Struct ActivationPatcher 

pub struct ActivationPatcher {
    pub config: PatchConfig,
}

Activation patcher – simulates the clean/corrupted dual-run experiment.

Fields

config: PatchConfig

Implementations

impl ActivationPatcher

pub fn new(config: PatchConfig) -> Self

pub fn corrupt(&self, prompt: &str, strategy: CorruptionStrategy) -> String

Corrupt a prompt according to the configured strategy.

In a real deployment this would call the tokeniser; here we operate on whitespace-split tokens as a transport-layer approximation.

pub fn run_patch_experiment( &self, clean_prompt: &str, corrupted_prompt: &str, target_token: &str, clean_logprobs: &[HashMap<String, f64>], corrupted_logprobs: &[HashMap<String, f64>], ) -> PatchingResult

Simulate a patching experiment using pre-computed logprob tables.

clean_logprobs[pos] and corrupted_logprobs[pos] should map target tokens to their log-probability at each position. In a live integration these are obtained from two separate API calls with logprobs=true.