Structured Rollout Summaries

• Structured rollout summaries are schema-driven representations that compress long agent trajectories into tuples capturing hypotheses, progress, and failure signatures.
• They enhance test-time scaling by replacing raw traces with compact summaries that improve pass rates and sample efficiency.
• Integration with RTV and PDR algorithms facilitates parallel benchmarking and iterative refinement, yielding notable empirical gains.

Structured rollout summaries are compact, schema-driven representations of agent trajectory data that distill complex, long-horizon interactions into a form optimized for selection and reuse during test-time scaling. Designed initially for agentic coding tasks involving extended action/observation histories, structured rollout summaries replace raw execution traces with tuples capturing the key hypotheses, progress, and failure signatures associated with each attempt. This abstraction facilitates both large-scale parallel benchmarking and inference-time refinement by enabling efficient downstream ranking and conditioning of rollouts, achieving substantial empirical gains in pass rates and sample efficiency for advanced code agents (Kim et al., 16 Apr 2026).

1. Formal Definition and Motivation

Let $$\tau = ((\mathcal T_1, \mathcal B_1, \mathcal O_1), \dots, (\mathcal T_M, \mathcal B_M, \mathcal O_M))$$ denote a rollout trajectory comprising agent thoughts $\mathcal T_i$, bash command sets $\mathcal B_i$, and environmental observations $\mathcal O_i$ at each of $M$ steps. The structured summary mapping $S$ compresses $\tau$ into a triple

$S(\tau) = \{H, P, F\}$

where:

• $H$ is a list of short natural language hypotheses about root causes of failures;
• $P$ is a sequence of progress markers (filename, diff) pairs reflecting substantive agent edits;
• $\mathcal T_i$0 enumerates salient failure modes witnessed (error messages, test failures).

Empirically, this form of post-hoc rollup enables efficient selection and reuse, directly addressing the bottleneck in long-horizon agentic evaluation: not generation, but filtering and propagation of the most promising sub-trajectories (Kim et al., 16 Apr 2026).

2. Summary Schema and Pruning Heuristics

The $\mathcal T_i$1 record adopts a fixed three-field schema:

• Hypotheses $\mathcal T_i$2: Short (≤2 sentences) explanations, e.g., "The IndexError arises because the code calls .pop() on an empty list."
• Progress Markers $\mathcal T_i$3: Each tuple provides a file name and a code diff hunk, e.g., $\mathcal T_i$4.
• Failure Modes $\mathcal T_i$5: Substrings of error outputs, e.g., "ModuleNotFoundError: no module named 'pytz'."

To maximize information density and minimize redundant context, pruning heuristics are employed:

• Terminal outputs or stack traces not referenced by any $\mathcal T_i$6 are excluded.
• Only the top-$\mathcal T_i$7 hypotheses by LM-scored association to failure observations are retained.
• Code diffs are deduplicated via canonicalized context lines, and progress markers pruned by edit frequency/novelty.

Summary field sizes are typically capped by prompt length, e.g., $\mathcal T_i$8 (Kim et al., 16 Apr 2026).

Field	Type	Capping
Hypotheses	List[str ≤ 2 sentences]	≤ 5
Progress	List[(filename, code diff)]	≤ 5
Failure Modes	List[str (error/test output substring)]	≤ 5

3. Construction Algorithms and Pseudocode

Structured rollout summaries are constructed via a deterministic LM summarization operator. A functional pseudocode form is given by: $\mathcal O_i$7 The $\mathcal T_i$9 can be instantiated via targeted LM prompts: e.g., "Given these thought–error pairs, how strongly does hypothesis $\mathcal B_i$0 explain these errors?" $\mathcal B_i$1 ranks diffs by occurrence count within $\mathcal B_i$2.

This pipeline fully automates the conversion of complex traces into bounded comparison and conditioning objects suitable for large-scale inference-time processing.

4. Integration with Selection and Refinement Algorithms

Structured summaries are critical enablers for two key inference-time scaling procedures:

Recursive Tournament Voting (RTV):

• Given $\mathcal B_i$3 rollout summaries, RTV partitions them into groups of size $\mathcal B_i$4 and, for $\mathcal B_i$5 votes, prompts an LM to choose which summary in each group is most likely to yield success.
• Winners advance to the next round; repeat until $\mathcal B_i$6.
• In LaTeX notation: $\mathcal B_i$7

Parallel-Distill-Refine (PDR):

• For each batch iteration, summarize rollouts, select top-$\mathcal B_i$8 using RTV, and condition next-rollout generation on the best summaries
• Context for the next round is formed from the top $\mathcal B_i$9 selected summaries, which are then concatenated into the LM input for subsequent rollouts.

Both methods boost effective test-time compute by leveraging the information density and modularity of the summary format, sidestepping the impracticality of raw trace comparison or gradient-based ranking in high-dimensional sequence space (Kim et al., 16 Apr 2026).

5. Hyperparameterization and Model Choices

Implementations are parameterized as follows:

• $\mathcal O_i$0 parallel rollouts/iteration, $\mathcal O_i$1 total iterations for refinement, $\mathcal O_i$2 context size for PDR, $\mathcal O_i$3 group size and $\mathcal O_i$4 votes/group for RTV.
• Field caps: $\mathcal O_i$5.
• Supported LLM families: Claude-4.5-Opus, Gemini-3.1-Pro, Claude-4.5-Sonnet, Gemini-3-Flash, GPT-5-0825.
• Benchmarked on SWE-Bench Verified and Terminal-Bench v2.0.

This standardized configuration enables fair benchmarking and robust scaling studies.

6. Empirical Results and Observations

Use of structured rollout summaries yields significant lift in pass@1 for state-of-the-art agents:

• Claude-4.5-Opus: 70.9% → 77.6% (SWE-Bench Verified), 46.9% → 59.1% (Terminal-Bench v2.0)
• Gemini-3.1-Pro: 72.3% → 76.6% (SWE-Bench Verified), 52.5% → 64.8% (Terminal-Bench v2.0).

Isolated RTV achieves 5–6% (SWE-Bench) and 8–12% (Terminal-Bench) gains without refinement; PDR yields 4–6% boost in round two, and full RTV→PDR→RTV brings improvements up to 12% absolute. Refined rollouts are about 50% shorter, attributing step-count compression to more focused search trajectories (Kim et al., 16 Apr 2026).

7. Comparison to Related Rollout-Based Summarization

While structured rollout summaries originate in test-time scaling of coding agents, related non-myopic acquisition function rollouts in Bayesian Optimization similarly distill $\mathcal O_i$6-step trajectory simulations via compact policy-driven representations. Efficient Monte Carlo rollouts in the Bayesian Optimization context utilize summary statistics, variance reduction (QMC, CRN, control variates), and policy-search selection to manage high-dimensional trajectory data (Lee et al., 2020). In both domains, the central insight is that selection/reuse bottlenecks can be breached by compressing sequential history to a bounded summary that accurately encodes the information required for ranking, conditioning, or reuse.

A plausible implication is that structured rollout summarization admits generalization to other domains where trajectory-level abstraction is required for sample-efficient search or robust evaluation under compute constraints.