Voting, Self-Consistency, and Debate Topology

> The cheapest aggregation: sample N independent agents, majority-vote. Wang et al. 2022 self-consistency did this with one model sampled N times. Multi-agent extends it with heterogeneous agents to escape monoculture — different models, different prompts, different temperatures, different contexts. Beyond majority vote, debate topology matters: MultiAgentBench (arXiv:2503.01935, ACL 2025) evaluated star / chain / tree / graph coordination and found graph best for research, with a "coordination tax" past ~4 agents. AgentVerse (ICLR 2024) documents two emergent patterns — volunteer behaviors and conformity behaviors — and conformity is both a feature (finding consensus) and a risk (groupthink, Lesson 24). This lesson maps the topology space, builds each variant, and measures the coordination tax.

Type: Learn + Build

Languages: Python (stdlib)

Prerequisites: Phase 16 · 07 (Society of Mind and Debate), Phase 16 · 14 (Consensus and BFT)

Time: ~75 minutes

Problem

Debate can improve accuracy (Du et al., arXiv:2305.14325). It can also degrade it. Whether debate helps depends on four structural choices:

1. Who talks to whom (topology).
2. How many rounds (Du 2023: both rounds and agents matter independently).
3. Whether agents are heterogeneous (different base models break monoculture).
4. Whether an adversarial voice is present (steel-manning vs. straw-manning).

Teams that bolt "run 5 agents and vote" onto a task often regress vs. a single agent. The failures are not random. They track topology and heterogeneity. This lesson is the topology map.

Concept

Self-consistency, the single-model baseline

Wang et al. 2022 ("Self-Consistency Improves Chain of Thought Reasoning") sampled the same model N times at temperature > 0 and majority-voted on reasoning-path answers. The result on GSM8K: substantial gains with N=40 samples over a single greedy decode. Self-consistency is the single-agent precursor to multi-agent voting.

Limit: self-consistency uses one base model. Errors are correlated by construction. If the model has a systematic bias, all N samples share it.

Multi-agent vote, the heterogeneous extension

Replace N samples with N *different* agents. Different base models (Claude, GPT, Llama), different prompts, different tool access. The benefit: uncorrelated errors. The cost: different agents cost different amounts; coordinating them adds overhead.

The canonical 2026 name for heterogeneous debate is A-HMAD — Adversarial Heterogeneous Multi-Agent Debate. Not universally adopted, but papers use the term for "different models debate, which reduces correlated errors from monoculture collapse."

The four topologies

star                chain               tree                graph

    ┌─A─┐           A─B─C─D         ┌──A──┐              A───B
    │   │                           │     │              │ × │
    B   C                           B     C              D───C
    │   │                          / \   / \
    D   E                         D   E F   G           (fully connected)

Star: one hub, all others talk only to hub. Equivalent to supervisor-worker without back-channel.

Chain: linear, each agent sees the prior one's output. Pipeline-like.

Tree: hierarchical, used by hierarchical agent systems (Lesson 06).

Graph: any-to-any. Includes fully-connected clique and arbitrary DAGs.

The coordination tax (MultiAgentBench)

MultiAgentBench (MARBLE, ACL 2025, arXiv:2503.01935) benchmarked star, chain, tree, graph on a task suite including research, coding, and planning. Key measured results:

• Graph topology wins on research tasks. Information flows any-to-any; agents can critique each other.
• Star wins on fast-answer factual tasks. Hub filters and consolidates.
• Chain wins on stepwise pipelines (staged refinement).
• Coordination tax appears past ~4 agents in graph topology. Wall-clock and token cost grow faster than quality.

The 4-agent ceiling is empirical, not fundamental. It reflects 2026 LLM context capacity: each agent's context fills with peers' outputs, and marginal value of adding agent N+1 drops once everyone can see everyone.

Multi-Agent Debate Strategies ("Should we be going MAD?")

arXiv:2311.17371 is the 2023 survey of MAD strategies. Key finding replicated by others: MAD variants that are *structurally similar* to self-consistency (independent sampling + aggregation) often underperform self-consistency when using the same budget. MAD helps most when agents are genuinely heterogeneous and the debate has adversarial structure (one agent argues against).

AgentVerse emergent patterns

AgentVerse (ICLR 2024, https://proceedings.iclr.cc/paper_files/paper/2024/file/578e65cdee35d00c708d4c64bce32971-Paper-Conference.pdf) documents two behaviors that emerge from multi-agent debate even without explicit design:

• Volunteer. An agent offers help ("I can take the next step") unprompted. Useful: it allocates work to the most-capable agent for a subtask.
• Conformity. An agent adjusts its stance to match a critic, even when the critic is wrong. This is the debate-equivalent of sycophancy (Lesson 14).

Conformity is why debate-until-agreement rewards bullies. Bounded rounds with a separate judge mitigate.

Heterogeneity: the actual knob that moves accuracy

A 2024-2026 pattern in the practical literature: swapping one of your N agents for a different base model gives a bigger accuracy bump than increasing N by 1. The intuition is monoculture — each new independent-error source is worth more than an additional correlated sample.

In the limit, heterogeneity beats numerosity. Three different models beat five copies of one model on most tasks that have clean ground truth.

Jury methods

The Sibyl framework (cited in Minsky-LLM literature) formalizes a "jury" — a small set of specialized agents that refine answers by voting at each stage. Unlike plain majority vote, a jury has roles: one agent cross-examines, one supplies context, one scores plausibility. Jury methods are a midpoint between plain vote (cheap, monoculture-prone) and full MAD (expensive, conformity-prone).

When vote-with-debate dominates

• The question has ground truth (fact, math, code behavior). Vote convergence is meaningful.
• Agents can access different sources or tools (heterogeneity is available).
• Rounds are bounded (2-3 typical) and there is a separate judge or verifier.
• Budget allows 3-5 agents. Beyond 5-7 on graph topology, coordination tax dominates.

When vote-with-debate hurts

• The question is opinion-shaped. Agents converge to whichever answer looks most confident, not most correct.
• All agents share a base model. Monoculture makes consensus meaningless.
• Rounds are unbounded. Conformity wins every time.
• The task is simple. A single agent with self-consistency at N=5 is cheaper and as accurate.

Build It

code/main.py implements:

• run_star(agents, hub, question) — hub polls each worker, aggregates.
• run_chain(agents, question) — sequential refinement.
• run_tree(root, children, question) — hierarchical with depth-2 aggregation.
• run_graph(agents, question, rounds) — all-to-all debate, bounded rounds.
• A scripted heterogeneity dial: each agent has an error_bias indicating its systematic wrongness.
• A measurement harness that runs each topology at N=3, 5, 7 and reports (accuracy, total_tokens, wallclock_simulated).

Run:

python3 code/main.py

Expected output: a table of topology × N → (accuracy, tokens, latency). Graph wins at N=3-5 on the research-style tasks; star wins on the fast-factual tasks; graph at N=7 shows the coordination tax (latency inflates faster than accuracy).

Use It

outputs/skill-topology-picker.md is a skill that reads a task description and recommends a topology (star / chain / tree / graph), an N (number of agents), a heterogeneity profile (base models to use), and a round bound.

Ship It

For any ensemble:

• Start with self-consistency at N=5 using one strong base model. It is the cheap baseline.
• Upgrade to heterogeneous voting at N=3 if accuracy matters. Measure the delta.
• Only upgrade to debate topology if the task has structure (research, multi-step) and bounded rounds are feasible.
• Always log the minority cluster. When a minority is persistently right, you have a diversity signal.
• Benchmark wall-clock and tokens alongside accuracy. "Better accuracy at 10x cost" is a business decision.

Exercises

1. Run code/main.py. Plot the coordination-tax curve for graph topology: accuracy vs N, tokens vs N. At what N does the curve inflect?
2. Implement A-HMAD: three agents with deliberately different biases. How does the all-same-bias baseline compare to A-HMAD on the monoculture attack from Lesson 14?
3. Add a "judge" role to the graph topology that does not vote, only scores the final consensus. Does this change the emergent conformity behavior?
4. Read the AgentVerse paper (ICLR 2024). Identify which emergent behavior your implementation exhibits most strongly. Can you elicit the opposite behavior by a prompt change?
5. Read MultiAgentBench (arXiv:2503.01935) Section 4 (topology experiments). Reproduce the "graph-wins-research" result on one task from the paper using your harness.

Key Terms

Further Reading

• Wang et al. — Self-Consistency Improves Chain of Thought Reasoning — single-model baseline
• Du et al. — Improving Factuality and Reasoning via Multiagent Debate — both agents AND rounds matter independently
• MultiAgentBench / MARBLE — topology benchmark showing graph best for research, chain for pipelines
• Should we be going MAD? — MAD-strategy survey; finds MAD often loses to self-consistency at equal budget
• AgentVerse (ICLR 2024) — volunteer and conformity emergent patterns
• MARBLE repo — reference benchmark implementation