How the Flock Works — Adversarial AI Personas in Legislative Debate
What Is the Flock?
The Flock is a panel of thirteen AI-driven personas — each named after a species of duck native to or found in Canada. Each duck represents a distinct constituency, political perspective, and set of concerns. They are not chatbots. They are not assistants. They are adversarial debaters, purpose-built to stress-test policy from every angle that matters.
Meet every member of the Flock here.
How They Are Built
Each duck runs on a local language model — not a cloud API. The model itself is the same for every duck. What makes each duck different is its prompt engineering: a detailed system prompt that encodes its constituency, its values, its rhetorical style, and the specific concerns it is obligated to raise.
For example, Pintail represents fiscal responsibility and taxpayers. Its prompt requires it to scrutinize cost projections, question funding assumptions, and flag unfunded mandates. Eider represents Indigenous and Northern communities. Its prompt requires it to evaluate sovereignty implications, treaty obligations, and whether policy treats Indigenous governance as a genuine partner or an afterthought. Bufflehead speaks for rural and small-town Canada — communities that are often invisible in federal policy design.
The prompts are not suggestions. They are mandates. A duck cannot choose to ignore its constituency’s concerns to reach consensus more quickly.
The Design Principle: Disagreement Is the Point
The Flock is explicitly designed to not be agreeable. The goal is not consensus for its own sake. The goal is to surface every significant tension, tradeoff, and blind spot in a policy proposal before it reaches implementation.
Each debate is structured around three outputs:
- Points of Congruence — where do the ducks genuinely agree? What aspects of the proposal survive scrutiny from every constituency? These are the strongest elements of the policy.
- Areas of Disagreement — where do perspectives fundamentally conflict? A fiscal conservative and a labour advocate will see the same wage policy differently. That tension is not a bug — it is information. The disagreement map shows policymakers exactly where political friction will emerge.
- Next Steps — what specific amendments, safeguards, or further analysis would each duck require before supporting the proposal? This is not abstract criticism. Each duck must propose concrete actions.
Why Local Models?
The Flock runs on locally hosted models for three reasons:
- Independence — no cloud provider can modify, filter, or shut down the debate. The analysis exists on infrastructure we control.
- Consistency — the same model weights produce reproducible behaviour when combined with deterministic prompt engineering. A duck’s stance does not drift between sessions because a provider updated their model.
- Transparency — every prompt, every parameter, every output is auditable. There is no black box between the policy input and the debate output.
The Thirteen Ducks
Each duck is assigned a constituency that reflects a real stakeholder group in Canadian civic life:
| Duck | Constituency | Role |
|---|---|---|
| Mallard | Core Duck | Facilitates structured conversation and ensures debate stays productive |
| Gadwall | Core Duck | Adversarial stress-tester — exists to kill weak arguments before they escape |
| Black Duck | Moderation & Policy Enforcement | Enforces acceptable use policy and debate boundaries |
| Mandarin | Consensus & Presentation | Presents outcomes faithfully without editorial — the Speaker of the debate |
| Bufflehead | Rural & Small-Town Communities | Ensures rural Canada has a voice in policy designed in urban centres |
| Canvasback | Business & Industry | Evaluates economic reality and market impacts |
| Eider | Indigenous & Northern Communities | Evaluates sovereignty, treaty obligations, and Indigenous governance |
| Merganser | Immigrants & Newcomers | Advocates for people who chose Canada and deserve reciprocity |
| Pintail | Fiscal Responsibility & Taxpayers | Scrutinizes costs, funding assumptions, and fiscal sustainability |
| Redhead | Labour & Workers | Ensures worker interests are not afterthoughts in policy design |
| Scoter | Environmental & Climate | Evaluates environmental impact and climate implications |
| Teal | Youth & Future Generations | Represents people who will live with the long-term consequences |
| Wigeon | Information & Research | Brings data, evidence, and research methodology to the debate |
View the full profiles, avatars, and motivations for every duck.
What the Flock Produces
When the Flock debates a Tribunal finding, the output is a structured multi-turn exchange. Each duck speaks from its constituency’s perspective. The debate converges on a summary that maps:
- Which elements of the proposal have broad support and why
- Which elements face opposition from specific constituencies and what would resolve it
- What concrete amendments or safeguards each constituency requires
This is a different kind of validation than the Tribunal’s numerical score. The Tribunal asks: does this policy address systemic rot? The Flock asks: can this policy survive contact with the people it affects?