening for PyPI publication.

Milestone: Platform Capability
Previous: v1.2.1 (Execution Planning Stabilization)
Next: Repository Understanding (v1.3)

---

Highlights

• LLM Provider foundation — LLMManager, ModelRegistry, ProviderRegistry
• Multi-provider support — OpenAI, DeepSeek, Anthropic
• Model Management CLI — full profile lifecycle via man1lab model
• First-run wizard — interactive model setup during man1lab init
• Portable profiles — man1lab model export / import (no secrets)
• Doctor LLM validation — connection health and configuration checks
• 614 unit tests passing
• MIT License — LICENSE at repository root; declared in pyproject.toml and CITATION.cff
• Release hardening — documentation sync, packaging validation, review directory normalization

---

Architecture Changes

No workflow or agent behavior changes. New platform layer:

CLI → Platform Facade → LLMManager → ModelRegistry → ProviderRegistry → LLMProvider

Business agents remain unaware of provider selection. Resolution happens inside LLMManager.

---

Execution Planning

Unchanged from v1.2.1 — six embedded providers with shared Decision Foundation (ADR-0018).

Workflow → Services → Providers → Decision Foundation → Builder → ExecutionStrategy

---

LLM Platform

Phase	Deliverable	Status
7.1	LLM Provider foundation	✅
7.2	Model Registry + persistence	✅
7.3	Anthropic provider	✅
7.4	Model Management CLI	✅
7.5	First-run Experience	✅

Component	Location
LLMManager	providers/llm/manager.py
ModelRegistry	providers/llm/registry.py
ProviderRegistry	providers/llm/provider_registry.py
Profile persistence	conf/llm/user_profiles.yaml
Providers	OpenAI, DeepSeek, Anthropic

---

CLI

Lifecycle

Command	Description
man1lab init	Workspace setup + optional first-model wizard
man1lab init --skip-model-config	Legacy init behavior
man1lab doctor	Environment + LLM validation
man1lab clean	Remove regeneratable artifacts

Model Management

Command	Description
man1lab model list	List configured profiles
man1lab model current	Show active profile
man1lab model use <profile>	Switch active profile
man1lab model add	Interactive profile creation
man1lab model remove <profile>	Remove profile
man1lab model rename <old> <new>	Rename profile
man1lab model test [profile]	Provider health check
man1lab model validate	Registry validation
man1lab model export <file>	Export portable config (no secrets)
man1lab model import <file>	Import profiles (--replace, --skip-existing)

Workflow (unchanged)

reproduce, analyze, discover, plan, execute, config, version

---

Model Management

Profiles are stored in conf/llm/user_profiles.yaml. API keys remain environment references only — never written to YAML.

Portable export format:

active: claude
profiles:
  claude:
    provider: anthropic
    model: claude-sonnet-4
    api_key_reference: ANTHROPIC_API_KEY
    enabled: true

---

First-run Experience

Recommended installation flow:

pip install man1lab
man1lab init          # configure first model interactively
man1lab doctor        # verify environment + LLM
man1lab reproduce paper.pdf

After successful init with wizard:

Workspace initialized.
Active model: default
Provider: Openai
Run `man1lab doctor` to verify your environment.

---

SDK

Unchanged public API:

from man1lab import Man1Lab, PLATFORM_VERSION, __version__

Model management is CLI-only in v1.2.2. SDK method exports planned for v1.3 polish.

---

Package Distribution

• Version: 1.2.2
• pip install man1lab
• Console script: man1lab
• Module entry: python -m man1lab
• Bundled: conf/, prompts/, .env.example
• Excluded: outputs/, logs/, mlruns/, workspace/tasks/

---

Bug Fixes

• Fixed DoctorCheck import path in CLI tests
• Synchronized version strings across application/version.py, pyproject.toml, and CLI --version
• Corrected stale test counts and version references in documentation

---

Breaking Changes

None. All v1.2.1 CLI commands, SDK methods, and workflow behavior remain compatible.

---

Migration Notes

From v1.2.1

No migration required. Existing .env and Hydra configuration continue to work.

Optional: run man1lab init in an existing workspace to add named profiles via the wizard, or use man1lab model add.

From v1.2.0 or earlier

Follow v1.2.1 migration notes — no additional steps for v1.2.2.

---

Known Limitations

Inherited from v1.2.1 — see CURRENT_STATUS.md:

• Discovery verification is shallow
• MCP / REST interfaces not implemented
• SDK does not expose model management methods
• execution_planning.enabled=false transitional path remains

---

Future Roadmap (v1.3)

Direction	Description
Repository Understanding	Semantic mapping of repo structure to analysis modules
RepositoryKnowledge artifact	New canonical object (design TBD)
SDK model management	Export list_models(), use_model(), etc. on Man1Lab
MCP / REST interfaces	Reserved in architecture

See ROADMAP.md.

---

Upgrade from v1.2.1

pip install --upgrade man1lab
man1lab --version   # 1.2.2
man1lab doctor

No configuration migration required.

---

Contributors

Man1Lab v1.2.2 Release Candidate — Platform Capability milestone.

Phase audits: reviews/7.1_llm_provider_foundation/ through reviews/7.5_first_run_experience/, reviews/1.2.2_release_candidate/.

Man1Lab v1.2.1

Man1Lab v1.2.1 — Execution Planning Stabilization

Release date: 2026-07-08
Repository: https://github.com/maniac1um/Man1Lab
Status: Stable platform capability release

---

Overview

v1.2.1 completes Execution Planning — the third Platform Capability. All six embedded engineering decision providers are implemented with a shared Decision Foundation. This release stabilizes architecture and documentation without changing engineering behavior.

v1.2.0 delivered platform interfaces and capability foundations. v1.2.1 delivers complete deterministic Execution Planning.

Milestone: Platform Capability
Previous: v1.2.0 (Release Candidate)
Next: Repository Understanding (v1.3)

---

Highlights

• Execution Planning complete — Strategy → Binding → Reuse → Adaptation → Generation → Risk
• Decision Foundation — shared ObservedFacts → DecisionDimensions → engineering decisions (ADR-0018)
• Six embedded providers — deterministic engineering decisions for all planning stages
• Architecture stabilization — legacy code removed, documentation synchronized
• 526 unit tests passing

---

Execution Planning Completion

Stage	Embedded provider	Status
Strategy Decision	EmbeddedStrategyProvider	✅
Resource Binding	EmbeddedResourceBindingProvider	✅
Reuse Planning	EmbeddedReuseProvider	✅
Adaptation Planning	EmbeddedAdaptationProvider	✅
Generation Planning	EmbeddedGenerationProvider	✅
Risk Assessment	EmbeddedRiskProvider	✅

Permanent architecture:

Workflow → Services → Providers → Decision Foundation → Builder → ExecutionStrategy

---

Decision Foundation

Internal package: providers/embedded/decision_foundation/

• facts.py — observed state extraction
• dimensions.py — engineering evaluation dimensions
• common.py — shared formatting helpers
• *_decision.py — per-stage engineering decisions

Not exported as canonical artifacts or public API.

---

GitHub Discovery

Unchanged from v1.2.0 — Collection, Evidence, Verification, Ranking providers (ADR-0016).

---

CLI

Unchanged API — man1lab init, doctor, reproduce, analyze, discover, plan, execute.

---

SDK

Unchanged API — from man1lab import Man1Lab.

---

Package Distribution

Unchanged from v1.2.0 — pip install man1lab, PEP 621 pyproject.toml.

---

Architecture Improvements

• Removed legacy execution_planning/stages.py
• Removed obsolete providers/embedded/execution_planning_skeleton.py
• Introduced decision_foundation/common.py for shared helpers
• Recorded Decision Foundation in ADR-0018
• Updated ADR-0017 for complete provider architecture

---

Documentation Improvements

• New architecture/EXECUTION_PLANNING.md
• Reorganized docs/reviews/ with numeric phase prefixes
• Synchronized README, CURRENT_STATUS, GETTING_STARTED, ARCHITECTURE
• Phase 7 architecture stabilization audit

---

Compatibility

• No breaking changes to CLI, SDK, or canonical artifact schemas
• No behavior changes to Execution Planning engineering decisions
• No changes to workflow ordering, builder semantics, or runtime contracts
• execution_planning.enabled=false transitional path unchanged

---

Known Limitations

Inherited from v1.2.0 — see CURRENT_STATUS.md:

• Discovery verification is shallow
• MCP / REST interfaces not implemented
• Review loop does not re-run Coder/Runner on patch
• Embedded Execution Planning uses deterministic rules only (no LLM)

---

Next Roadmap

Milestone	Focus
v1.3	Repository Understanding — RepositoryKnowledge artifact
Future	LLM-assisted planning providers (optional backends)
Future	MCP and REST platform interfaces

See ROADMAP.md.

---

Upgrade from v1.2.0

pip install --upgrade man1lab

No migration steps required. Engineering decision outputs are behaviorally identical to v1.2.0 embedded provider phases.

---

Related Documents

• CURRENT_STATUS.md
• EXECUTION_PLANNING.md
• ADR-0018
• Phase 7 audit

Man1Lab v1.2.0

Man1Lab v1.2.0 — Platform Capability Release

Release date: 2026-07-03
Codename: Platform Capability
Repository: https://github.com/maniac1um/Man1Lab
Status: Release Candidate

---

Overview

v1.2.0 completes Platform Capability — the first release where Man1Lab is a unified, installable platform with public interfaces (CLI, Python SDK), integrated Discovery and Execution Planning, and the GitHub Discovery Provider.

v1.1.0 delivered infrastructure foundation. v1.2.0 delivers capabilities and interfaces users interact with directly.

Milestone: Platform Capability
Previous: Foundation Complete (v1.1.0)
Next: Repository Understanding (v1.3)

---

Highlights

• Platform Facade — Man1Lab as the single composition root for all interfaces
• CLI — man1lab Typer application (init, doctor, reproduce, analyze, discover, plan, execute, config, version)
• Python SDK — pip install man1lab · from man1lab import Man1Lab
• Package distribution — PEP 621 pyproject.toml, console script, python -m man1lab
• Lifecycle commands — man1lab init, man1lab doctor
• Discovery integrated — DiscoveryWorkflow → ResearchResourceDiscovery (ADR-0013)
• GitHub Discovery Provider — Collection, Evidence, Verification, Ranking (ADR-0016)
• Execution Planning integrated — ExecutionPlanningWorkflow → ExecutionStrategy (ADR-0014)
• Strategy-driven Planner — TaskModel from ExecutionStrategy
• 419 unit tests passing

---

Architecture Changes

Platform interfaces

CLI  ·  Python SDK  ·  (Future MCP)  ·  (Future REST)
                    ↓
          Man1Lab (Platform Facade)
                    ↓
       TrackedWorkflowOrchestrator
                    ↓
  Reader → Discovery → Execution Planning → Planner
                    ↓
       Coder → Runner → Verification → Review → Report

Canonical artifact pipeline (implemented)

Stage	Artifact
Analysis	PaperReproductionAnalysis
Discovery	ResearchResourceDiscovery
Execution Planning	ExecutionStrategy
Planning	TaskModel
Implementation	Workspace
Execution	ExecutionResult
Reporting	ReportModel

Future artifacts (planned — not in v1.2.0)

Artifact	Milestone
RepositoryKnowledge	v1.3 Repository Understanding
Consolidated ReproductionReport	Future reporting evolution

ExecutionResult and ReportModel are implemented runtime/report outputs — not renamed to roadmap names above.

Configuration switches

Hydra settings (default enabled):

Switch	Effect
discovery.enabled	Run Discovery workflow; else empty discovery artifact
execution_planning.enabled	Run Execution Planning; else legacy Planner.run_legacy()

---

CLI Introduction

pip install man1lab
man1lab init
man1lab doctor
man1lab reproduce paper.pdf
man1lab analyze paper.pdf
man1lab discover paper.pdf
man1lab plan paper.pdf
man1lab version

All commands delegate to Man1Lab — no direct orchestrator access.

---

SDK Introduction

from man1lab import Man1Lab

client = Man1Lab()
client.init()
client.doctor()
report = client.reproduce("paper.pdf")

Public exports: Man1Lab, PLATFORM_VERSION, __version__.

---

GitHub Discovery Provider

First production external Discovery backend:

Stage	Scope
Collection	Paper-embedded URLs + conditional search
Evidence	Metadata, README, license, activity signals
Verification	Exists, accessible, archived, owner match (no clone)
Ranking	Deterministic rules — no LLM

Design: github-discovery-provider.md

---

Package Distribution

Artifact	Location
Package metadata	pyproject.toml
Console script	man1lab → interfaces.cli.app:run_cli
Module entry	python -m man1lab
Bundled resources	conf/, prompts/, .env.example

---

Breaking Changes

Change	Migration
Recommended entry point	Use man1lab reproduce or SDK — not python app.py
Planner input (when execution planning enabled)	Consumes ExecutionStrategy instead of implicit strategy from analysis alone
Orchestrator pipeline order	Discovery and Execution Planning stages run before Planner when enabled

Workflow artifacts remain backward-compatible. PaperReproductionAnalysis unchanged (ADR-0009).

app.py retained for maintainers only — not documented as user entry.

---

Migration Guide

From v1.1.x (developers)

1. pip install -e . or pip install man1lab
2. man1lab init — create workspace and .env
3. man1lab doctor — validate environment
4. Replace PYTHONPATH=. python app.py with man1lab reproduce paper.pdf
5. Optional: set GITHUB_TOKEN in .env for GitHub discovery

From v1.1.x (programmatic)

# Before (maintainer-only)
# PYTHONPATH=. python app.py

# After
from man1lab import Man1Lab
Man1Lab().reproduce("paper.pdf")

---

Compatibility

Area	Status
Pixi development	pixi install + pixi run man1lab
pip install	pip install man1lab
Legacy app.py	Maintainers only
MLflow tracking	Optional; TRACKING_BACKEND=noop to disable
Discovery / Execution Planning disable	Hydra flags for transitional paths

---

Known Limitations

Unchanged from prior releases — see CURRENT_STATUS.md:

• Full training reproduction not guaranteed on benchmark papers
• Review loop does not re-invoke Coder/Runner on patch
• RAG blocks delivery defects but not runtime API breakage
• Shallow Discovery verification does not guarantee runnable repos

---

Tests

419 unit tests passing (pixi run test).

Platform coverage: facade, CLI, SDK, package distribution, discovery, execution planning, GitHub provider, platform integration.

---

Related Documents

Document	Purpose
CURRENT_STATUS.md	Live state
ARCHITECTURE.md	Platform architecture
ROADMAP.md	Future milestones
v1.1.0	Foundation release
CHANGELOG.md	Detailed changelog

---

Documentation Freeze Audit

Audit performed for v1.2.0 Release Candidate documentation freeze (2026-07-03).

Updated documents

Document	Status
README.md	✅ Rewritten — homepage, CLI/SDK primary
docs/GETTING_STARTED.md	✅ Platform workflow
docs/architecture/ARCHITECTURE.md	✅ v1.2 interfaces, artifacts, scope
docs/CURRENT_STATUS.md	✅ v1.2.0, 419 tests, next milestone
docs/releases/v1.2.0.md	✅ Finalized
ROADMAP.md	✅ Created
docs/releases/README.md	✅ Updated
docs/README.md	✅ v1.2.0 index
CHANGELOG.md	✅ [1.2.0] section
docs/adr/README.md	✅ Platform ADR overview
CONTRIBUTING.md	✅ ADR range + install reference

Architecture consistency

Check	Result
Interface → Facade → Workflow diagram	✅ Consistent across README, ARCHITECTURE, CURRENT_STATUS
Discovery + Execution Planning in pipeline	✅
Canonical artifacts documented	✅ Implemented vs planned distinguished
GitHub Provider scope (no clone)	✅

README review

Check	Result
CLI as primary entry	✅
SDK examples	✅
app.py demoted to maintainer notes	✅
Version 1.2.0	✅
Roadmap link	✅

Getting Started review

Check	Result
init → doctor → reproduce flow	✅
No app.py as primary	✅
SDK section	✅

Roadmap review

Check	Result
v1.0–v1.2 completed	✅
v1.3–v1.5 planned	✅
Future interfaces marked not implemented	✅

Release Notes review

Check	Result
Highlights complete	✅
Migration guide	✅
Breaking changes documented	✅

Terminology consistency

Term	Consistent
Execution Planning (not "Execution Planner")	✅
Research Resource Discovery / Discovery	✅
ExecutionStrategy	✅
Platform Facade / Man1Lab	✅
ResearchResourceDiscovery	✅

Diagram consistency

Check	Result
No v1.1-only pipeline without Discovery	✅
Interface layer shown	✅

Outdated content removed

Item	Status
v1.1.0 as current version in README citation	✅ Fixed → 1.2.0
app.py as recommended quick start	✅ Removed from user paths
docs/README v1.1.0 header	✅ Updated
ARCHITECTURE "Repository discovery planned" in §6	✅ Updated — Discovery complete

Remaining documentation debt

Item	Notes
docs/architecture/CAPABILITIES.md	Per-agent detail; not fully refreshed in this freeze
docs/api/README.md	Partial refresh — facade as primary entry
ADR-0013/0014/0016 status still Draft	Promote to Accepted at release tag
DEVELOPMENT.md	Maintainer doc; partial interface table may need sync
Benchmark table dates	Historical M8.x runs unchanged
Future MCP/REST	Correctly marked not implemented

---

Verdict

...

Man1Lab v1.1.0

Man1Lab v1.1.0 — Foundation Release

Release date: 2026-06-30
Codename: Foundation
Repository: https://github.com/maniac1um/Man1Lab

---

Overview

v1.1.0 is not a feature-expansion release. It completes Platform Foundation — the infrastructure and architectural substrate on which future Man1Lab capabilities will be built.

v1.0.0 delivered an end-to-end MVP reproduction pipeline. v1.1.0 refactors and hardens that pipeline: canonical analysis artifact, external infrastructure adoption via ports and adapters, experiment tracking, and documentation governance.

Milestone: Foundation Complete
Next: Platform Capability (v1.2)

---

Highlights

• Analysis pipeline refactor — PaperReproductionAnalysis replaces PaperModel as the sole domain object (ADR-0009)
• Docling default parsing — structured Markdown via ports & adapters (ADR-0008)
• Hydra configuration layer — composed settings behind SettingsProvider (ADR-0010)
• Pixi developer environment — lockfile-driven repo environment (ADR-0011)
• MLflow experiment tracking — thin tracking via ExperimentTracker (ADR-0012)
• Documentation governance — public docs/ for decisions; local private/ for research process
• Infrastructure governance — infrastructure.md adoption matrix

---

Architecture Changes

Canonical domain object

Before (v1.0.0)	After (v1.1.0)
Flat PaperModel	Modular PaperReproductionAnalysis
WorkflowHistory.paper	WorkflowHistory.analysis
Reader → flat fields	Reader → goal / resources / method / evaluation / gaps

All downstream agents (Planner, Coder, Reviewer, Reporter) consume the analysis artifact through structured context views.

Layered pipeline

Paper (PDF)
    ↓ Parsing (Docling / PyMuPDF)     → ParsedDocument
    ↓ Analysis (Reader)              → PaperReproductionAnalysis
    ↓ Planning                         → TaskModel
    ↓ Implementation (Coder)           → Workspace
    ↓ Execution (Runner)               → ExecutionResult
    ↓ Verification → Review → Report
    ↓ Experiment Tracking (MLflow)     → parent run + nested stage runs

Infrastructure layers attach via Provider / Adapter — business agents do not import Docling, Hydra, Pixi, or MLflow.

---

Infrastructure Adoption

Capability	Tool	Status	ADR
Paper Parsing	Docling (+ PyMuPDF fallback)	Adopted	ADR-0008
Configuration	Hydra	Adopted	ADR-0010
Environment	Pixi	Adopted	ADR-0011
Experiment Tracking	MLflow	Adopted	ADR-0012
Dataset Versioning	DVC	Pending	—
Workflow Engine	TBD	Research	—

See infrastructure.md for the full adoption matrix.

---

Governance

Change	Detail
Public documentation	ADRs, architecture, API, guides, current status — in Git
Private documentation	Adoption reviews, audits, benchmarks, roadmaps — private/ (gitignored)
Decision rule	Technology reviews inform ADRs; ADRs are the durable public record
ADR index	ADR-0001 through ADR-0012

---

Breaking Changes

Change	Migration
Reader.run() returns PaperReproductionAnalysis	ADR-0009 supersedes ADR-0002 return type
WorkflowHistory.paper removed	Use WorkflowHistory.analysis
PaperModel removed from runtime pipeline	Legacy test files retained only
Composition root wires TrackedWorkflowOrchestrator	Tracking disable: TRACKING_BACKEND=noop

No change to TaskModel, Workspace, or orchestrator stage order.

---

Compatibility

Area	Status
Pixi (recommended)	pixi install + pixi run test
pip legacy	requirements.txt still supported
Parser fallback	PARSER_BACKEND=pymupdf
Tracking opt-out	TRACKING_ENABLED=false or TRACKING_BACKEND=noop
LLM providers	OpenAI-compatible + Anthropic unchanged

Tests: 172 passing (pixi run test)

Product limitations from v1.0.0 remain — see CURRENT_STATUS.md.

---

Native Platform Components (v1.1.0)

These remain Man1Lab-native — not replaced by infrastructure tools:

Component	Role
Analysis	Reader → PaperReproductionAnalysis
Planning	Planner → TaskModel
Coding	Coder → Workspace (+ GQ-1, RAG)
Review	Reviewer + PatchPlanner
Report	Reporter → ReportModel
Orchestration	WorkflowOrchestrator (topology frozen, ADR-0001)

---

Future Roadmap — Platform Capability (v1.2)

Foundation complete. Next milestone focuses on platform capabilities, not infrastructure:

Capability	Direction
Repository Discovery	Consume reproduction_gaps + metadata
Environment Generation	Beyond workspace venv bootstrap
Verification	Deeper alignment with analysis evaluation module
Failure Recovery	Close review loop — re-invoke Coder/Runner
DVC	Dataset versioning (pending adoption review)
Workflow Engine	Research only — Man1Lab orchestrator remains native

Working roadmap: private/roadmap/ (local).

---

Quick Start

git clone https://github.com/maniac1um/Man1Lab.git
cd Man1Lab
pixi install
pixi run test
pixi run run

Optional MLflow UI:

mlflow ui --backend-store-uri sqlite:///mlruns/mlflow.db

Full guide: GETTING_STARTED.md

---

Related Documents

Document	Purpose
CURRENT_STATUS.md	Live capabilities and benchmarks
ARCHITECTURE.md	Platform architecture
CHANGELOG.md	Version history
v1.0.0 release notes	MVP baseline

Man1Lab v1.0.0 (Research Prototype)

Man1Lab v1.0.0 (Research Prototype)

Release date: 2026-06-29
Repository: https://github.com/maniac1um/Man1Lab

---

Project Overview

Man1Lab is a research prototype that automates the engineering workflow for reproducing a machine learning paper from PDF. The system reads a paper, plans implementation tasks, generates a Python reproduction repository, installs dependencies, runs the training script, verifies the outcome, reviews failures with an LLM, and writes a final report.

This is the first public release of the MVP. It demonstrates a complete autonomous pipeline with real LLM integration. It does not guarantee successful end-to-end training on arbitrary papers.

---

Highlights

• Full pipeline — Reader through Reporter runs without manual intervention
• Real LLM integration — OpenAI-compatible API (tested with DeepSeek)
• Deterministic routing — TaskRouter maps engineering tasks to repository files
• Repository Acceptance Gate (RAG) — Rejects structurally defective repositories before execution
• 126 unit tests — Agents, services, coder quality, and acceptance gate covered
• Documented architecture — ADRs, capability reference, milestone review archive

---

Implemented Capabilities

Stage	Component	Output
Ingestion	Reader	PaperModel
Planning	Planner	TaskModel
Code generation	Coder	Workspace
Execution	Runner	ExecutionResult
Verification	VerificationService	VerificationResult
Review	Reviewer	ReviewReport
Patch planning	PatchPlanner	PatchPlan
Reporting	Reporter	ReportModel

The orchestrator schedules all stages. Agents do not communicate directly; they pass typed Pydantic artifacts.

---

Repository Acceptance Gate

The RAG is the final step inside Coder.run(). Before Runner executes, the generated repository must pass:

Check	What it catches
Import closure	Third-party imports missing from requirements.txt
Framework binding	Mixed or forbidden framework imports
Internal imports	Scripts importing symbols not in the interface registry
Training entrypoint	Missing or empty scripts/train.py

When rejected, Coder raises RepositoryAcceptanceError and Runner is not invoked. Warnings (config drift, README issues) do not block acceptance.

On the DeiT benchmark re-run, RAG accepted a repository with complete requirements.txt (torch, timm, torchvision, PyYAML, tqdm). Execution then failed at a timm runtime API issue — outside RAG scope.

Details: docs/reviews/repository_acceptance_gate/implementation_review.md

---

Benchmark Summary

Runs used scripts/run_integration_m7_1.py with DeepSeek API.

Benchmark	Paper	Result
M8.1	ResNet (1512.03385v1.pdf)	Pipeline complete; execution failed (No module named 'torch') — pre-GQ-1
M8.2	DeiT (2012.12877v2.pdf)	Pipeline complete; stub requirements.txt — pre-RAG
RAG	DeiT (re-run)	RAG ACCEPTED; execution failed (timm _pil_interp runtime error)

Takeaway: The pipeline reliably completes on real papers. Delivery defects (missing deps, broken imports) are caught at Coder. Successful training reproduction on benchmark papers remains an open goal.

Reports: M8.1 · M8.2

---

Known Limitations

1. No guaranteed reproduction — Training may fail due to runtime, paper-specific, or LLM generation issues
2. Review loop not closed — PatchPlan is produced but Coder/Runner are not re-invoked
3. Framework coverage — Binding profiles for PyTorch, TensorFlow, JAX, Caffe only
4. External API dependency — Reviewer can fail on LLM timeouts unrelated to code quality
5. Research prototype — Single maintainer; pull requests not accepted (issues welcome)

Full list: docs/CURRENT_STATUS.md

---

Future Work

Area	Direction
v1.1	Close review loop — re-run Coder/Runner on PatchPlan
v1.2	GitHub repository initialization for generated workspaces
v1.3	Multi-model LLM provider support
v2.0	Memory, human-in-the-loop, multi-agent collaboration

Roadmap: docs/roadmap/ROADMAP.md

---

Quick Start

git clone https://github.com/maniac1um/Man1Lab.git
cd Man1Lab
pip install -r requirements.txt
PYTHONPATH=. python -m pytest tests/ -v
PYTHONPATH=. python app.py

Set PAPER_PATH to a PDF. Copy .env.example to .env for real LLM calls.

Guide: docs/GETTING_STARTED.md

---

Documentation

Resource	Link
Current status	docs/CURRENT_STATUS.md
Architecture	docs/architecture/ARCHITECTURE.md
Changelog	CHANGELOG.md
Contributing	CONTRIBUTING.md

---

Citation

@software{man1lab_2026,
  author       = {maniac1um},
  title        = {Man1Lab: An Autonomous Research Paper Reproduction Pipeline},
  year         = {2026},
  version      = {1.0.0},
  url          = {https://github.com/maniac1um/Man1Lab}
}