Anthropic Opus 4.6 Hackathon

Never settle forone-sided

Multiple AI agents argue every angle of your question, poke holes in each other's logic, and back claims with real evidence.
What survives is a verdict you can actually trust.

Argument GraphCross-ExaminationDecision Matrix

Enter Access Codeor enter access code →

See it in action

Pick a scenario, watch agents reason

Prompt

Should we migrate from monolith to microservices this quarter?

Architecture Scope

Platform constraints

Inputquestion:clarification

Setup + Personas

agent:intro / protocol:assignments

Opusstage:setup

Reliability Lead

Failure domains

Sonnetstage:opening

Cross Examination

forced claim assignments

Sonnetstage:cross-exam

FinOps Reviewer

Infra cost impact

Sonnetstage:closing

Feasibility Check

Haiku convergence

Haikuconsensus:check

Fact Check

evidence verification pass

Haikuanalysis:fact_check

Architecture Verdict

Opus synthesis

Opusverdict:complete

Architecture Scope

Platform constraints

Inputquestion:clarification

Setup + Personas

agent:intro / protocol:assignments

Opusstage:setup

Reliability Lead

Failure domains

Sonnetstage:opening

Cross Examination

forced claim assignments

Sonnetstage:cross-exam

FinOps Reviewer

Infra cost impact

Sonnetstage:closing

Feasibility Check

Haiku convergence

Haikuconsensus:check

Fact Check

evidence verification pass

Haikuanalysis:fact_check

Architecture Verdict

Opus synthesis

Opusverdict:complete

Capabilities

Structured Debate Engine

Watch a production-style engine window with rolling transcript cards, cross-exam state changes, and synthesis output in a continuous preview loop.

Engine Preview

Real-time debate transcript window

Follow the exact flow users see in Debate view: orchestrator setup, agent arguments, and synthesis updates in one timeline.

Agents

Rounds

Latency

~28s

verdict-engine / transcript

DebatingRound 2/3Balanced

Should we launch a freemium plan this quarter or keep paid-only?

Debate FeedDebatingOpus Orchestrator analyzing...

Agents3Rounds3Latency~28s

✓Always-On Debate Modenow

Debating pipeline active

Setup personasOpening argumentsCross-examinationConsensus checkFinal synthesis

+0:04Opus OrchestratorOpus 4.6

Setup

Personas assigned
Budget guard: $0.70
Target decision: pricing strategy

+0:11The EconomistSonnet 4.5

Claim A

Freemium can 2x top-of-funnel growth.

Risk: support costs spike if activation is weak.

+0:16The AdvocateSonnet 4.5

Counter

Launch freemium with strict limits.

Upsell at team collaboration
Upsell at advanced exports

+0:22The CriticSonnet 4.5

Rebuttal requested

Show conversion assumptions and churn impact by month 3.

+0:28Consensus CheckHaiku 4.5

Drift detected

Growth upside and margin risk are unresolved.

Routing targeted cross-exam.

+0:33Opus SynthesizerOpus 4.6

Interim synthesis

Direction: limited freemium test in one segment.

Confidence moving 64% -> 74%.

analyzingOpus Orchestratorthinking...

Preview mode — full transcript, evidence links & exports in Dashboard

Open app →

Active personas

The EconomistThe AdvocateThe Critic

Multi-Agent Debate

Spawn distinct personas that research, challenge, and refine claims from multiple perspectives.

Smart Model Routing

Opus for orchestration, Sonnet for depth, and Haiku for consensus checks to keep quality high and costs low.

Emergent Decision Matrix

Criteria and weights are generated from the debate itself, not rigid templates.

Argument Graph

Claims are tracked as standing, challenged, rebutted, or conceded in a structured graph.

Evidence-Grounded

Agents search and cite supporting or contradictory evidence before final synthesis.

Live Streaming

Watch stage progress, token usage, and routing decisions while the debate runs.

How It Works

From question to verdict in three steps

Ask Your Question

Submit a decision problem with context, constraints, and desired outcome.

Run Structured Debate

Agents execute opening arguments, forced cross-exam, and rebuttal rounds.

Get Decision Verdict

Receive recommendation, confidence, dissenting view, blind spots, and source-backed rationale.

Why Verdict

Stop asking the same question 20 times

Traditional AI gives you one perspective. You refine, re-prompt, and still miss blind spots. Verdict gets it right the first time.

The old way

Single-AI chat

One perspective at a time. You keep re-prompting to cover angles.

Should we use microservices?

Yes, microservices offer better scalability and...

Wait, what about the downsides?

Well, there is added complexity and DevOps...

Compare costs with monolith?

Monoliths are cheaper initially but harder to...

What would a CTO recommend?

A CTO might prioritize team velocity over...

Security implications?

Each microservice expands the attack surface...

How about team readiness?

You need strong DevOps culture and...

What does the data say?

Studies show 60% of migrations encounter...

+13 more re-prompts

You manually cover each angle

No one challenges weak claims

No evidence verification

20+

prompts to maybe get a balanced answer

Still missing perspectives you didn't think to ask about

The Verdict way

Multi-agent debate

You ask once. Agents argue every angle for you.

“Should we migrate to microservices?”

agents activate

Advocate

FOR

Critic

AGAINST

Analyst

DATA

cross-examining

Opening arguments

Agents present initial positions

Cross-examination

Agents challenge each other

Evidence search

Web search validates claims

Final synthesis

Structured verdict with confidence

Verdict: Proceed with caution87%

Decision matrixDissent reportEvidence linksRisk analysis

Multiple agents cover every angle

Claims are challenged and stress-tested

Real evidence backs every argument

prompt. Every angle covered automatically.

Agents find blind spots you didn't know existed

Model Routing

Multi-tier execution strategy

Target Cost

$0.29 - $0.72 per debate

Opus 4.6

Orchestrator + final synthesis

$15 in / $75 out per 1M tokens

Sonnet 4.5

Debate rounds + cross-exam

$3 in / $15 out per 1M tokens

Haiku 4.5

Consensus + classification

$0.8 in / $4 out per 1M tokens

All-Opus baselineRouted savings

Approximate 60% cost reduction with tiered routing.

Deliverables

Shipping as four packages

@verdict/enginenpm

SDK for protocol orchestration, routing, argument graph, and scoring.

README

verdict-webweb

Premium web interface for live debate runs, reports, and dashboards.

README

verdict-clicli

Terminal workflow for scripted debates and fast local experimentation.

README

verdict-mcpmcp

MCP server exposing Verdict as tools inside Claude Code and Desktop.

README

CLI Demo

Terminal flow in action

Added the recorded `verdict-cli` run directly on the landing page so judges can immediately see the terminal UX.

Open CLI README

Startup + model routing

Live argument turns

Final verdict summary

Packages

40+

UI Features

Novel Algorithms

MIT

Licensed

Ready to try it?

Run a real debate, inspect the protocol flow, and export a report your team can act on.

Enter Access Code

Pick a scenario, watch agents reason

Prompt

Should we migrate from monolith to microservices this quarter?

Architecture Scope

Platform constraints

Inputquestion:clarification

Setup + Personas

agent:intro / protocol:assignments

Opusstage:setup

Reliability Lead

Failure domains

Sonnetstage:opening

Cross Examination

forced claim assignments

Sonnetstage:cross-exam

FinOps Reviewer

Infra cost impact

Sonnetstage:closing

Feasibility Check

Haiku convergence

Haikuconsensus:check

Fact Check

evidence verification pass

Haikuanalysis:fact_check

Architecture Verdict

Opus synthesis

Opusverdict:complete

Architecture Scope

Platform constraints

Inputquestion:clarification

Setup + Personas

agent:intro / protocol:assignments

Opusstage:setup

Reliability Lead

Failure domains

Sonnetstage:opening

Cross Examination

forced claim assignments

Sonnetstage:cross-exam

FinOps Reviewer

Infra cost impact

Sonnetstage:closing

Feasibility Check

Haiku convergence

Haikuconsensus:check

Fact Check

evidence verification pass

Haikuanalysis:fact_check

Architecture Verdict

Opus synthesis

Opusverdict:complete

Structured Debate Engine

Watch a production-style engine window with rolling transcript cards, cross-exam state changes, and synthesis output in a continuous preview loop.

Engine Preview

Real-time debate transcript window

Follow the exact flow users see in Debate view: orchestrator setup, agent arguments, and synthesis updates in one timeline.

Agents

Rounds

Latency

~28s

verdict-engine / transcript

DebatingRound 2/3Balanced

Should we launch a freemium plan this quarter or keep paid-only?

Debate FeedDebatingOpus Orchestrator analyzing...

Agents3Rounds3Latency~28s

✓Always-On Debate Modenow

Debating pipeline active

Setup personasOpening argumentsCross-examinationConsensus checkFinal synthesis

+0:04Opus OrchestratorOpus 4.6

Setup

Personas assigned
Budget guard: $0.70
Target decision: pricing strategy

+0:11The EconomistSonnet 4.5

Claim A

Freemium can 2x top-of-funnel growth.

Risk: support costs spike if activation is weak.

+0:16The AdvocateSonnet 4.5

Counter

Launch freemium with strict limits.

Upsell at team collaboration
Upsell at advanced exports

+0:22The CriticSonnet 4.5

Rebuttal requested

Show conversion assumptions and churn impact by month 3.

+0:28Consensus CheckHaiku 4.5

Drift detected

Growth upside and margin risk are unresolved.

Routing targeted cross-exam.

+0:33Opus SynthesizerOpus 4.6

Interim synthesis

Direction: limited freemium test in one segment.

Confidence moving 64% -> 74%.

analyzingOpus Orchestratorthinking...

Preview mode — full transcript, evidence links & exports in Dashboard

Open app →

Active personas

The EconomistThe AdvocateThe Critic

Multi-Agent Debate

Spawn distinct personas that research, challenge, and refine claims from multiple perspectives.

Smart Model Routing

Opus for orchestration, Sonnet for depth, and Haiku for consensus checks to keep quality high and costs low.

Emergent Decision Matrix

Criteria and weights are generated from the debate itself, not rigid templates.

Argument Graph

Claims are tracked as standing, challenged, rebutted, or conceded in a structured graph.

Evidence-Grounded

Agents search and cite supporting or contradictory evidence before final synthesis.

Live Streaming

Watch stage progress, token usage, and routing decisions while the debate runs.

From question to verdict in three steps

Ask Your Question

Submit a decision problem with context, constraints, and desired outcome.

Run Structured Debate

Agents execute opening arguments, forced cross-exam, and rebuttal rounds.

Get Decision Verdict

Receive recommendation, confidence, dissenting view, blind spots, and source-backed rationale.

Stop asking the same question 20 times

Traditional AI gives you one perspective. You refine, re-prompt, and still miss blind spots. Verdict gets it right the first time.

The old way

Single-AI chat

One perspective at a time. You keep re-prompting to cover angles.

Should we use microservices?

Yes, microservices offer better scalability and...

Wait, what about the downsides?

Well, there is added complexity and DevOps...

Compare costs with monolith?

Monoliths are cheaper initially but harder to...

What would a CTO recommend?

A CTO might prioritize team velocity over...

Security implications?

Each microservice expands the attack surface...

How about team readiness?

You need strong DevOps culture and...

What does the data say?

Studies show 60% of migrations encounter...

+13 more re-prompts

You manually cover each angle

No one challenges weak claims

No evidence verification

20+

prompts to maybe get a balanced answer

Still missing perspectives you didn't think to ask about

The Verdict way

Multi-agent debate

You ask once. Agents argue every angle for you.

“Should we migrate to microservices?”

agents activate

Advocate

FOR

Critic

AGAINST

Analyst

DATA

cross-examining

Opening arguments

Agents present initial positions

Cross-examination

Agents challenge each other

Evidence search

Web search validates claims

Final synthesis

Structured verdict with confidence

Verdict: Proceed with caution87%

Decision matrixDissent reportEvidence linksRisk analysis

Multiple agents cover every angle

Claims are challenged and stress-tested

Real evidence backs every argument

prompt. Every angle covered automatically.

Agents find blind spots you didn't know existed