arguments-jiachen

name	arguments-jiachen
description	VA argument index for Jiachen (Amber) Liu — searchable verifiable arguments across all research domains. Each VA follows the 7-field claim schema with citekeys.
version	1

Verifiable Arguments — Jiachen (Amber) Liu

"No fabricated anecdotes. If you cannot trace a claim to a source, mark it [unverified]. If you are unsure, do not include it."

---

Search Guide

Arguments are organized by domain. Use grep or browse the domain directories.

Domains:

• llm-serving/ — LLM serving systems, QoE, energy
• federated-learning/ — FL benchmarking, optimization, resource management
• ai-agents/ — AI agents for science, Curie, EXP-Bench
• evaluation/ — Benchmarking philosophy, evaluation frameworks

To find arguments by topic:

grep -rn 'Claim.*keyword' arguments/**/*.md

To count total VA entries:

grep -rc '\*\*Claim ID\*\*' arguments/**/*.md

---

Argument Index

LLM Serving Systems

ID	Claim	Domain	File
VA001	QoE metrics are more important than raw throughput for conversational LLM serving	llm-serving	llm-serving/andes-qoe.md
VA002	Text streaming services require a fundamentally different serving framework than batch inference	llm-serving	llm-serving/andes-qoe.md
VA003	Energy consumption should be a first-class metric in LLM serving optimization	llm-serving	llm-serving/ml-energy.md

Federated Learning

ID	Claim	Domain	File
VA004	FL must be benchmarked at scale with realistic device heterogeneity	federated-learning	federated-learning/fedscale.md
VA005	Scalable client clustering significantly improves FL training efficiency	federated-learning	federated-learning/auxo.md
VA006	Model transformation enables efficient FL across heterogeneous clients	federated-learning	federated-learning/fedtrans.md
VA007	Multi-job FL resource management is an under-studied systems problem	federated-learning	federated-learning/venn.md

AI Agents for Science

ID	Claim	Domain	File
VA008	AI agents can automate rigorous scientific experimentation end-to-end	ai-agents	ai-agents/curie.md
VA009	AI research capability should be evaluated with end-to-end experiment benchmarks	ai-agents	ai-agents/exp-bench.md
VA010	Domain-specific engineering skills are the bottleneck for AI research agents	ai-agents	ai-agents/ai-research-skills.md

Evaluation & Infrastructure

ID	Claim	Domain	File
VA011	Infrastructure-as-Code needs its own code generation benchmark	evaluation	evaluation/iac-eval.md
VA012	Building evaluation infrastructure is itself a research contribution	evaluation	evaluation/meta-pattern.md
VA013	The trend in ML systems is predictable by analyzing application trajectories	evaluation	evaluation/predict-position.md

---

Total VA Count

Run: grep -rc '\*\*Claim ID\*\*' arguments/**/*.md

Target range: 15–25 VA entries for initial persona.