Joint Workflow and Prompt Optimization for User Behavior Simulation

Nipun B Nair; Tongtong Wu; Hongzhi Yin; Hui Li; Weiqing Wang

cs.MAOct 6, 2026

Joint Workflow and Prompt Optimization for User Behavior Simulation

Authors: Nipun B Nair, Tongtong Wu, Hongzhi Yin, Hui Li, Weiqing Wang

Organizations: Monash University · The University of Queensland · Xiamen University

Abstract

User behavior simulation is the computational modeling of user interactions within information systems through the use of simulated agents in place of live users. It supports system testing and evaluation, decision-making and forecasting, and user experience design. Existing simulators rely on hand-crafted rules or domain expertise that transfers poorly across tasks. SWORD (Simulation-driven Workflow and Prompt Optimization with Role-based Design) is introduced as a framework that jointly optimizes multi-agent workflow topology and natural-language prompts. It is guided solely by a scalar task metric, without domain initialization or task-specific engineering. The experimental results demonstrate that SWORD achieves statistically significant gains over prompt-only, workflow-only, and staged-optimization baselines under a controlled, identical-backbone comparison. Against the strongest published domain-specific baseline, SWORD further improves accuracy while using a smaller backbone model, substantially less training data, and a very reasonable API cost ($4--$6 for each dataset). Beyond predictive performance, SWORD autonomously discovers domain-relevant signals, review-sentiment mapping rules and epidemiological decay priors, purely from scalar error feedback, establishing textual gradients as a mechanism for unsupervised feature-importance discovery in user behavior modeling.

Figures & tables

Figure 1. Comparison of LLM based user simulation paradigms and the SWORD framework. (A) Agent-per-User : instantiates isolated LLM agents per user, preserving personalization but sharing no knowledge. (B) Code-Generation : achieves personalization via generating executable Python simulation code for each user through shared agent workflow, while shifting the burden of workflow design and simulation logic validation to domain experts, limiting task transferability. (C) SWORD (Ours) : is a further step forward compared with Code-Generation to eliminate the dependence on domain experts for different user simulation tasks via automated workflow and prompt optimization and task agnostic modular agent pool design. A three-panel diagram comparing paradigms of user behavior simulation. Panel A shows the Agent-per-User paradigm with isolated LLM agents. Panel B shows the Code-Generation paradigm where an LLM generates a shared Python execution function. Panel C shows the SWORD paradigm featuring a shared workflow DAG (Analyzer, Planner, Generator, Critic, Synthesizer) optimized via an interleaved co-evolution loop of workflow evolution and prompt optimization.

RQ	Addressed in	Evidence
RQ1	§ 6.1 , 6.2	Tables 2 , 3
RQ2	§ 6.5	Table 5
RQ3	§ 6.6	Figure 2 , Table 6
RQ4	§ 6.7	Table 8

Table 1. Research question to section mapping.

Category	Method	Agent MAE $\downarrow$	Mask RMSE $\downarrow$
Prior work	AI Scientist-v2 Yamada et al.,2025	$0.89_{\pm 0.10}$	$0.45_{\pm 0.07}$
	YuLan-OneSim Wang et al.,2025a	$0.72_{\pm 0.08}$	$0.42_{\pm 0.04}$
	G-SIM-ES Holt et al.,2025	$0.60_{\pm 0.07}$	$0.30_{\pm 0.08}$
	G-SIM-SBI Holt et al.,2025	$0.69_{\pm 0.08}$	${0.20}_{\pm 0.10}$
	Reflexion Shinn et al.,2023	${0.57}_{\pm 0.08}$	$0.36_{\pm 0.04}$
LLM-based	SOCIA- $\nabla$ (as reported) Hua et al.,2026	${0.54}_{\pm 0.06}$	${0.22}_{\pm 0.07}$

Table 2. Main results on Agent Society (MAE

\downarrow

) and Mask Adoption (RMSE

\downarrow

). Results for prior work are sourced from SOCIA-

\nabla

Hua et al., 2026 . Note: prior-work baselines use GPT-5-class models; SWORD uses GPT-5.4-mini, so this comparison is a cost-efficiency benchmark rather than a controlled accuracy head-to-head. Table 3 provides backbone-controlled ablation comparisons. Best result per column in bold . SWORD results are means over

n=10

seeds.

Method	Mask RMSE $\downarrow$	$\Delta$ [95% CI]	Agent MAE $\downarrow$	$\Delta$ [95% CI]
Workflow-Only (e.g., Zhang et al., 2025e )	$0.0686_{\pm 0.0722}$	$[-0.004,0.098]$	$0.456_{\pm 0.129}$	$[0.061,0.336]^{*}$
Prompt-Only (e.g., Chen et al., 2025c )	$0.0232_{\pm 0.0021}$	$[0.0002,0.0026]^{*}$	$0.301_{\pm 0.087}$	$[-0.040,0.128]$
Staged Opt. (e.g., Zhou et al., 2026 )	$0.1795_{\pm 0.1698}$	$[0.036,0.279]^{*}$	$0.510_{\pm 0.274}$	$[0.079,0.427]^{*}$
SWORD (Ours)	$\textbf{0.0217}_{\pm 0.0017}$	—	$\textbf{0.257}_{\pm 0.105}$	—

Table 3. Ablation study (

n{=}10

seeds,

N_{\text{train}}{=}30

N_{\text{test}}{=}100

). All methods share the identical GPT-5.4-mini backbone and identical seeds (paired design). We report mean

\pm

std and the 95% confidence interval of the paired difference vs. SWORD (

\Delta

, positive means worse than SWORD; undefined for SWORD itself). SWORD’s own 95% CIs of the mean are

[0.0206,0.0229]

RMSE and

[0.182,0.332]

MAE (Appendix C ). ∗ marks comparisons significant after per-benchmark Holm–Bonferroni correction (

\alpha{=}0.05

). Best per column in bold .

Task	Metric	SWORD	SOCIA- $\nabla$ (reproduced)
agent_society	MAE $\downarrow$	$0.2567\pm 0.1049$	$0.2275$ ( $n{=}1$ )
mask_adoption	RMSE $\downarrow$	$0.0217\pm 0.0017$	$0.5547$ ( $n{=}1$ )

Table 4. SWORD vs. SOCIA-

\nabla

(our backbone-matched re-execution) on both tasks. SWORD statistics are mean

\pm

std over

n=10

independent trials; the re-executed baseline reflects a single successfully executed, non-placeholder run (

n=1

) due to run-to-run execution instability in its automated code generation.

Seed	Task	Iter	Pre Workflow	Post Workflow	Improvement
1	Agent Society	T0→T1	p→a→s→c→g	p→a→g→t→e	+17.3%
1	Mask Adoption	T1→T2	a→p→g→c→v→s	a→p→g→v→s	+11.9%
2	Agent Society	T2→T3	a→p→g→t→c→v→s	a→p→g→t→v→s	+12.7%
2	Mask Adoption	T0→T1	s→a→g→t→e	a→p→g→c→v→s	+1.2%
3	Agent Society	T2→T3	a→g→s	a→p→g→v→s	+1.9%
3	Mask Adoption	T2→T3	a→p→g→c→v→s	a→p→g→c→t→v→s	+7.8%

Table 5. Rank Inversion Analysis: All 10 seeds (

n=10

) showing workflow transitions that drive performance improvements. Notation: abbreviated agent names (

a

=analyzer,

p

=planner,

g

=generator,

c

=critic,

v

=validator,

s

=synthesizer,

e

=executor,

t

=transformer) connected by

\to

showing execution order. Improvements are percentage gains in task score (Agent Society: MAE reduction; Mask Adoption: RMSE reduction).

Figure 2. Textual gradient content taxonomy across optimization iterations. Left : Mask Adoption. Role-reframing gradients dominate early and fade as workflow grows more complex, confirming the coarse-to-fine dynamic. Right : Agent Society. 72% of gradients are sentiment-related, autonomously discovered from MAE signals without task-specific initialization. Heat map showing gradient direction category distributions across iterations for both tasks.

Iter	Prompt excerpt
0	“You are a generator agent. Combine all upstream signals into ONE final prediction.”
1	“Priority: (1) historical_rates + recent_trend, (2) intervention status + decay, (3) population priors. Do not omit peer influence or habit dynamics.”
2	“Priority: (1) historical_rates + recent_trend, (2) intervention status + decay, (3) population priors. Do not omit peer influence or habit dynamics.”
3	“If historical_rates empty, apply low-baseline prior (0.08–0.10). In zero-history/no-intervention cases, do not infer high adoption. Intervention effects near-zero before day 10.”
4	“Decision hierarchy: 1) historical/recent trend 2) intervention (exponential decay from start) 3) conservative baseline 4) small population adjustments 5) calibrated params as weak diagnostics only. Clamp to [0, 1].”

Table 6. Generator prompt evolution across iterations on Mask Adoption. Epidemiological domain knowledge (decay, prior, timing) emerges from RMSE signals alone.

Task	$T=1$	$T=2$	$T=3$	$T=4$	$T=5$
Mask Adoption	$0.8037\pm 0.0095$	$0.8127\pm 0.0108$	$0.8217\pm 0.0121$	$0.8380\pm 0.0145$	$0.8450\pm 0.0167$
Agent Society	$0.8950\pm 0.0089$	$0.8952\pm 0.0091$	$0.8955\pm 0.0093$	$0.8958\pm 0.0095$	$0.8960\pm 0.0097$

Table 7. Best-so-far score across

n=10

seeds (mean

\pm

std per iteration). Low variance confirms robust, seed-independent convergence by

T=5

Task	$T=1$	$T=2$	$T=3$	$T=5$	$T=7$	$T=10$
Mask Adoption	0.8037	0.8127	0.8217	0.8450	0.8533	0.8565
Agent Society	0.8950	0.8952	0.8955	0.8960	0.8967	0.8967

Table 8. Extended convergence trajectory (single seed). Shows saturation persists beyond

T=5

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Source	Examples touched	Calls
Workflow scoring	All 30, all $K$ topologies	$5\!\times\!30\!\times\!4\!\times\!5=3{,}000$
Gradient	1 worst, elite only	$5\!\times\!3\!\times\!2\!\times\!5=\phantom{0,0}150$
Topology proposal	None	$5\!\times\!2=\phantom{3,0}10$
Total		$\mathbf{\approx 3{,}160}$

Appendix

\bar{A}\approx 5

(mean agents, range

[2,8]

);

F{=}1

(gradient runs every iteration); gradient targets single worst-scoring example.

Category	%	Iter	Example direction
Role reframing	34	0–1	“Shift from extraction to direct prediction”
Calibration	28	2–4	“Add low-baseline prior for zero-history cases”
Schema alignment	18	0–2	“Output single float, not multi-field JSON”
Signal priority	12	2–4	“historical_rates $>$ intervention $>$ network degree”
Intervention hdl	8	1–4	“Apply decay only if applied=true and day $\geq$ start”

Appendix

Table 9. Gradient direction taxonomy for Mask Adoption. The high proportion of role-reframing gradients (34%) at early iterations confirms that SWORD discovers qualitatively new specialisations, not merely incremental calibrations.

Category	%	Iter	Example direction
Sentiment prioritisation	41	0–1	“Weight review sentiment above numeric priors”
Sentiment mapping	31	1–4	“Strongly positive review: add $+$ 0.8–1.2 stars”
Conflict resolution	16	2–4	“If sentiment disagrees with rating, weight sentiment 2:1”
Calibration	12	3–4	“Platform-mean anchor adjustment: $\pm$ 0.1”

Appendix

Table 10. Gradient direction taxonomy for Agent Society. The dominance of sentiment-related gradients (72% combined) shows the textual gradient mechanism correctly identifies review text as the most informative feature.

Seed	SWORD	Staged	Prompt-Only	Workflow-Only
1	0.2445	0.4310	0.3249	0.2965
2	0.3122	0.5078	0.3404	0.3618
3	0.2502	0.7685	0.4649	0.4222
4	0.2000	0.1311	0.2944	0.6161
5	0.1870	0.7424	0.3214	0.4384
6	0.1611	0.2248	0.2562	0.7424

Appendix

Table 11. Per-seed Agent Society MAE

\downarrow

for all methods.

Seed	SWORD	Staged	Prompt-Only	Workflow-Only
1	0.0217	0.2977	0.0218	0.0902
2	0.0201	0.4194	0.0249	0.0254
3	0.0207	0.4194	0.0217	0.0430
4	0.0200	0.0227	0.0230	0.0220
5	0.0202	0.0678	0.0206	0.0394
6	0.0219	0.0200	0.0221	0.0269

Appendix

Table 12. Per-seed Mask Adoption RMSE

\downarrow

for all methods.

Explore similar work

UserProxyBench: Evaluating LLM User Simulators for Agent Benchmarks and Training
Sep 29, 2026Ashish Jain, Armaan SandhuUser Simulation Agentic Benchmarks
Sarvam AI
Prompt Optimization for User Simulation in Conversational Recommender Systems: A Multi-Objective Framework
May 8, 2026Nipun B Nair, Tongtong Wu, Weiqing WangUser Simulation Prompt Engineering
Monash University, Australia
MIMESIS: Learning User Simulators as Training Environments for Interactive Agents
Oct 7, 2026Hoang Phan, Dat Huynh, Andrey Zhmoginov +7User Simulation
Meta Superintelligence Labs · New York University · University of Wisconsin - Madison

Sep 29, 2026, cs.AIJ/K move · Enter open · S save

UserProxyBench: Evaluating LLM User Simulators for Agent Benchmarks and Training

Ashish Jain, Armaan Sandhu

Sarvam AI

User Simulation Agentic Benchmarks Desiderata Rubrics

Open paper AskPDF arXiv

Interactive agent benchmarks and multi-turn reinforcement learning increasingly place a second language model in the role of the user. This simulated user controls what information the agent receives and when, yet current benchmarks score only the agent and do not directly measure whether the user correctly executed its assigned role. We introduce UserProxyBench, an evaluation layer over the tau-bench family, and the User Fidelity Score (UFS), which measures adherence to the benchmark's private user instructions using task-grounded rubric criteria scored independently of agent success. Holding the agent fixed at GPT-5.5 and varying only the user proxy across 375 enterprise tasks changes mean task reward by 15.2 points, while 24.4% of successful episodes contain a user-specification violation. The dominant failure is premature disclosure: users provide information before it is requested. This behavior has little effect on task reward, yet among successful episodes it causes the agent to make 1.06 fewer tool calls on average, changing the interaction being evaluated while preserving the reward. Finally, across seven proxies we identify an empirical cost-fidelity frontier, enabling practitioners to select the least expensive simulator that satisfies a required fidelity level.

Abstract

Explore similar work

UserProxyBench: Evaluating LLM User Simulators for Agent Benchmarks and Training

Prompt Optimization for User Simulation in Conversational Recommender Systems: A Multi-Objective Framework

MIMESIS: Learning User Simulators as Training Environments for Interactive Agents