Skip to content

Add manuscript review system and port simulation dashboards#236

Open
chenhaozhou1996 wants to merge 9 commits intostreamlit:masterfrom
chenhaozhou1996:claude/manuscript-review-coordinator-wz0mC
Open

Add manuscript review system and port simulation dashboards#236
chenhaozhou1996 wants to merge 9 commits intostreamlit:masterfrom
chenhaozhou1996:claude/manuscript-review-coordinator-wz0mC

Conversation

@chenhaozhou1996
Copy link

@chenhaozhou1996 chenhaozhou1996 commented Mar 1, 2026

Summary

This PR introduces two major new applications to the codebase:

  1. Manuscript Review System (manuscript_review_app.py): A distributed multi-agent system for academic manuscript peer review coordination
  2. Port Operations Simulation (streamlit_app.py): A discrete-event simulation dashboard for container port operations with 3D/2D visualization

Key Changes

Manuscript Review System (New File)

  • Distributed Storage Coordinator: Implements a multi-partition storage system (manuscripts, reviews, messages, agent states, coordination metadata) with distributed locking for state synchronization across agents
  • Multi-Agent Architecture:
    • EIC Agent: Coordinates the review workflow, assigns reviewers, synthesizes feedback, and makes final publication decisions
    • Reviewer Agents: Evaluate manuscripts across multiple dimensions (novelty, methodology, writing quality, significance, data analysis) with personality-based scoring
    • Author Agent: Processes feedback, generates point-by-point responses, and revises manuscripts
  • Review Workflow: Implements iterative review cycles with configurable maximum rounds, supporting accept/minor revision/major revision/reject decisions
  • Knowledge Base: Includes reviewer profiles, review templates, revision action templates, and dimension-based evaluation criteria
  • Message System: Asynchronous message passing between agents with full audit logging

Port Operations Simulation (Refactored)

  • Complete rewrite from placeholder template to full discrete-event simulation
  • SimPy-based Engine: Implements ship arrival generation, berth allocation, crane operations, and container handling
  • Data Models: Defines Ship, Berth, Crane, Container, and SimulationStats dataclasses for state management
  • 3D Visualization: Creates interactive 3D port layout with water surface, quay walls, berths, cranes, ships, and container yard using Plotly
  • Professional Styling: Custom CSS for dark-themed dashboard with metric cards, status indicators, and responsive layout
  • Real-time Analytics: Tracks utilization metrics, throughput, queue lengths, and event logs
  • Configuration System: Parameterized simulation with configurable berths, cranes, arrival rates, and crane speeds

Dependencies

  • Added plotly>=5.18.0 for 3D/2D visualization
  • Added simpy>=4.1.1 for discrete-event simulation
  • Added numpy>=1.24.0 for numerical operations
  • Updated streamlit, pandas, and altair to recent stable versions

Notable Implementation Details

  • Distributed Coordination: The manuscript review system uses Streamlit session state as a distributed store with lock mechanisms to coordinate state changes across multiple agents
  • Stochastic Simulation: Port simulation uses exponential distributions for ship arrivals and random service times for realistic port behavior
  • Iterative Review Cycles: Manuscript system supports multiple review rounds with automatic score improvements when authors address feedback
  • Event Logging: Both systems maintain comprehensive audit trails of all state changes and agent interactions

https://claude.ai/code/session_01RevzPW3GHZfXrsm6AufG36

claude and others added 5 commits December 12, 2025 12:19
@claude
Add professional port operations simulation with 3D/2D visualization
6fdf7e7 
- Implement discrete-event simulation using SimPy for port operations
- Add 3D port visualization with Plotly (ships, cranes, containers, water)
- Create 2D top-down view for operational overview
- Design analytics dashboard with KPIs (ships processed, containers moved)
- Include berth utilization charts, queue gauges, and throughput metrics
- Add professional dark theme CSS styling
- Support configurable parameters (berths, cranes, arrival rates)
- Track simulation events and ship statistics in data tables
@chenhaozhou1996
Merge pull request #1 from chenhaozhou1996/claude/anylogic-port-simul…
83d19b2 
…ation-01B7D51cciFTrGpRaghDNXcv
@claude
Add distributed manuscript review coordination agent system
cab0877 
Implement multi-agent review coordination model with:
- EIC (Editor-in-Chief) agent: orchestrates review process, synthesizes
  feedback, assigns reviewers, makes accept/reject decisions
- Reviewer agents (3 profiles): evaluate manuscripts across 5 dimensions
  (novelty, methodology, writing, significance, data analysis)
- Author agent: processes feedback, generates point-by-point responses,
  revises manuscript

Key architecture:
- DistributedStore: distributed state coordination with partitioned storage
  for manuscripts, reviews, messages, and agent states
- ReviewCoordinator: manages multi-round review cycles until convergence
- Streamlit UI: interactive dashboard with workflow visualization, score
  trends, message flow, and event logging

The system runs iterative rounds: EIC assigns reviewers -> reviewers
evaluate -> EIC synthesizes feedback -> author revises -> repeat until
reviewers are satisfied or max rounds reached.

https://claude.ai/code/session_01RevzPW3GHZfXrsm6AufG36
@claude
Enhance review coordination with consensus analysis, comment tracking…
fc53ac6 
…, and smart loop termination

Major improvements to the distributed manuscript review system:

1. EIC Decision Intelligence:
   - Expertise-weighted scoring (1.5x weight for expert dimensions)
   - Reviewer disagreement detection (variance > 2.0 flags)
   - Outlier reviewer identification
   - Score trajectory analysis with convergence prediction
   - Consensus level metric (0-100%)

2. Reviewer Re-Review Verification:
   - Check actual author response against previous comments
   - Resolution tracking: ADDRESSED / PARTIALLY_ADDRESSED / NOT_ADDRESSED
   - Score-comment consistency enforcement (low score must have matching severity)
   - Expertise-based confidence calculation

3. Author Strategy:
   - Conflict detection between reviewers on same dimension
   - Rebuttal generation for conflicting feedback
   - Priority-ordered responses (critical > major > minor)
   - Section-specific revisions based on feedback categories

4. Decision Loop Fix:
   - is_process_complete() now enforces max_rounds
   - Minor revision after revision cycle with no critical issues → accept
   - Diminishing returns detection (< 0.5 improvement for 2 rounds)
   - Formal make_final_decision() integrated into UI flow

5. New UI Components:
   - Comment resolution tracker dashboard with progress bar
   - Version diff viewer (section-level comparison)
   - Convergence chart with accept/reject threshold lines
   - Reviewer disagreement heatmap
   - Fixed workflow diagram bug (ValueError on missing phase)

6. Bug Fixes:
   - Fixed line 552: ManuscriptState.get() crash (dataclass not dict)
   - Fixed workflow diagram: safe phase ordering without ValueError

https://claude.ai/code/session_01RevzPW3GHZfXrsm6AufG36
@claude
Add .gitignore for Python cache files
d7d0ab8 
https://claude.ai/code/session_01RevzPW3GHZfXrsm6AufG36
chenhaozhou1996
chenhaozhou1996 commented Mar 1, 2026
View reviewed changes
Copy link
Author

@chenhaozhou1996 chenhaozhou1996 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a

claude added 4 commits March 1, 2026 22:21
@claude
Add download/export feature for review documents (TXT, JSON, CSV)
971d532 
Users can now download complete review history in three formats:
- TXT: Full review report with all rounds, comments, and decisions
- JSON: Complete structured data for programmatic analysis
- CSV: Score summary table (per reviewer, per round, per dimension)

Download panel appears in the completion phase after review concludes.

https://claude.ai/code/session_01RevzPW3GHZfXrsm6AufG36
@claude
Add Chinese usage guide for running the review system locally
408a726 
Step-by-step instructions for Mac users covering git clone,
pip install, streamlit run, and UI walkthrough with FAQ.

https://claude.ai/code/session_01RevzPW3GHZfXrsm6AufG36
@claude
Add MCP plugin and Agent SDK multi-agent review system
bf634af 
Two new entry points for the review coordination system:

1. mcp_review_server.py - MCP Server Plugin
   - 8 tools: init_review, run_review, author_revise, next_round,
     get_status, list_sessions, export_report, auto_run
   - Simulated review logic (no API cost)
   - Configurable via Claude Code MCP settings

2. review_agent.py - Claude Agent SDK version
   - Real AI-powered multi-agent system using Claude API
   - EIC, Reviewer (x3), Author agents with distinct prompts
   - Multi-round orchestration with convergence detection
   - CLI interface with --title, --abstract, --export flags
   - Also usable as Python module

Plus usage guide (README_插件使用指南.md) with setup instructions.

https://claude.ai/code/session_01RevzPW3GHZfXrsm6AufG36
@claude
Integrate Claude API into Streamlit app for real AI-powered review
2b07715 
Adds dual-mode support to the web interface:

- Simulation mode (default): Template-based review (no API needed)
- AI mode: Real Claude-powered review via Anthropic API

New features:
- AIReviewEngine class with ai_review(), ai_synthesize(), ai_author_respond()
- Sidebar toggle for AI mode with API key input and model selector
- ReviewCoordinator now accepts optional ai_engine parameter
- AI reasoning and key issues displayed in EIC synthesis tab
- Mode badge shown during review execution

Usage: Toggle "AI 审稿模式" in sidebar, enter API key, select model
(Haiku for speed/cost, Sonnet for quality)

https://claude.ai/code/session_01RevzPW3GHZfXrsm6AufG36
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@chenhaozhou1996 @claude