# PRECEPT: Planning Resilience via Experience, Context Engineering & Probing Trajectories A Unified Framework for Test-Time Adaptation with Compositional Rule Learning and Pareto-Guided Prompt Evolution Source: [PRECEPT: Planning Resilience via Experience, Context Engineering & Probing Trajectories A Unified Framework for Test-Time Adaptation with Compositional Rule Learning and Pareto-Guided Prompt Evolution]() Collection index: [llms.txt]() Data updated: 2026-03-12T05:00:12.408Z Authors: Arash Shahmansoori Published: 2026-03-10T13:16:45.000Z ## Abstract LLM agents that store knowledge as natural language suffer steep retrieval degradation as condition count grows, often struggle to compose learned rules reliably, and typically lack explicit mechanisms to detect stale or adversarial knowledge. We introduce PRECEPT, a unified framework for test-time adaptation with three tightly coupled components: (1) deterministic exact-match rule retrieval over structured condition keys, (2) conflict-aware memory with Bayesian source reliability and threshold-based rule invalidation, and (3) COMPASS, a Pareto-guided prompt-evolution outer loop. Exact retrieval eliminates partial-match interpretation errors on the deterministic path (0% by construction, vs 94.4% under Theorem~B.6's independence model at N=10) and supports compositional stacking through a semantic tier hierarchy; conflict-aware memory resolves static--dynamic disagreements and supports drift adaptation; COMPASS evaluates prompts through the same end-to-end execution pipeline. Results (9--10 seeds): PRECEPT achieves a +41.1pp first-try advantage over Full Reflexion (d\>1.9), +33.3pp compositional generalization (d=1.55), 100% $P\_1$ on 2-way logistics compositions (d=2.64), +40--55pp continuous learning gains, strong eventual robustness under adversarial static knowledge (100% logistics with adversarial SK active; partial recovery on integration), +55.0pp drift recovery (d=0.95, p=0.031), and 61% fewer steps. Core comparisons are statistically significant, often at p\<0.001. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Continuous Learning on Integration domain](): Continuous Learning; Integration Domain; P1 Score: 70.4. - [Rule drift adaptation on Integration Domain](): Rule drift adaptation; Integration Domain; P1 Score: 60.7. - [Rule drift adaptation on Logistics Domain](): Rule drift adaptation; Logistics Domain; P1 Performance Change (pp): 83.3. - [Compositional Generalization on Logistics](): Compositional Generalization; Logistics; P1: 100. - [Compositional Generalization on Integration](): Compositional Generalization; Integration; P1 Score: 49. - [Agentic Workflow Automation on Integration domain](): Agentic Workflow Automation; Integration Domain; Overall Success Rate ($P\_t$): 83.3. - [Agentic Workflow Automation on Booking domain](): Agentic Workflow Automation; Booking domain; Success Rate ($P\_t$): 99.4. - [Agentic Workflow Automation on Logistics domain](): Agentic Workflow Automation; Logistics Domain; Overall Success Rate ($P\_t$): 100. - [Continual Learning on Logistics 1st Encounter](): Continuous Learning; Logistics 1st Encounter; P1: 44.8. - [Continual Learning on Logistics 2nd Encounter](): Continuous Learning; Logistics 2nd Encounter; P1: 82.5. - [Continuous Learning on Logistics 3rd Encounter](): Continuous Learning; Logistics 3rd Encounter; P1 Score: 87.5. - [Continuous Learning on Logistics 4th Encounter](): Continuous Learning; Logistics 4th Encounter; P1 Score: 100. - [Continuous Learning on Logistics Delta 1st to 4th](): Continuous Learning; Logistics Delta 1st to 4th; Delta P1 (pp): 55.2. - [First-try Success on Integration domain (test)](): First-try Success; Integration domain (test); P1: 80. - [First-try Success on Booking domain (test)](): First-try Success; Booking domain (test); P1 Success Rate: 94.1. - [First-try Success on Logistics domain (test)](): First-try Success; Logistics domain (test); P1 Success Rate: 95. - [First-try Success on Average across domains](): First-try Success; Average across domains; P1 Success Rate: 89.7. - [Rule persistence after session restart on Integration domain Experiment 5 (test)](): Rule persistence after session restart; Integration domain Experiment 5 (test); Enc. 1 P1: 100. - [Rule persistence after session restart on Logistics domain Experiment 5 (test)](): Rule persistence after session restart; Logistics domain Experiment 5 (test); P1 Score (Enc. 1): 88.5. ## Original sources and implementation [Original paper]() [PDF]()