Skip to content
This repository was archived by the owner on Jul 14, 2026. It is now read-only.

🧹 Remove unused 'should_reject_hedged' method #27

Open
LeandroPG19 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

🧹 Remove unused 'should_reject_hedged' method #27

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 0 additions & 38 deletions cuba_cognitive_engine/src/engine/ewma_reward.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -237,44 +237,6 @@ impl EwmaTracker {
self.value < self.budget_mode.mcts_threshold()
}

/// V5-4: Hedged MCTS Rejection — stochastic threshold zone.
///
/// Prevents Reward Hacking (NeurIPS 2025): LLMs mathematically discover
/// deterministic thresholds and engineer outputs to surf at threshold+ε.
/// The sigmoid zone makes the exact cutoff unknowable.
///
/// Zones:
/// - diff < -0.05: deterministic reject (saves compute)
/// - diff > +0.05: deterministic accept
/// - [-0.05, +0.05]: sigmoid probability P(reject) = 1/(1+e^(20·diff))
///
/// Seed: O(1) from EWMA state — no RNG dependency.
#[allow(dead_code)]
pub fn should_reject_hedged(&self) -> bool {
if self.step_count <= 3 {
return false; // Warmup guard
}

let threshold = self.budget_mode.mcts_threshold();
let diff = self.value - threshold;

// Deterministic zones (avoid unnecessary computation)
if diff < -0.05 {
return true;
}
if diff > 0.05 {
return false;
}

// Hedging zone: sigmoid rejection probability
// At diff=0 → P=50%, diff=-0.04 → P≈68%, diff=+0.04 → P≈31%
let rejection_prob = 1.0 / (1.0 + (20.0 * diff).exp());

// O(1) deterministic seed from EWMA state — reproducible but unpredictable
let seed = (self.value * 10000.0).fract().abs();
seed < rejection_prob
}

/// V5-5: Process Advantage Verifier (PAV) — ICLR 2026.
///
/// Measures how much the current step exceeds the historical baseline
Expand Down