The DOI matches its Crossref record; 23 of the 23 references it cites were found in a registry or an index; none of the 11 sources its claims cite is checked for support yet.
AEL: Evolving Agent Harness in Open-Ended Environments
Open accessAbstract
LLM Agents Harnesses are hand-designed and stay fixed, so agents accumulate experience but never learn how to use it: which memories to retrieve, when retrieved evidence is misleading, and when the retrieval strategy itself should change.We introduce Agent Evolving Learning (AEL), a two-timescale framework that evolves the harness, recasting memory use as online policy selection.A fast Thompson Sampling bandit selects among memory-retrieval policies episode by episode, while slow LLM reflection follows a diagnose-before-prescribe principle: it first explains why performance degraded, then injects a targeted new retrieval policy as a bandit arm when the current pool plateaus.AEL outperforms ten self-improving and non-LLM baselines on a sequential portfolio benchmark, lifting Sharpe by 27% over the strongest memory-only variant with the lowest variance among all stochastic methods, and generalizes to a support-ticket routing stream, where it improves accuracy by 18% over reflectionfree Thompson Sampling and by 51% over the best prior baseline.Mechanism studies further show that the gains are causal: reflection helps precisely when regimes demand different retrieval behavior, and is provably no-harm/nogain when the best policy is stable.Code is available at 1 .
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
AEL is a two-timescale agent framework that uses a Thompson Sampling bandit to select memory retrieval policies at the fast timescale and LLM-driven reflection to diagnose failures and inject causal insights at the slow timescale. On a sequential portfolio benchmark, AEL achieves a Sharpe ratio of 2.13±0.47, outperforming five prior self-improving methods and all non-LLM baselines while maintaining the lowest variance among LLM-based approaches.
Key findings
- AEL achieves the highest Sharpe (2.13), Sortino (4.08), and Calmar (10.40) on the D-full benchmark, with the tightest MaxDD (-1.53%) and lowest Sharpe variance (±0.47) among all stochastic/LLM methods.
- Incremental build: Stateless 1.35 → +Memory 1.68 (+24%) → AEL 2.13 (+27%), indicating memory enables cross-episode learning and reflection adds a further diagnostic gain.
- A nine-variant ablation shows every added mechanism degrades performance: removing warm-up Δ-0.88, removing reflection Δ-0.45; adding planner evolution Δ-1.72, per-tool selection Δ-1.70, cold-start Δ-1.31, skills Δ-1.11, FCC Δ-1.09, and LLM-FCC Δ-0.64.
- Uniform credit outperforms sophisticated credit assignment; prior methods show high seed variance, e.g., EvoTool 1.37±1.74, ExpeL 0.76±1.93, Reflexion -0.59±1.33, while the deterministic momentum-weighted baseline reaches 1.44.
Methods
- Sequential portfolio allocation benchmark D-full: 10 sector-diverse tickers across 7 GICS sectors, 1-hour bars, 208 episodes (140 train / 40 val / 28 test), 5 random seeds; test set contains a bear-to-bull transition, and all learning is frozen at test time.
- Baselines include four non-LLM methods, five prior self-improving methods (Reflexion, ExpeL, FactorMiner, Meta-Reflexion, EvoTool), Hyper-Agent, and an AEL incremental build; all methods share the same 12 tools, backbone LLM (Claude Haiku 4.5), and data split.
- Main analysis reports frozen test-phase Sharpe, Sortino, Calmar, cumulative Return%, MaxDD%, win rate, and tail ratio, with component ablations and credit-assignment comparisons.
Limitations
- The paper states AEL is domain-agnostic but evaluated only on a single financial domain with a 2-week test period; cross-domain validation on coding and web benchmarks is described as the key next step.
- The paper notes cost estimates at 5, 10, and 20 bp are approximate, based on average turnover statistics from logged portfolio histories.
- The paper reports that pre-fix credit results were partially driven by buggy interactions; after credit-system bug fixes, FCC and LLM-FCC collapsed (-0.07 and -0.53 Sharpe), while uniform-credit AEL remained at 2.13. Evident from the text: the benchmark is short-horizon (208 episodes, 28 test bars) and high-noise, which the authors themselves emphasize as the setting where adaptive mechanisms incur exploration overhead.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
47 Own result · 38 Own interpretation · 11 Own method · 6 No basis given · 3 Sources not found · 2 Not checked yet
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It also attains the tightest MaxDD (-1.53%)”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“-per-tool selection | 0.61 | -1.34 0.63 | 1.63 | -0.21 | 0.26”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In this pre-fix setting, uniform credit (1.96) outperformed both FCC (1.51) and LLM-FCC (1.87).”
§ AEL: Agent Evolving Learning for Open-Ended Environments > G Controlled Single-Variable Experiments
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“reflection adds a further +27%”
- Also stated in: § Evaluation & Protocol. · plain
“reflection further enhances performance (+27%)”
- Also stated in: § Evaluation & Protocol. · plain
“reflection providing a 27% jump over memory alone.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > 1 Introduction
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § Evaluation & Protocol. · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Even LLM-driven credit (LLM-FCC, ∆-0.64), the least harmful addition, still underperforms uniform credit.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Simplicity beats complexity.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“On a sequential portfolio benchmark (10 sector-diverse tickers, 208 episodes, 5 random seeds), AEL achieves a Sharpe ratio of 2.13±0.47”
- Also stated in: § 1 Introduction · plain
“On a sequential portfolio benchmark (10 tickers, 208 episodes, 5 seeds), AEL achieves Sharpe 2.13±0.47”
- Also stated in: § Evaluation & Protocol. · plain
“AEL achieves the highest Sharpe (2.13)”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Abstract
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 1 Introduction · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The incremental build (Stateless 1.35 → +Memory 1.68 → AEL 2.13)”
- Also stated in: § Evaluation & Protocol. · plain
“Stateless (1.35) → +Memory (1.68) → AEL (2.13)”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § Evaluation & Protocol. · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“yet every additional mechanism we test (planner evolution, per-tool selection, cold-start initialization, skill extraction, and three credit assignment methods) degrades performance.”
- Also stated in: § 4.4 Ablation and Credit Analysis (Q2, Q3) · plain
“every modification to AEL degrades performance.”
- Also stated in: § Simplicity beats complexity. · plain
“every addition to the AEL configuration degrades performance.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Abstract
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 4.4 Ablation and Credit Analysis (Q2, Q3) · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The deterministic momentum-weighted baseline (1.44) outperforms all prior LLM methods on Sharpe”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“memory and reflection together produce a 58% cumulative improvement over the stateless baseline”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Abstract
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- The quote only partly states this claim. 2
- It may state two separate claims as one. 38
- Probably not a substantial claim. 25
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“and all non-LLM baselines”
- Also stated in: § Evaluation & Protocol. · plain
“outperforming all 10 baselines”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Abstract
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The quote only partly states this claim.
- Also stated in: § Evaluation & Protocol. · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“ExpeL (0.76±1.93) and Reflexion (-0.59±1.33)both exhibit extreme seed dependence: individual outlier seeds inflate their means”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The best seed (s456=0.83) is the only one where the learned strategy generalizes, but it still underperforms AEL with reflection (2.13).”
§ AEL: Agent Evolving Learning for Open-Ended Environments > J.1 Evolution Log
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“With the schema fix, generation 0 now improves training Sharpe from -0.974 to -0.44 by actually using tool signals.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > J.1 Evolution Log
plaindirectSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AEL achieves the highest Sharpe (2.13), Sortino (4.08), and Calmar (10.40), outperforming all 10 baselines.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“-warm-up episodes | -1.23 -1.22 2.57 | 0.50 | -1.32 -0.14”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Third, the momentum-weighted baseline is relatively cost-resilient (Sharpe 1.0 at 20 bp)”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“+ LLM-FCC credit | 3.10 | 1.11 | 3.48 -0.94 | 2.60 | 1.87”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“+ uniform credit only | 3.02 | 1.58 | 1.88 | 2.69 | 0.61 | 1.96”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Among prior methods, EvoTool is the strongest on mean Sharpe (1.37) but with the highest variance (±1.74)”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“First, AEL remains the top-performing method at all cost levels tested, maintaining a Sharpe above 1.3 even at 20 bp.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Test Sharpe (frozen) | 0.496”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plainSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“HyperAgent (0.72±0.38) achieves the lowest variance among prior methods, but its recursive code generation fails to escape the initial equalweight template”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Component ablation from AEL (2.13): removing reflection (→1.68) or memory (→1.35) degrades performance.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AEL full (FCC baseline) | 2.77 | 1.32 | 1.95 | 0.97 | 0.53 | 1.51”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Removing per-tool Thompson selection (-1.25) or warm-up (-1.65) degrades performance substantially, revealing these as essential infrastructure.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > G Controlled Single-Variable Experiments
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“After those fixes, both FCC and LLM-FCC collapsed (-0.07 and -0.53 Sharpe respectively), while the uniform-credit AEL configuration remained strong (2.13).”
§ AEL: Agent Evolving Learning for Open-Ended Environments > G Controlled Single-Variable Experiments
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Adding any complexity to AEL hurts: planner evolution (∆-1.72),per-tool selection (∆-1.70), and cold-start (∆-1.31) are most harmful.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Test result (5 seeds) | 0.46±0.41 | 2.13±0.47”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AEL achieves the highest Sharpe (2.13±0.47)with the lowest variance among LLM methods.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plainn = 5- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Planner evolution (∆-1.72),pertool selection (∆-1.70), and cold-start (∆-1.31) are particularly harmful”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Simplicity beats complexity.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“Despite 20 generations, it achieves only a 0.46 Sharpe ratio-below equal-weight (0.70) due to code-generation failures (50% invalid), batch evaluation masking progress, and the absence of modular credit signals.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
What the sources say
- Not matchedcites Zhang et al., 2026
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Not matched
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, most subsequent generations fail code validation (the LLM generates imports that are blocked by the sandbox), and the strategies that do pass tend to overfit the training regime.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > J.1 Evolution Log
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Removing per-tool selection or warm-up collapses performance.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Uniform credit provides a modest gain; LLM-FCC also improves over FCC.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“All modifications degrade performance”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The quote only partly states this claim.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“FCC tuned (intv=40) | 2.82 | -1.05 0.65 | 2.54 | 1.51 | 1.29”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Prior methods evolve a single module (memory, tools, or planning) while holding the rest fixed.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“New retrieval policies are generated only when the existing pool is inadequate, preventing changes from noisy signals.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The contextual planner bandit does not assume any one reasoning style is globally best; instead, it learns which planning style works better under which market”
§ AEL: Agent Evolving Learning for Open-Ended Environments > M Planner and Memory Policy Families
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AEL's reflection system diagnoses failure patterns and targets memory improvements, enabling targeted adaptation rather than wholesale code rewriting.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > J.2 Root Cause Analysis
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Both artifacts were validated via AST parsing and accepted into the module pool by LinUCB, which subsequently selected them when their context features matched.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Additional information
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AEL is domain-agnostic (requiring only a context vector and scalar reward), but we evaluate on a single financial domain with a 2-week test period.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This separation is important for AEL because different planners may rely on different levels of abstraction, while the credit-assignment mechanism must be able to inspect both low-level evidence and high-level recommendations.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > K Dataset Details L Tool Descriptions
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
- Rests on an assumption: Different planners may rely on different levels of abstraction.
- Rests on an assumption: The credit-assignment mechanism must be able to inspect both low-level evidence and high-level recommendations.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The improvements HyperAgent would require-modularity, per-component signals, and online updates-are exactly what AEL provides, suggesting these are fundamental requirements for self-improving agents (Appendix J).”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
hedgeddirectSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“the best retrieval strategy depends on the task context and on the maturity of the memory store itself.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > M Planner and Memory Policy Families
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“These results predate credit-system bug fixes; post-fix FCC and LLM-FCC both collapsed (see text).”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“Modern agents comprise a planner for reasoning (Huang et al., 2024), tools for external API access (Schick et al., 2023), and memory for storing past experience (Xu et al., 2025), yet do not reliably transform experience into improved behavior.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > 1 Introduction
What the sources say
- Not checked yetcites Huang et al., 2024 source
- Not matchedcites Schick et al., 2023
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Xu et al., 2025
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AEL treats the agent as a coupled system where a Thompson Sampling bandit learns how to access experience and LLM-driven reflection diagnoses when structural change is needed.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“All methods share the same 12-tool finance registry for fair comparison, so performance differences cannot be attributed to tool access alone.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > K Dataset Details L Tool Descriptions
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“suggesting that AEL's design reflects fundamental requirements for self-improving agents in complex sequential tasks.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > J.3 Structural Comparison
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“During test, all learning is disabled: bandit posteriors are frozen, memory is read-only, and no evolution occurs, ensuring that test results reflect the agent's generalized architecture.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > D Implementation Protocol
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“the cost adjusted sharpe function in the codebase enables exact computation from individual run logs.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Fast-timescale bandit updates adapt the memory retrieval policy episode by episode, while slow-timescale LLM reflection aggregates trajectories to produce causal diagnoses.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This illustrates how LLM credit can distinguish”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.1 LLM Credit Assignment Example
boosteddirectSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In open-ended environments, agent competence emerges from the interaction of planning, tool use, and memory; when multiple modules change simultaneously, the resulting performance shift is ambiguous.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > 1 Introduction
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The cost estimates at 5, 10, and 20 bp are based on average turnover statistics from logged portfolio histories and should be interpreted as approximate”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Our central finding is that self-diagnosis is the key enabler: memory alone modestly improves over stateless baselines, but reflection produces a qualitative jump (27% Sharpe), while every additional mechanism degrades performance.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > 6 Conclusion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“those disagreements are exactly the kind of cross-module evidence that makes credit assignment informative in our setting.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > K Dataset Details L Tool Descriptions
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
- Rests on an assumption: Momentum, technicals, analyst sentiment, and composite scoring may all point in the same direction during a strong trend, but they diverge during regime shifts.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The key insight is that modular evolution with credit assignment preserves locality: changes to one component do not destroy progress in others.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > J.3 Structural Comparison
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“All metrics are computed on frozen test-phase returns {r 1 , . . ., r T }.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“we retroactively apply proportional transaction costs to the recorded portfolio weight changes at each bar.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“All LLM methods share the same backbone, so data contamination is symmetric and does not confound AEL-vs-baseline comparisons.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“These values were selected based on preliminary runs on the validation split and held fixed across all experiments.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“All baselines use the same LLM backbone (Claude Haiku 4.5), tool set (12 financial tools), and portfolio allocation interface.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
The paper relies on this without support.
The sequential portfolio benchmark (10 sector-diverse tickers, 208 episodes) is a valid and representative testbed for open-ended environments, such that results generalize to other domains.
“While AEL is domain-agnostic, financial prediction provides objective ground truth and regime diversity that stress-tests adaptation”
The paper relies on this without support.
The 208-episode horizon constitutes a 'short-horizon, high-noise' domain where exploration overhead outweighs benefit.
“in short-horizon, high-noise domains, the exploration overhead of adaptive mechanisms outweighs their benefit.”
The paper relies on this without support.
The test set contains a bear-to-bull transition, testing whether the learned architecture generalizes to unseen conditions.
“the test set contains a bear-to-bull transition, testing whether the learned architecture generalizes to unseen conditions.”
The paper relies on this without support.
The reward signal s_t ∈ [-1, 1] converted to uniform reward r_t = clip((s_t + 1)/2, 0, 1) is an appropriate learning signal for the bandit.
“each episode produces a scalar outcome s t ∈ [-1, 1] that is converted into a uniform reward rt = clip((s t + 1)/2, 0, 1) and used directly to update the memory-policy bandit posterior”
The paper relies on this without support.
The 12-tool finance registry is sufficient to represent the relevant information for portfolio allocation decisions.
“All methods share the same 12-tool finance registry for fair comparison”
The paper relies on this without support.
Different planners may rely on different levels of abstraction.
“different planners may rely on different levels of abstraction”
The paper relies on this without support.
The credit-assignment mechanism must be able to inspect both low-level evidence and high-level recommendations.
“the credit-assignment mechanism must be able to inspect both low-level evidence and high-level recommendations.”
The paper relies on this without support.
Momentum, technicals, analyst sentiment, and composite scoring may all point in the same direction during a strong trend, but they diverge during regime shifts.
“momentum, technicals, analyst sentiment, and composite scoring may all point in the same direction during a strong trend, but they diverge during regime shifts”
The paper relies on this without support.
Frequent large weight shifts are penalized by the market.
“frequent large weight shifts are penalized by the market.”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
“Figure 3b confirms that AEL achieves the highest mean with the lowest variance (±0.47) among all LLM methods, while prior methods exhibit high instability.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
“| | HyperAgent | 0.72±0.38 | 1.00±0.53 | 3.90±2.37 | +0.34±0.20 | -2.49±0.15 | 0.49 | 0.76 |”
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
“| | HyperAgent | 0.72±0.38 | 1.00±0.53 | 3.90±2.37 | +0.34±0.20 | -2.49±0.15 | 0.49 | 0.76 |”
§ AEL: Agent Evolving Learning for Open-Ended Environments > J.1 Evolution Log
“Despite these improvements, HyperAgent achieves a mean test Sharpe of only 0.46±0.41across 5 seeds.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
“Among prior methods, EvoTool is the strongest on mean Sharpe (1.37) but with the highest variance (±1.74), indicating that its evolutionary tool policy is sensitive to initialization.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
“| | ExpeL | 0.76±1.93 | 1.68±3.93 | 3.97±9.79 | +0.24±0.63 | -1.76±0.16 | 0.45 | 1.05 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
“| AAPL | Technology | Mega | 60/20/20 |”
§ AEL: Agent Evolving Learning for Open-Ended Environments > 4.1 Experiments Setup
“The D-full benchmark comprises 10 sector-diverse tickers spanning 7 GICS sectors at 1-hour resolution (208 episodes: 140 train / 40 val / 28 test).”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ AEL: Agent Evolving Learning for Open-Ended Environments > Evaluation & Protocol.
“It also attains the tightest MaxDD (-1.53%) and the lowest Sharpe variance (±0.47) among all stochastic methods.”
§ AEL: Agent Evolving Learning for Open-Ended Environments > O.2 Code Evolution Examples
“| | EvoTool | 1.37±1.74 | 2.73±3.58 | 6.28±7.70 | +0.34±0.46 | -1.46±0.15 | 0.47 | 1.17 |”
Citations
Cites
Finite-time Analysis of the Multiarmed Bandit Problem
Auer, P., Cesa-Bianchi, N., & Fischer, P. (2002) · Machine Learning · doi:10.1023/a:1013689704352 · Cited by 6,033
Chain-Of-Thought Prompting Elicits Reasoning in Large Language Models
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q. V., & Zhou, D. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-1800 · Cited by 4,505
A contextual-bandit approach to personalized news article recommendation
Li, L., Chu, W., Langford, J., & Schapire, R. E. (2010) · Proceedings of the 19th international conference on World wide web · doi:10.1145/1772690.1772758 · Cited by 2,562 · cited 2×
“The framework additionally supports two further bandit levels, evaluated in the ablation (Table 3): Planner selection (LinUCB (Li et al., 2010)).”
Tree of Thoughts: Deliberate Problem Solving with Large Language Models
Yao, S., Yu, D., Zhao, J., Shafran, I., Griffiths, T. L., Cao, Y., & Narasimhan, K. (2023) · arXiv · doi:10.48550/arxiv.2305.10601 · Cited by 1,158
Toolformer: Language Models Can Teach Themselves to Use Tools
Schick, T., Dwivedi-Yu, J., Dessì, R., Raileanu, R., Lomeli, M., Zettlemoyer, L., Cancedda, N., & Scialom, T. (2023) · arXiv · doi:10.48550/arxiv.2302.04761 · Cited by 1,150
Reflexion: Language Agents with Verbal Reinforcement Learning
Shinn, N., Cassano, F., Berman, E., Gopinath, A., Narasimhan, K., & Yao, S. (2023) · arXiv · doi:10.48550/arxiv.2303.11366 · Cited by 1,087
Analysis of Thompson Sampling for the multi-armed bandit problemPreprint
Agrawal, S., & Goyal, N. (2011) · arXiv (Cornell University) · doi:10.48550/arxiv.1111.1797 · Cited by 854
Voyager: An Open-Ended Embodied Agent with Large Language Models
Wang, G., Xie, Y., Jiang, Y., Mandlekar, A., Xiao, C., Zhu, Y., Fan, L., & Anandkumar, A. (2023) · arXiv · doi:10.48550/arxiv.2305.16291 · Cited by 222
Expel: LLM agents are experiential learners
Zhao, A., Huang, D., Xu, Q., Lin, M., Liu, Y.-J., & Huang, G. (2024) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v38i17.29936 · Cited by 153
MemGPT: Towards LLMs as Operating Systems
Packer, C., Wooders, S., Lin, K., Fang, V., Patil, S. G., Stoica, I., & Gonzalez, J. E. (2023) · arXiv · doi:10.48550/arxiv.2310.08560 · Cited by 64
Research works printed but not linked: 8
- Olivier Chapelle and Lihong Li. An empirical evaluation of thompson sampling. Advances in Neural Information Processing Systems, 24, 2011.
- Amirata Ghorbani and James Zou. Data shapley: Equitable valuation of data for machine learning. Proceedings of the 36th International Conference on Machine Learning, pp. 2242-2251, 2019.
- Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. SWE-bench: Can language models resolve real-world GitHub issues? Proceedings of the 12th International Conference on Learning Representations, 2024.
- Scott M. Lundberg and Su-In Lee. A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 30, 2017.
- Lloyd S. Shapley. A Value for n-Person Games, volume 2. Princeton University Press, 1953.
- Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models. Proceedings of the 11th International Conference on Learning Representations, 2023.
- Jenny Zhang, Bingchen Zhao, Wannan Yang, Jakob Foerster, Jeff Clune, Minqi Jiang, Sam Devlin, and Tatiana Shavrina. Hyperagents. arXiv, 2026.
- Shuyan Zhou, Frank F. Xu, Hao Zhu, Xuhui Zhou, Robert Lo, Abishek Sridhar, Xianyi Cheng, Tianyue Ou, Yonatan Bisk, Daniel Fried, Uri Alon, and Graham Neubig. Webarena: A realistic web environment for building autonomous agents. Proceedings of the 12th International Conference on Learning Representations, 2024.
Cited by
Sorted by citations
Self-Improvements in Modern Agentic Systems: A SurveyPreprint
Ren, Z., Chen, Y., Guo, D., Rong, G., Li, T., Xiong, R. B., Lan, Q., Wang, W., Nanbo, L., Yang, Y., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.13104 · Cited by 32
CoMAP: Co-Evolving World Models and Agent Policies for LLM Agents
Liu, Y., Wang, J., Wang, H., & Li, W. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.1200 · Cited by 9
Co-Evolving Skill Generation and Policy OptimizationPreprint
Zhang, Z., Lin, Y., Kuang, N. L., Wu, L., Li, X., Liu, S., & Ma, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.08755 · Cited by 9
Individual Turing Test: A Case Study of LLM-based Simulation Using Longitudinal Personal Data
Guo, M., Ye, Z., Xu, W., Zhu, X., Hua, W., & Metaxas, D. (2026) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3805712.3809868 · Cited by 6
Self-citationNode-as-Agent: Graph Agentic NetworkPreprint
Guo, M., Zhu, X., Jiao, Q., Liu, X., Xue, H., Zhang, C., Lin, S., Huang, J., Ye, Z., & Zhang, Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.00429 · Cited by 4
Self-citation
Metadata
- Authors
- Xu, W., Han, J., Guo, M., Mei, K., Zhu, X., Zhang, H., & Metaxas, D. N.
- Affiliations
- Rutgers, The State University of New Jersey (US)
- Venue
- Findings of the Association for Computational Linguistics: EMNLP 2026
- Published
- 2026
- Pages
- 27857-27885
- OpenAlex
- W7155488304
- Cited by
- 10
- References
- 13
- Citations by year
- 2026: 9
- FWCI
- 53.61
- Type
- conference paper
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Social Sciences › Decision Sciences › Management Science and Operations Research › Advanced Bandit Algorithms Research
- Physical Sciences › Computer Science › Artificial Intelligence › AI-based Problem Solving and Planning
- Physical Sciences › Computer Science › Artificial Intelligence › Data Stream Mining Techniques
- Keywords
- LLM agents, self-improving agents, portfolio optimization, open-ended environments
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-11-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
