Skip to content

BACKLOG_README.md

Latest commit

 

History

History
225 lines (173 loc) · 8.38 KB

BACKLOG_README.md

File metadata and controls

225 lines (173 loc) · 8.38 KB

📋 Entrainer Selection Framework - Development Backlog

Overview

This document provides a comprehensive overview of the development backlog for the Safety-by-Design Framework for Ethanol-Water Separation Entrainer Selection.

🎯 RECOMMENDED EXECUTION SEQUENCE

Primary Path (Follow Phase Order):

1. INF-001 to INF-008  (Environment Setup + Data Schemas)
        ↓
2. P1-xxx: Phase I (Domain Mapping)
        ↓
3. P2A-xxx: Phase II-A (Graph-RAG)
        ↓
4. P2B-xxx: Phase II-B (TRIZ Agents)
        ↓
5. P2C-xxx: Phase II-C (Cheminformatics)
        ↓
6. P3-xxx: Phase III (Graph Traversal)
        ↓
7. P4-xxx: Phase IV (Bayesian Optimization)
        ↓
8. P5-xxx: Phase V (DWSIM Simulation)  ←── DWSIM risk contained here

Risk Clarification (Updated from Consultation #8)

Risk Scope Impact on Phases 1-4 Mitigation
DWSIM COM fails Phase 5 ONLY ❌ None Plan-B fallbacks
Oracle slow (>1s) Phase 4 ❌ None on 1-3 Caching strategy
LLM hallucinations Phase 2A/2B ❌ Contained API verification

Key Insight: DWSIM is for VALIDATION only. Phases 1-4 produce a ranked candidate list independently. If DWSIM automation fails, manual validation or alternative simulators can be used.

Optional Risk Spikes (Can Be Deferred)

These were recommended by Consultation #8 but can be deferred to Phase 5:

Task Original Priority Updated Priority Rationale
INF-009: DWSIM Spike 🔴 CRITICAL 🟢 Deferred to P5 Risk contained to Phase 5
INF-010: Oracle Benchmark 🔴 Critical 🟡 Before Phase 4 Only needed before MOBO
INF-012: Simulation Watchdog 🟡 High 🟢 Deferred to P5 Only needed for DWSIM

🚨 Critical Fixes (Must Implement)

Based on the technical review in Implementation Plan/Feedback to Improve.md AND Consultations #7 and #8, the following critical fixes MUST be implemented:

1. Phase V Simulation Engine (CRITICAL)

  • Issue: FUG shortcut method assumes constant relative volatility
  • Impact: Invalid for extractive distillation where entrainer changes α along column
  • Fix: Implement DWSIM COM automation to solve full MESH equations
  • Status: ⬜ Not Started
  • Backlog: backlog/BACKLOG_05_Phase_5_Simulation.md

2. Tanimoto Threshold (HIGH)

  • Issue: 0.5 threshold returns functionally distinct molecules
  • Impact: Graph traversal returns noise instead of similar candidates
  • Fix: Tighten to 0.75-0.85 for similarity, keep 0.5 for scaffold hopping only
  • Status: ⬜ Not Started
  • Backlog: backlog/BACKLOG_02C_Phase_2C_Cheminformatics.md

3. Safety Data Verification (HIGH)

  • Issue: LLMs can hallucinate GHS categories
  • Impact: Safety barrier function fails if data is wrong
  • Fix: Query PubChem PUG REST API first, LLM as fallback only
  • Status: ⬜ Not Started
  • Backlog: backlog/BACKLOG_02A_Phase_2A_GraphRAG.md

4. Ternary Azeotrope Check (HIGH)

  • Issue: Solvent may form ternary azeotrope at finite concentrations
  • Impact: Good selectivity at infinite dilution but fails in practice
  • Fix: Add ternary azeotrope check in Phase IV Oracle, not just Phase V
  • Status: ⬜ Not Started
  • Backlog: backlog/BACKLOG_04_Phase_4_Optimization.md

📁 Backlog Structure

backlog/
├── BACKLOG_00_Infrastructure.md      # Core setup, config, databases
├── BACKLOG_01_Phase_1_Domain.md      # Phase I: Domain Mapping
├── BACKLOG_02A_Phase_2A_GraphRAG.md  # Phase II-A: Graph-RAG Engine
├── BACKLOG_02B_Phase_2B_TRIZ.md      # Phase II-B: TRIZ Multi-Agent
├── BACKLOG_02C_Phase_2C_Cheminformatics.md  # Phase II-C: Clustering
├── BACKLOG_03_Phase_3_Traversal.md   # Phase III: Graph Traversal
├── BACKLOG_04_Phase_4_Optimization.md # Phase IV: Bayesian Optimization
└── BACKLOG_05_Phase_5_Simulation.md  # Phase V: Process Simulation

🎯 Phase Summary

Phase Description Key Deliverables Critical Fixes
Infrastructure Core setup Config, DB connections, models -
Phase I Domain Mapping PubChem queries, SMARTS patterns -
Phase II-A Graph-RAG Neo4j graph, ChromaDB embeddings Safety verification
Phase II-B TRIZ Agents 6 specialized LLM agents -
Phase II-C Clustering RDKit fingerprints, diversity Tanimoto threshold
Phase III Traversal Graph expansion, similarity network Tanimoto threshold
Phase IV Optimization BoTorch MOBO, Pareto frontier Ternary azeotrope
Phase V Simulation DWSIM automation, validation DWSIM (not FUG)

🔄 Development Workflow

  1. Pick a task from the relevant backlog file
  2. Update status to "🔄 In Progress"
  3. Implement following the implementation notes
  4. Test using the specified test approach
  5. Update status to "✅ Complete" with date

📊 Progress Tracking

Overall Progress

  • Infrastructure: ⬜ 0%
  • Phase I: ⬜ 0%
  • Phase II-A: ⬜ 0%
  • Phase II-B: ⬜ 0%
  • Phase II-C: ⬜ 0%
  • Phase III: ⬜ 0%
  • Phase IV: ⬜ 0%
  • Phase V: ⬜ 0%

Sprint Goals

  • Sprint 1: Infrastructure + Phase I
  • Sprint 2: Phase II (A, B, C)
  • Sprint 3: Phase III + Phase IV
  • Sprint 4: Phase V + Integration Testing

🔗 Related Documentation

📝 Task Status Legend

  • ⬜ Not Started
  • 🔄 In Progress
  • ⏸️ Blocked
  • ✅ Complete
  • ❌ Cancelled

📋 Consultation #8 Recommendations Summary

Date: 2026-01-11 | Scorecard: Docs 9/10, Plan 6/10, Backlog 9/10

High Priority Items (Incorporated)

# Recommendation Location Status
1 DWSIM Feasibility Spike INF-009 ⬜ Added
2 Oracle Latency Benchmark INF-010 ⬜ Added
3 Data Contracts (Pydantic Schemas) INF-008 ⬜ Added
4 Name-to-SMILES Resolver INF-011, P2B-009 ⬜ Added
5 Simulation Watchdog INF-012 ⬜ Added
6 Solver Strategy Pattern P5-007 ⬜ Added
7 Reorder Execution Sequence This README ✅ Done

Medium Priority Items (Incorporated)

# Recommendation Location Status
1 Split settings.yaml config/infra_config.yaml, config/science_config.yaml ✅ Done
2 Configurable Safety Mode science_config.yaml → strict_safety_mode ✅ Done
3 SMILES Validation Gateway P2B-009 ⬜ Added
4 Move MCP Server .tools/advanced_consultation_mcp_server ✅ Done

Deferred Items

# Recommendation Reason
1 API Rate Limit Orchestrator Deferred per user request
2 Delete/simplify TRIZ agents Retained - user expertise

Key Insights from Consultation

  1. "Windows Trap": Docker + DWSIM COM = incompatible. Decision required.
  2. Pre-Mortem Warning: Likely failure modes are "Empty Set Optimization" (aggressive safety) or deployment failure.
  3. Backlog > Plan: The Backlog is more technically mature than the Implementation Plan.

⚠️ PENDING DECISIONS

The following decisions are required before significant coding begins:

1. Docker vs. Windows Deployment

  • Issue: DWSIM COM requires Windows, but proposal promises "Dockerized Virtual Lab"
  • Options: Abandon Docker / Use DWSIM CLI / REST Service / Windows Container
  • Status: ⏳ PENDING

2. Safety Mode Configuration

  • Issue: Strict safety may filter out global optimums early
  • Current: Configurable via strict_safety_mode in science_config.yaml
  • Status: ✅ RESOLVED (configurable)

📊 Updated Task Counts

Backlog Original Added Total
Infrastructure 7 5 12
Phase I - - -
Phase II-A - - -
Phase II-B 8 1 9
Phase II-C - - -
Phase III - - -
Phase IV - - -
Phase V 6 1 7