arxiv: 2605.02178 · v1 · submitted 2026-05-04 · 💻 cs.AI

Recognition: 2 theorem links

· Lean Theorem

T²PO: Uncertainty-Guided Exploration Control for Stable Multi-Turn Agentic Reinforcement Learning

Haixin Wang , Hejie Cui , Chenwei Zhang , Xin Liu , Shuowei Jin , Shijie Geng , Xinyang Zhang , Nasser Zalmout

show 2 more authors

Zhenyu Shi Yizhou Sun

Authors on Pith no claims yet

Pith reviewed 2026-05-08 19:34 UTC · model grok-4.3

classification 💻 cs.AI

keywords explorationmulti-turnprogressuncertaintyfine-grainedinstabilitylearninglevel

0 comments

The pith

T²PO improves stability and performance in multi-turn agentic RL by using uncertainty dynamics at token and turn levels to guide exploration and avoid wasted rollouts.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Multi-turn reinforcement learning lets AI agents interact with environments over many steps, like shopping online or answering questions by searching. But training often collapses because the agent keeps picking actions that give little new information. T²PO watches how the agent's uncertainty about its next action changes with each token it generates. When uncertainty stops dropping much, it forces the agent to think longer before acting. At the level of whole turns, it detects interactions that made almost no progress and reruns them instead of continuing down a dead end. The approach was tested on WebShop, ALFWorld, and Search QA tasks. The abstract reports more stable training curves and higher final performance compared with prior stabilization techniques.

Core claim

We evaluate T²PO in diverse environments, including WebShop, ALFWorld, and Search QA, demonstrating substantial gains in training stability and performance improvements with better exploration efficiency.

Load-bearing premise

That marginal uncertainty change at the token level and negligible exploration progress at the turn level can be measured reliably enough to trigger interventions without introducing new sources of instability or bias in the policy updates.

Figures

Figures reproduced from arXiv: 2605.02178 by Chenwei Zhang, Haixin Wang, Hejie Cui, Nasser Zalmout, Shijie Geng, Shuowei Jin, Xin Liu, Xinyang Zhang, Yizhou Sun, Zhenyu Shi.

**Figure 1.** Figure 1: Training instability of SOTA baselines under different environment initialization random seeds. We can observe that success rate drops while internal signals like KL divergence and gradient norms explode (shown in orange background). deviate from the successful action space at an early stage, yet continue executing numerous repetitive and unproductive turns, leaving little chance of recovery within a limi… view at source ↗

**Figure 2.** Figure 2: Overview of the proposed Uncertainty-Guided Exploration Control at both token and turn levels. acts with the environment to obtain an observation represented as the state s k ∈ S, where S denotes the environment-defined state space. Based on this state, the agent generates an action a k ∈ Vn, V n is the action space formed by the LLM tokenizer vocabulary V. Typically, base LLMs fine-tuned with chain-of-th… view at source ↗

**Figure 3.** Figure 3: Contour of Ht fails to discriminate highly uncertain distributions near uniformity, while Ct ignores variations in tail probabilities. The proposed signal Mt integrates both measures, producing non-degenerate contour geometry that distinguishes distributions sharing identical top-k probability but differing residual mass. Self-calibrated uncertainty signal. Based on the above analysis, Ct and Ht are compl… view at source ↗

**Figure 4.** Figure 4: (a) Uncertainty dynamics of self-calibrated signal Mt on response length. (b) Word cloud of tokens with the highest uncertainty. (c) Colormap of the uncertainty signal aggregated by the sliding window. When the signal falls below ϵ (corresponding to the brightest token ‘Then’), thinking cutoff is triggered. The contour lines of Mt are no longer piecewise-linear and degenerate under the max operator compare… view at source ↗

**Figure 5.** Figure 5: We evaluate both task performance and exploration efficiency. (a) shows that T2 PO enables performance to steadily improve without collapse on three different env seeds. In (b), the bar chart shows that the distribution of token consumption for successful trajectories generated by T2 PO is substantially lower than that of SOTA baseline. Meanwhile, the line plot indicates that the exploration efficiency of … view at source ↗

**Figure 6.** Figure 6: (a) illustrates the change in GiGPO’s average output length under different maximum response length settings. (b) illustrates the change in T2 PO ’s average output length under different maximum response length settings. (c) shows the proportion of truncated outputs for GiGPO under different maximum response length settings. (d) shows the proportion of truncated outputs for T2 PO under different maximum re… view at source ↗

**Figure 7.** Figure 7: Additional efficiency analysis on Alfworld. that in the last 50 training steps, our method rarely triggers maximum-length clipping, indicating that it avoids generating redundant or uninformative text and substantially mitigates over-thinking. (3) T2PO more effectively stimulates meaningful interaction-driven reasoning. From view at source ↗

read the original abstract

Recent progress in multi-turn reinforcement learning (RL) has significantly improved reasoning LLMs' performances on complex interactive tasks. Despite advances in stabilization techniques such as fine-grained credit assignment and trajectory filtering, instability remains pervasive and often leads to training collapse. We argue that this instability stems from inefficient exploration in multi-turn settings, where policies continue to generate low-information actions that neither reduce uncertainty nor advance task progress. To address this issue, we propose Token- and Turn-level Policy Optimization (T$^2$PO), an uncertainty-aware framework that explicitly controls exploration at fine-grained levels. At the token level, T$^2$PO monitors uncertainty dynamics and triggers a thinking intervention once the marginal uncertainty change falls below a threshold. At the turn level, T$^2$PO identifies interactions with negligible exploration progress and dynamically resamples such turns to avoid wasted rollouts. We evaluate T$^2$PO in diverse environments, including WebShop, ALFWorld, and Search QA, demonstrating substantial gains in training stability and performance improvements with better exploration efficiency. Code is available at: https://github.com/WillDreamer/T2PO.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces T²PO, an uncertainty-aware framework for controlling exploration in multi-turn agentic RL for LLMs. It argues that pervasive training instability arises from inefficient exploration, where policies generate low-information actions that neither reduce uncertainty nor advance task progress. T²PO intervenes at the token level by monitoring uncertainty dynamics and triggering thinking interventions once marginal uncertainty change falls below a threshold; at the turn level, it identifies turns with negligible exploration progress and dynamically resamples them to avoid wasted rollouts. Evaluations on WebShop, ALFWorld, and Search QA are claimed to demonstrate substantial gains in training stability, performance, and exploration efficiency.

Significance. If the reported gains in stability and performance are robust and the interventions do not introduce uncorrected biases, T²PO could provide a practical heuristic for addressing a key source of instability in multi-turn RL for agentic LLMs. The dual token/turn uncertainty control targets a plausible mechanism and could improve exploration efficiency in interactive reasoning tasks, but its significance depends on empirical validation showing that benefits exceed artifacts from altered trajectory distributions.

major comments (2)

Abstract (turn-level resampling description): The dynamic resampling of turns with negligible exploration progress alters the distribution of trajectories supplied to policy optimization. No importance sampling, adjusted loss terms, or other bias-correction mechanisms are described, which risks biased gradient estimates and could produce apparent stability gains that are artifacts of data rebalancing rather than genuine uncertainty-aware control.
Abstract (evaluation claim): The abstract asserts 'substantial gains in training stability and performance improvements with better exploration efficiency' across WebShop, ALFWorld, and Search QA, yet supplies no quantitative results, baselines, error bars, ablation studies, or implementation details, rendering the central empirical claim unevaluable from the provided text.

minor comments (2)

Abstract: The method is presented as a heuristic framework; explicit definitions or equations for uncertainty (e.g., how marginal uncertainty change is computed at the token level) and the threshold are absent, hindering assessment of whether the approach is parameter-free or introduces new free parameters.
Abstract: No discussion of how token-level interventions interact with credit assignment in the underlying multi-turn RL objective, which could affect the validity of stability claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed review and constructive feedback on our work. We address each major comment below and outline the revisions we will make to the manuscript.

read point-by-point responses

Referee: Abstract (turn-level resampling description): The dynamic resampling of turns with negligible exploration progress alters the distribution of trajectories supplied to policy optimization. No importance sampling, adjusted loss terms, or other bias-correction mechanisms are described, which risks biased gradient estimates and could produce apparent stability gains that are artifacts of data rebalancing rather than genuine uncertainty-aware control.

Authors: We acknowledge the validity of this concern. The turn-level resampling is intended to improve exploration efficiency by reallocating rollouts to turns with higher uncertainty reduction potential. In the full manuscript, we describe the resampling process but do not explicitly discuss bias correction. To address this, we will revise the paper to include an analysis of the trajectory distribution shift, provide theoretical justification for why the uncertainty-guided resampling does not introduce harmful bias in this context, and add empirical ablations showing performance with and without resampling. If necessary, we will incorporate importance sampling weights in future iterations. revision: yes
Referee: Abstract (evaluation claim): The abstract asserts 'substantial gains in training stability and performance improvements with better exploration efficiency' across WebShop, ALFWorld, and Search QA, yet supplies no quantitative results, baselines, error bars, ablation studies, or implementation details, rendering the central empirical claim unevaluable from the provided text.

Authors: We agree that the abstract is high-level and does not include specific numbers due to length constraints typical for abstracts. The full manuscript contains comprehensive experimental results, including quantitative metrics, comparisons to baselines, error bars from multiple runs, ablation studies on the token- and turn-level components, and implementation details in the appendix. To improve the abstract, we will revise it to briefly mention key quantitative improvements, such as relative gains in success rates and stability metrics on the evaluated benchmarks. revision: yes

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on standard RL assumptions plus two new heuristics whose parameters and reliability are not detailed in the abstract.

free parameters (1)

uncertainty change threshold
Triggers token-level thinking intervention; value and selection method unspecified in abstract.

axioms (1)

domain assumption Model uncertainty can be estimated from token probabilities or logits in a way that correlates with exploration value
Core premise of the token-level component; invoked implicitly throughout the abstract.

pith-pipeline@v0.9.0 · 5535 in / 1079 out tokens · 52477 ms · 2026-05-08T19:34:10.688031+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AlphaCoordinateFixation.lean alpha_pin_under_high_calibration unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

M_t = α(H̃_t) + (1−α)(1−C̃_t), α∈[0,1] ... we vary α from 0.2 to 0.4, 0.6, 0.8, and observe that α=0.4 yields the best performance.
IndisputableMonolith/Cost (Jcost = ½(x + x⁻¹) − 1) washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Token entropy H_t = −Σ p_i log p_i and confidence C_t = −(1/j)Σ log p_i used as exploration signals.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

33 extracted references · 24 canonical work pages · 10 internal anchors

[1]

GPT-4 Technical Report

Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

work page internal anchor Pith review arXiv
[2]

anthropic.com

URL https://www. anthropic.com. Large language model. Cai, H. J., Wang, J., Chen, X., and Dhingra, B. How much backtracking is enough? exploring the interplay of sft and rl in enhancing llm reasoning.arXiv preprint arXiv:2505.24273,

work page arXiv
[3]

Seed-grpo: Semantic entropy enhanced grpo for uncertainty-aware policy optimization.arXiv preprint arXiv:2505.12346, 2025

Chen, M., Chen, G., Wang, W., and Yang, Y . Seed-grpo: Semantic entropy enhanced grpo for uncertainty-aware policy optimization.arXiv preprint arXiv:2505.12346,

work page arXiv
[4]

Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

Comanici, G., Bieber, E., Schaekermann, M., Pasupat, I., Sachdeva, N., Dhillon, I., Blistein, M., Ram, O., Zhang, D., Rosen, E., et al. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261,

work page Pith review arXiv
[5]

Agentic entropy-balanced policy optimization.arXiv preprint arXiv:2510.14545,

Dong, G., Bao, L., Wang, Z., Zhao, K., Li, X., Jin, J., Yang, J., Mao, H., Zhang, F., Gai, K., et al. Agentic entropy-balanced policy optimization.arXiv preprint arXiv:2510.14545,

work page arXiv
[6]

Group-in-Group Policy Optimization for LLM Agent Training

Feng, L., Xue, Z., Liu, T., and An, B. Group-in-group policy optimization for llm agent training.arXiv preprint arXiv:2505.10978,

work page internal anchor Pith review arXiv
[7]

Areal: A large-scale asynchronous reinforcement learning system for language reasoning.arXiv preprint arXiv:2505.24298,

Fu, W., Gao, J., Shen, X., Zhu, C., Mei, Z., He, C., Xu, S., Wei, G., Mei, J., Wang, J., Yang, T., Yuan, B., and Wu, Y . Areal: A large-scale asynchronous reinforcement learning system for language reasoning, 2025a. URL https://arxiv.org/abs/2505.24298. Fu, Y ., Wang, X., Tian, Y ., and Zhao, J. Deep think with confidence.arXiv preprint arXiv:2508.15260, ...

work page arXiv 2011
[8]

Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning

Jin, B., Zeng, H., Yue, Z., Yoon, J., Arik, S., Wang, D., Zamani, H., and Han, J. Search-r1: Training llms to reason and leverage search engines with reinforcement learning.arXiv preprint arXiv:2503.09516,

work page Pith review arXiv
[9]

TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension

Joshi, M., Choi, E., Weld, D. S., and Zettlemoyer, L. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension.arXiv preprint arXiv:1705.03551,

work page internal anchor Pith review arXiv
[10]

DeepSeek-V3 Technical Report

Liu, A., Feng, B., Xue, B., Wang, B., Wu, B., Lu, C., Zhao, C., Deng, C., Zhang, C., Ruan, C., et al. Deepseek-v3 technical report.arXiv preprint arXiv:2412.19437,

work page Pith review arXiv
[11]

Learn to reason efficiently with adaptive length-based reward shaping.arXiv preprint arXiv:2505.15612, 2025

Liu, L., Yao, F., Zhang, D., Dong, C., Shang, J., and Gao, J. Flashrl: 8bit rollouts, full power rl, August 2025a. URL https://fengyao.notion.site/flash-rl. Liu, W., Zhou, R., Deng, Y ., Huang, Y ., Liu, J., Deng, Y ., Zhang, Y ., and He, J. Learn to reason efficiently with adaptive length-based reward shaping.arXiv preprint arXiv:2505.15612, 2025b. 9 Unc...

work page arXiv
[12]

A., and Lewis, M

Press, O., Zhang, M., Min, S., Schmidt, L., Smith, N. A., and Lewis, M. Measuring and narrowing the composi- tionality gap in language models. InFindings of the As- sociation for Computational Linguistics: EMNLP 2023, pp. 5687–5711,

2023
[13]

Proximal Policy Optimization Algorithms

Schulman, J., Wolski, F., Dhariwal, P., Radford, A., and Klimov, O. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347,

work page internal anchor Pith review arXiv
[14]

rstar2-agent: Agentic reasoning technical report, 2025

Shang, N., Liu, Y ., Zhu, Y ., Zhang, L. L., Xu, W., Guan, X., Zhang, B., Dong, B., Zhou, X., Zhang, B., et al. rstar2- agent: Agentic reasoning technical report.arXiv preprint arXiv:2508.20722,

work page arXiv
[15]

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Shao, Z., Wang, P., Zhu, Q., Xu, R., Song, J., Bi, X., Zhang, H., Zhang, M., Li, Y ., Wu, Y ., et al. Deepseekmath: Push- ing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300,

work page internal anchor Pith review arXiv
[16]

HybridFlow: A Flexible and Efficient RLHF Framework

Sheng, G., Zhang, C., Ye, Z., Wu, X., Zhang, W., Zhang, R., Peng, Y ., Lin, H., and Wu, C. Hybridflow: A flexi- ble and efficient rlhf framework.arXiv preprint arXiv: 2409.19256,

work page internal anchor Pith review arXiv
[17]

ALFWorld: Aligning Text and Embodied Environments for Interactive Learning

Shridhar, M., Yuan, X., C ˆot´e, M.-A., Bisk, Y ., Trischler, A., and Hausknecht, M. Alfworld: Aligning text and embodied environments for interactive learning.arXiv preprint arXiv:2010.03768,

work page internal anchor Pith review arXiv 2010
[18]

Zerosearch: Incentivize the search capability of llms without searching, 2025

Sun, H., Qiao, Z., Guo, J., Fan, X., Hou, Y ., Jiang, Y ., Xie, P., Zhang, Y ., Huang, F., and Zhou, J. Zerosearch: In- centivize the search capability of llms without searching. arXiv preprint arXiv:2505.04588,

work page arXiv
[19]

Team, K., Du, A., Gao, B., Xing, B., Jiang, C., Chen, C., Li, C., Xiao, C., Du, C., Liao, C., et al. Kimi k1. 5: Scaling reinforcement learning with llms.arXiv preprint arXiv:2501.12599,

work page internal anchor Pith review arXiv
[20]

Qwen3 Technical Report

URL https: //arxiv.org/abs/2505.09388. Trivedi, H., Balasubramanian, N., Khot, T., and Sab- harwal, A. Musique: Multihop questions via single- hop question composition, 2022.URL https://arxiv. org/abs/2108.00573,

work page internal anchor Pith review arXiv 2022
[21]

arXiv preprint arXiv:2509.09265 , year=

Wang, J., Liu, J., Fu, Y ., Li, Y ., Wang, X., Lin, Y ., Yue, Y ., Zhang, L., Wang, Y ., and Wang, K. Harnessing un- certainty: Entropy-modulated policy gradients for long- horizon llm agents, 2025a. URL https://arxiv. org/abs/2509.09265. Wang, X., Zhang, H., Wang, H., Shi, Y ., Li, R., Han, K., Tong, C., Deng, H., Sun, R., Taylor, A., et al. Arlarena: A ...

work page arXiv
[22]

RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning

Wang, Z., Wang, K., Wang, Q., Zhang, P., Li, L., Yang, Z., Yu, K., Nguyen, M. N., Liu, L., Gottlieb, E., Lam, M., Lu, Y ., Cho, K., Wu, J., Fei-Fei, L., Wang, L., Choi, Y ., and Li, M. Ragen: Understanding self-evolution in llm agents via multi-turn reinforcement learning, 2025b. URLhttps://arxiv.org/abs/2504.20073. Wang, Z., Zheng, X., An, K., Ouyang, C....

work page internal anchor Pith review arXiv
[23]

Sim- pletir: End-to-end reinforcement learning for multi-turn tool-integrated reasoning.arXiv preprint arXiv:2509.02479,

Xue, Z., Zheng, L., Liu, Q., Li, Y ., Zheng, X., Ma, Z., and An, B. Simpletir: End-to-end reinforcement learning for multi-turn tool-integrated reasoning.arXiv preprint arXiv:2509.02479,

work page arXiv
[24]

Yang, Z., Qi, P., Zhang, S., Bengio, Y ., Cohen, W., Salakhut- dinov, R., and Manning, C. D. Hotpotqa: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 conference on empirical methods in natural language processing, pp. 2369–2380,

2018
[25]

DAPO: An Open-Source LLM Reinforcement Learning System at Scale

10 Uncertainty-Guided Exploration Control for Stable Multi-Turn Agentic RL Yu, Q., Zhang, Z., Zhu, R., Yuan, Y ., Zuo, X., Yue, Y ., Dai, W., Fan, T., Liu, G., Liu, L., et al. Dapo: An open-source llm reinforcement learning system at scale.arXiv preprint arXiv:2503.14476,

work page Pith review arXiv
[26]

arXiv preprint arXiv:2512.01374 , year=

Zheng, C., Dang, K., Yu, B., Li, M., Jiang, H., Lin, J., Liu, Y ., Yang, A., Zhou, J., and Lin, J. Stabilizing rein- forcement learning with llms: Formulation and practices. arXiv preprint arXiv:2512.01374, 2025a. Zheng, C., Liu, S., Li, M., Chen, X.-H., Yu, B., Gao, C., Dang, K., Liu, Y ., Men, R., Yang, A., et al. Group sequence policy optimization.arXi...

work page arXiv
[27]

I want a red shirt

11 Uncertainty-Guided Exploration Control for Stable Multi-Turn Agentic RL APPENDIX A. More Task Details A.1. Evaluation Metrics A.1.1. WEBSHOP We adopt six complementary evaluation metrics to comprehensively assess task completion quality. (1)Task Scoreis defined as 10×avg. reward , measuring the average accumulated reward per episode. (2)Success Rateis ...

2020
[28]

This is also why parameter-efficient tuning methods, such as Wang et al

suggests that RL in reasoning models does not primarily benefit from memorizing correct solution trajectories, but rather from internalizing structured search behaviors embedded in demonstration traces. This is also why parameter-efficient tuning methods, such as Wang et al. (2023; 2024), can work. In particular, backtracking, where the model explicitly r...

2023
[29]

is a stabilization strategy for thinking control designed to improve the robustness of multi-turn policy optimization. In multi-turn reasoning, the accumulation of low-probability tokens and high sampling stochasticity often produces void turns,i.e., responses that contain neither a valid final answer nor a complete executable structure. Typical void turn...

2024
[30]

Each reward component evaluates a distinct aspect of task completion, including correct product title identification (Title Score), accurate category matching (reward type), attribute fulfillment (reward attribute), and final option selection (reward option). We observe that prompting-based and instruction-tuned baselines exhibit limited performance on fi...

2039
[31]

Codebase Building upon the existing codebase verl (Sheng et al., 2024), our codebase introduces targeted modifications to both the vLLM (Kwon et al.,

and obtain turn-level observation signal: Φk = TkY t=1 M k t 1/Tk 11:ifk >1then 12:Monitor temporal variation across turnsΓ k =|Φ k −Φ k−1| 13:else 14:SetΓ k ←+∞ 15:end if 16:b←b+ 1 17:untilΓ k ≥ηorb≥B max 18:Parse and execute actions in environment: (sk+1, rk)← E.step(a k) 19:Store(s k,a k, rk)intoτ 20:ifIs all done=Truethen 21:break 22:end if 23:end for...

2024
[32]

inference engine and the agent interaction loop, enabling seamless integration with verl while preserving its scalability and modularity. Specifically, we redesign the decoding and rollout pipeline to support fine-grained uncertainty-aware control during generation, while maintaining full compatibility with the step-wise multi-turn training paradigm and m...

1982
[33]

The next step is to execute this search

That should narrow down the results to the relevant products. The next step is to execute this search. </think> <action>search[heather grey men classic fit 3x large under 30]</action> " ⇑ Falling into repeated failures across multiple interaction turns T2PO Enables Decisive, Non-Redundant Actions. The following figure shows a successful WebShop interactio...

1982