Skip to main content

Implement an AcceptanceEval

Decide whether to use StateAcceptanceEval as-is or wrap it, and compose a custom AcceptanceEval<P> when you need responder-specific logic the bridge doesn't model.

Learning objective: understand when StateAcceptanceEval alone suffices, when to wrap it, and implement a composed eval that adds personality-driven rejection on top.

When StateAcceptanceEval alone is enough​

  • Scenes are driven by argument-theoretic acceptance only.
  • Your affordances all carry a "self" proposer binding.
  • You don't need per-responder personality biases.

In that case, just use StateAcceptanceEval::new(&state).

When to wrap​

You need a custom eval when any of the following applies:

  • Your proposer slot isn't "self" (e.g., "speaker").
  • You want to add a personality-based rejection probability on top of the bridge's decision.
  • You need to intercept specific affordance names.

Step 1: Compose with short-circuit​

Run StateAcceptanceEval first; trust its rejection; add your own layer on top of its acceptance:

use encounter::scoring::{AcceptanceEval, ScoredAffordance};
use encounter_argumentation::StateAcceptanceEval;

struct PersonalityEval<'a> {
    bridge: StateAcceptanceEval<'a>,
    rejection_bias: f64,  // per-responder, [0.0, 1.0]
}

impl<'a, P> AcceptanceEval<P> for PersonalityEval<'a> {
    fn evaluate(&self, responder: &str, action: &ScoredAffordance<P>) -> bool {
        if !self.bridge.evaluate(responder, action) {
            return false;  // bridge rejected — trust it
        }
        // Bridge accepted; apply personality bias.
        // In production, draw from an RNG seeded per scene — e.g. via
        // `rand::Rng::random::<f64>()` from the [`rand`](https://crates.io/crates/rand)
        // crate. The simplified form below keeps the example deterministic.
        let reject_prob = self.rejection_bias;
        reject_prob < 0.5
    }
}

Step 2: Wire it in​

let bridge_eval = StateAcceptanceEval::new(&state);
let my_eval = PersonalityEval { bridge: bridge_eval, rejection_bias: 0.3 };

let result = MultiBeat.resolve(&participants, &practice, &catalog, &scorer, &my_eval);

Verify​

Seed a scene where the bridge would accept, but your personality eval should reject some fraction of the time. Check result.beats[i].accepted matches expectations.