ReForge: Keeping ABR Algorithms Never Finished with Verified Large Language Model Edits

Zhiqiang He; Zhi Liu

cs.AIAug 15, 2026

ReForge: Keeping ABR Algorithms Never Finished with Verified Large Language Model Edits

Authors: Zhiqiang He, Zhi Liu

Organizations: The University of Electro-Communications Tokyo, Japan

Abstract

Designing an ABR algorithm for one network scenario takes an engineer months, and large language models now do this work in hours, matching or beating hand-built designs. But either way, the design fits only the world visible at its birth, and fails on the world that arrives after. We ask whether an ABR algorithm can keep pace with the world, redesigned in minutes as each scenario arrives, with every change proven harmless to every scenario already served. In this work, we propose ReForge, a continual heuristic learning framework that adapts to continuously changing scenarios. ReForge runs that routine with a large language model (LLM) in the loop. Each round the LLM reads where the current design falls short and proposes one small edit, and a replay over every network served so far decides. Specifically, what it edits is a single page of fuzzy rules that routes every decision to one of a frozen pool of pre-trained policies. The LLM writes the first page from measurements alone, then keeps improving it on its own. Each round it reads where the current rules fall short and proposes one small edit, and a replay over every network served so far decides whether the edit lands. We evaluate ReForge on nine real-world network families arriving one at a time as 3G, 4G, then 5G. A few edits per arrival lift mean QoE from 1.23 to 1.74, past the best single policy at 1.66 and to 94% of an oracle, and even repair families the loop never saw, one rising from 0.30 to 0.80. All code, data, and experiment records will be open-sourced upon cleanup.

Figures & tables

Figure 2. Trained in one world, deployed in many. (a) Training curves of the three learned policies;

\star

marks the frozen checkpoint. (b) QoE each policy gives up against that family’s best; zero, boxed, marks the best.

Figure 3. Why the LLM’s first router fails. (a) Each point is one method;

x

is mean QoE on the five in-pool families,

y

on the four unseen families (both on test traces); dashed lines are equal nine-family means. The router the LLM wrote (1.23) sits below fixed RobustMPC (1.66). (b, c) The cause; policy rankings flip across the boundary, the in-pool winner is the wrong choice out-of-pool, and at design time only the in-pool axis is visible.

Operator	Mechanical checks (excerpt)
add_rule	features, labels, and policy must exist
remove_rule	index must match the prompt’s numbering
edit_rule	same checks as add, applied to an index
retune_membership	new breakpoints must satisfy $a\leq b\leq c\leq d$
split_label	split point must lie in the band’s plateau;
	rules referencing the old label are copied

Table 1. The edit language. One edit per round; every edit is mechanically validated before any evaluation, and an invalid edit costs the round.

Figure 4. Final performance pooled over the nine test families. Left to right: the per-episode QoE CDF, three pairwise trade-offs with penalty axes reversed so upper right is better, and the QoE decomposition, where ours is the last bar of each group.

System	QoE $\uparrow$	Bitrate $\uparrow$	Rebuf. $\uparrow$	Smooth $\uparrow$
Classical controllers (outside the pool)
Rate-based	1.374	2.203	$-0.508$	$-0.321$
BBA	1.379	1.977	$-0.181$	$-0.417$
BOLA	0.531	2.075	$-0.681$	$-0.863$
Frozen pool policies, each deployed alone
FDASH	$-1.852$	2.302	$-4.050$	$-0.103$

Table 2. Nine-family means, QoE and its three components.

Figure 5. The design trajectory. Top: probe QoE, which moves only on accepted rounds and never descends. Bottom: the edit ledger, color by operator, filled when accepted. Right: per-family regret after each phase.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Signal	Label	$[a,b,c,d]$
throughput mean (kbps)	Low	$[0,0,900,1500]$
	Mid	$[900,1500,4800,6400]$
	High	$[4800,6400,10^{5},10^{5}]$
throughput variability	Steady	$[0,0,0.10,0.20]$
	Choppy	$[0.10,0.20,0.35,0.55]$
	Volatile	$[0.35,0.55,5,5]$

Appendix

Table 3. Membership bands of the Stage-1 router.

#	mean	variability	buffer	drift	$\to$ policy
0	High	Steady	Full	Holding	MPC
1	High	Steady	Full	Draining	PA-MoE
2	High	Steady	Healthy		PA-MoE
3	High	Steady	Critical		PA-MoE
4	High	Choppy			PA-MoE
5	High	Volatile			PA-MoE

Appendix

Table 4. The eight rules; a blank cell means the rule ignores that signal. Metadata: the fence allows MPC only when variability is at most 0.35, the buffer holds at least 10 s, and the drift is no worse than

-3

s per chunk; stickiness keeps the incumbent unless a challenger leads by 0.15 for 4 chunks; the warm-up and no-rule default is PA-MoE.

Continual Learning

Continual Learning studies how models acquire knowledge from changing data streams while retaining or selectively revising what they learned before.

See all 305 papers

Who Plans, Who Edits? Verified, Cost-Aware Division of Labor for LLM-Driven Analog Layout Repair
Oct 16, 2026Xilai Dai, Saeed MohammadiLLM Agents Electronic Design Automation
Elmore Family School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907, USA
Let AI Agents Translate Networks, Not Reason About Them
Jul 24, 2026Hongyu Hè, Maria Apostolaki
Princeton University
AlgoBench: Benchmarking Algorithmic Adaptation in Code Generation
Jun 30, 2026Xinyuan Song, Zekun Cai, Liang ZhaoSynthetic Benchmark Generation Code Generation Evaluation
Emory University, Atlanta, GA, USA · The University of Tokyo, Tokyo, Japan · LocationMind, Tokyo, Japan

Oct 16, 2026J/K move · Enter open · S save

Who Plans, Who Edits? Verified, Cost-Aware Division of Labor for LLM-Driven Analog Layout Repair

Xilai Dai, Saeed Mohammadi

Elmore Family School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907, USA

LLM Agents Electronic Design Automation LLM Agent Workflow Optimization LLM Agent Efficiency

Open paper AskPDF arXiv

Language-model agents can repair analog layouts when sign-off tools check every edit, but which model should plan a repair, which should edit it, and at what cost remain open questions. In our tool-grounded repair agent, models only choose bounded layout operators; sign-off tools rerun each task's contract checks after every edit, so failing layouts are never accepted. On layout-versus-schematic (LVS), design-rule checking (DRC), and post-layout repairs of three designs (a production high-voltage digital-to-analog converter (DAC), a 5-V oscillator, and a SKY130 operational transconductance amplifier (OTA)) with six tier-matched models from two vendors, 182 blind runs show that cheap models execute plans but rarely succeed alone (the cheapest model: 1 of 11 runs alone, 16 of 17 under a flagship planner; p < 0.001), cheap planners fail more often, and no fixed assignment is cheapest on every task. We propose a learning controller that selects models for each role and the escalation order from the first tool evaluation and updates Bayesian estimates after every attempt. With rules and initial estimates frozen before testing, the controller fixed 7 of 8 episodes on four sealed held-out tasks, matching the OpenAI flagship alone, at 15-40% of the flagship's spend per fix on three tasks but 8% higher spend overall. The cheapest mid-tier model alone fixed 6 of 8 at the lowest overall spend per fix but missed the hardest task. All 32 extracted DAC repairs match the human-guided LVS repair's integral nonlinearity (INL) within 0.005 least significant bits (LSB) at every code.

Abstract

Continual Learning

Explore similar work

Who Plans, Who Edits? Verified, Cost-Aware Division of Labor for LLM-Driven Analog Layout Repair

Let AI Agents Translate Networks, Not Reason About Them

AlgoBench: Benchmarking Algorithmic Adaptation in Code Generation