The Thousand-Graph Hypothesis: A Testable Hypothesis of Task-Conditioned Relation Materialization in Repository-Level Code Reasoning

Fei Ding

cs.SEAug 27, 2026

The Thousand-Graph Hypothesis: A Testable Hypothesis of Task-Conditioned Relation Materialization in Repository-Level Code Reasoning

Authors: Fei Ding

Organizations: Alibaba Group

Abstract

Large software repositories are often beyond model context limits. Training repository knowledge into models is costly and quickly stale, while local retrieval can miss scattered requirements, and explicit relation graphs add ongoing maintenance burden. We propose an entity-only external interface with task-conditioned relation materialization during inference. A two-layer index separates global routing from local entity focus and is evaluated on DeepSeek-V4-Flash and SWE-bench Verified. The base, one-layer, and two-layer conditions achieve 92.1%, 94.2%, and 95.6% success, respectively, under zero pre-built entity-relation edges.

Figures & tables

Panel of Figure 1 — Figure 1: Task-conditioned attention interpretation of The Thousand-Graph Hypothesis. The shared entity base and different task contexts induce different relation structures; $\Phi$ aggregates cross-token, layer, and head interactions for each task.

Panel of Figure 2 — Figure 2: Task telescope for two-layer entity interface. Task $q$ narrows repository entities through two lenses to context $C_{q}$ without persistent relation edges.

Dimension	Explicit graph	This work
External state	Entities, edges, schema, graph database	Two-layer entity index
Where relations come from	External extraction or inference over source graph	Hypothesis: task-conditioned attention
Task adaptation	Fixed graph with local subgraph query	Task-specific transient organization in inference
Repository updates	Recompute entities and edges with consistency	Update entities and indexes only

Table 1: Boundary comparison between explicit external graphs and this approach

Condition	Pre-built relation edges	Success rate (%)
Base system	none	92.1
One-layer index	none	94.2
Two-layer index	none	95.6

Table 2: DeepSeek-V4-Flash on SWE-bench Verified.

Evidence level	Records
Author-reported overall	Half-year-plus usage; 200+ projects; several GitHub fixes
Auditable local snapshot	6 repositories, 52 artifacts
Public GitHub subset	1 repository; 29 merged PRs; 27 explicit-fix PRs
Index evolution in repair	19 index-sync commits; 11 close index+fix sequences

Table 3: Reported scale vs. auditable subset

LLM Reasoning with Graphs

Large language model reasoning with graphs uses structured relationships to support multi-step inference, evidence retrieval, and planning.

See all 211 papers

DeepDiscovery: A Location-Inference Framework for Task-Level Repository Understanding
Jun 22, 2026Jiawei He, Weisong Sun, Mengyu Shi +4LLM Grounding Software Engineering Agents
AMAP, Alibaba Group · Nanyang Technological University · Nanjing University +1
LARGER: Lexically Anchored Repository Graph Exploration and Retrieval
May 8, 2026Yuntong Hu, Tongli Su, Liang Zhao +2AI Coding Agents Agentic Retrieval
Emory University
DeepRepoQA: Code Repository Question Answering with Deep Agent Exploration
Aug 25, 2026Weihan Peng, Yuling Shi, Yingwei Ma +3Question Answering Software Engineering Agents
Shanghai Jiao Tong University · The Hong Kong University of Science and Technology · University of California San Diego

Jun 22, 2026J/K move · Enter open · S save

DeepDiscovery: A Location-Inference Framework for Task-Level Repository Understanding

Jiawei He, Weisong Sun, Mengyu Shi, Jie Jia+3

AMAP, Alibaba Group · Nanyang Technological University · Nanjing University+1

LLM Grounding Software Engineering Agents

Open paper AskPDF arXiv

Large language models have shown strong performance on software engineering (SE) tasks, yet understanding large industrial repositories remains challenging. Existing methods often retrieve only local fragments and fail to recover the broader task-relevant context needed for complex repository-level tasks. We present DeepDiscovery, a task-level repository-understanding method for large industrial codebases. DeepDiscovery uses a two-stage Location--Inference framework to localize high-confidence task anchors and recover broader task-relevant context over multi-relational repository structure under budget constraints. Across controlled method-level evaluation, organization-internal industrial repository-understanding scenarios, and end-to-end evaluation on SWE-bench Verified, DeepDiscovery consistently improves task-relevant file recovery and downstream SE performance. On 27 medium-scale tasks, DeepDiscovery achieves the best file recovery quality among five representative baselines without offline preprocessing. On organization-internal industrial tasks from a production-scale integrated codebase ecosystem, including 27 medium-scale tasks and 40 large-scale tasks, DeepDiscovery improves Full Recall Rate across multiple AI coding systems, with absolute gains ranging from 1.6 to 9.2 percentage points on large subprojects and from 2.5 to 7.4 percentage points on medium-scale subprojects. In a controlled end-to-end evaluation on SWE-bench Verified, a system equipped with DeepDiscovery achieves a 78.6% Solve Rate, outperforming the corresponding baseline by 8.2 percentage points. These results suggest that stronger task-level repository understanding can improve coding-agent performance on complex SE tasks.

Abstract

LLM Reasoning with Graphs

Explore similar work

DeepDiscovery: A Location-Inference Framework for Task-Level Repository Understanding

LARGER: Lexically Anchored Repository Graph Exploration and Retrieval

DeepRepoQA: Code Repository Question Answering with Deep Agent Exploration