Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Contract2Plan: Verified Contract-Grounded Retrieval-Augmented Optimization for BOM-Aware Procurement and Multi-Echelon Inventory Planning

Sahil Agarwal

TL;DR

The paper tackles the risk that LLM-based extraction of contract terms into planning inputs can produce infeasible or non-compliant procurement plans due to BOM coupling and heterogeneous document sources. Contract2Plan offers a verified GenAI-to-optimizer pipeline that inserts a solver-based compliance gate, grounding extracted constraints with provenance, compiling them into a BOM-aware MILP, and verifying feasibility with targeted repairs or abstention when automation is unsafe. It formalizes a taxonomy of constraint classes with conservative repair guarantees and explicit abstention rules, accompanied by auditable decision cards linking decisions to evidence spans. A synthetic microbenchmark demonstrates heavy-tailed cost and compliance risk from extraction errors, motivating verification as a first-class component of contract-grounded planning systems.

Abstract

Procurement and inventory planning is governed not only by demand forecasts and bills of materials (BOMs), but also by operational terms in contracts and supplier documents (e.g., MOQs, lead times, price tiers, allocation caps, substitution approvals). LLM-based extraction can speed up structuring these terms, but extraction-only or LLM-only decision pipelines are brittle: missed clauses, unit errors, and unresolved conflicts can yield infeasible plans or silent contract violations, amplified by BOM coupling. We introduce Contract2Plan, a verified GenAI-to-optimizer pipeline that inserts a solver-based compliance gate before plans are emitted. The system retrieves clause evidence with provenance, extracts a typed constraint schema with evidence spans, compiles constraints into a BOM-aware MILP, and verifies grounding, eligibility, consistency, and feasibility using solver diagnostics, triggering targeted repair or abstention when automation is unsafe. We formalize which clause classes admit conservative repair with contract-safe feasibility guarantees and which require human confirmation. A self-contained synthetic micro-benchmark (500 instances; T=5) computed by exact enumeration under an execution model with MOQ uplift and emergency purchases shows heavy-tailed regret and nontrivial MOQ-violation incidence for extraction-only planning, motivating verification as a first-class component of contract-grounded planning systems.

Contract2Plan: Verified Contract-Grounded Retrieval-Augmented Optimization for BOM-Aware Procurement and Multi-Echelon Inventory Planning

TL;DR

The paper tackles the risk that LLM-based extraction of contract terms into planning inputs can produce infeasible or non-compliant procurement plans due to BOM coupling and heterogeneous document sources. Contract2Plan offers a verified GenAI-to-optimizer pipeline that inserts a solver-based compliance gate, grounding extracted constraints with provenance, compiling them into a BOM-aware MILP, and verifying feasibility with targeted repairs or abstention when automation is unsafe. It formalizes a taxonomy of constraint classes with conservative repair guarantees and explicit abstention rules, accompanied by auditable decision cards linking decisions to evidence spans. A synthetic microbenchmark demonstrates heavy-tailed cost and compliance risk from extraction errors, motivating verification as a first-class component of contract-grounded planning systems.

Abstract

Procurement and inventory planning is governed not only by demand forecasts and bills of materials (BOMs), but also by operational terms in contracts and supplier documents (e.g., MOQs, lead times, price tiers, allocation caps, substitution approvals). LLM-based extraction can speed up structuring these terms, but extraction-only or LLM-only decision pipelines are brittle: missed clauses, unit errors, and unresolved conflicts can yield infeasible plans or silent contract violations, amplified by BOM coupling. We introduce Contract2Plan, a verified GenAI-to-optimizer pipeline that inserts a solver-based compliance gate before plans are emitted. The system retrieves clause evidence with provenance, extracts a typed constraint schema with evidence spans, compiles constraints into a BOM-aware MILP, and verifies grounding, eligibility, consistency, and feasibility using solver diagnostics, triggering targeted repair or abstention when automation is unsafe. We formalize which clause classes admit conservative repair with contract-safe feasibility guarantees and which require human confirmation. A self-contained synthetic micro-benchmark (500 instances; T=5) computed by exact enumeration under an execution model with MOQ uplift and emergency purchases shows heavy-tailed regret and nontrivial MOQ-violation incidence for extraction-only planning, motivating verification as a first-class component of contract-grounded planning systems.
Paper Structure (43 sections, 2 theorems, 7 equations, 5 figures, 4 tables, 1 algorithm)

This paper contains 43 sections, 2 theorems, 7 equations, 5 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Illustrative failure vignette (synthetic but realistic).
  3. Related Work and Positioning
  4. Background and Problem Setting
  5. BOM-aware planning
  6. Multi-echelon inventory planning
  7. Contractual procurement constraints
  8. Task Definition: Contract-Grounded Planning
  9. Inputs
  10. Outputs
  11. Evaluation axes
  12. Method: Contract2Plan
  13. System architecture
  14. Constraint schema with provenance
  15. Document ingestion and table handling
  16. ...and 28 more sections

Key Result

Theorem 1

Fix a planning instance (BOM, network, demand). For each Class A constraint type $\tau$, let $\mathcal{V}_\tau$ be the retrieved candidate values applicable to the same scope and effective window, and let $\widehat{c}_\tau=\mathrm{merge}_\tau(\mathcal{V}_\tau)$. Let $\widehat{\mathcal{C}}$ denote th Then any plan feasible under $\widehat{\mathcal{C}}$ is feasible under the true constraints $\mathc

Figures (5)

  • Figure 1: Contract2Plan pipeline. Constraints are not trusted until they pass grounding, consistency, eligibility, and solver-based feasibility checks. Failures trigger targeted repair or explicit human confirmation.
  • Figure 2: Verifier and repair logic. Contract2Plan loops on targeted evidence collection and constrained re-extraction. It escalates to humans when automation is unsafe.
  • Figure 3: Synthetic contract excerpt used for the walkthrough (written to resemble typical addendum structure).
  • Figure 4: Example schema-constrained extraction output with evidence pointers to the excerpt lines.
  • Figure 5: Example BOM with an alternate component. Substitution is compliance-sensitive and is enforced as a constraint.

Theorems & Definitions (4)

  • Definition 1: Restrictiveness order
  • Theorem 1: Conservative feasibility implies contract-safe feasibility for Class A
  • proof : Proof sketch
  • Proposition 1: Tier eligibility prevents ineligible discount claims