Try Hyperspace — Lattices

arxiv.org

arXivBrowseSubjectsHelp

cs > cs.AI

Tideline: Phase-Aligned Scheduling for Long-Horizon Agents

PDFOther formats

Authors: R. Okafor, M. Lindqvist, S. Tan, et al.

[v1] 4 Nov 2026 · arXiv:2611.04127

Abstract

We present Tideline, a scheduler that aligns agent subtasks to shared phases of a planning horizon. Instead of interleaving work greedily, Tideline defers each task to the next slack phase, which keeps long-horizon plans consistent while reducing redundant tool calls. Across three planning suites it improves task completion at equal compute.

Figure 1. Three task phases converge on a shared horizon.

Comments: 14 pages, 6 figures
Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)

tideline — tide.rs

TIDELINE

src

Rtide.rs

Rscheduler.rs

Rphase.rs

Cargo.toml

MREADME.md

tide.rsscheduler.rs

12345678910111213141516171819

use std::time::Duration;pub struct Tide<T> {    phases: Vec<Phase<T>>,    horizon: Duration,}impl<T: Task> Tide<T> {    pub fn new(horizon: Duration) -> Self {        Self { phases: Vec::new(), horizon }    }    /// Place a task on the next slack phase.    pub fn schedule(&mut self, task: T) -> PhaseId {        let slot = self.slack_phase();        self.phases[slot].push(task);        PhaseId(slot)    }}

main0 problemsRustLn 15, Col 41

Lattices Hyperspace, simulated window layout demo