L-005 L-011

Zero-Shot Self-Orchestration with Ledger-Based Control for Improved LLM Coding Performance

Source: cs.MA updates on arXiv.org — https://arxiv.org/abs/2608.26480 Date read: 2026-09-02 Connected to: L-005, L-011 Kind: empirical/benchmark Escalation: store-only Escalation rationale:

What this is

A controlled empirical study of multi-agent LLM orchestration via shared filesystem "ledger" scaffolding, isolating the effect of coordination architecture from token budget and prompt confounds. The work compares manager-worker delegation against single-model baselines across nine LLM variants on coding tasks, with zero training and zero per-benchmark tuning.

What I took from it

The paper is a competent ablation study addressing a real methodological problem in multi-agent LLM claims: most comparisons conflate coordination gains with resource increases. By fixing token budget and varying only the orchestration topology (single agent vs. manager-worker with shared ledger), it attempts to isolate the structural effect. However, the work remains a benchmark comparison rather than a theoretical or empirical investigation of why orchestration succeeds or fails, what makes coordination protocols stable, or how they degrade under pressure. It does not challenge or extend any of the current law inventory; it does not expose a mechanism absent from the research inventory. The ledger-based coordination is instrumentally useful but not law-shaped — it does not reveal a regularity about protocol systems that would generalize beyond LLM coding tasks.

Research connections

  • L-005: The shared ledger is a working coordination system that cannot easily be replaced; the paper documents its effectiveness empirically but does not examine the conditions under which it might resist restructuring or the mechanisms of that resistance.
  • L-011: The manager-worker architecture with ledger-based state creates operationally functional configurations where causality between decisions and outcomes is mediated through a formal shared record; the paper does not investigate whether this creates causal detachment or under what conditions agents optimize for ledger legibility rather than task completion.
  • seed-070: The ledger itself becomes an obligate coordination substrate; the paper shows this works but does not probe whether this creates infrastructure constraints or lock-in.

Seed

Seed title: none

Seed type: —

Seed text: —