diff --git a/crates/effects/src/fidelity.rs b/crates/effects/src/fidelity.rs
new file mode 100644
index 00000000..31243bd0
--- /dev/null
+++ b/crates/effects/src/fidelity.rs
@@ -0,0 +1,47 @@
+//! Deterministic rendered-frame comparison helpers for the fidelity corpus.
+
+use effectcraft_raster::Image;
+
+#[derive(Clone, Copy, Debug, PartialEq)]
+pub struct FrameScore {
+    pub compared_pixels: usize,
+    pub mean_absolute_error: f64,
+    pub max_error: f32,
+}
+
+/// Compare two premultiplied rendered frames. A size mismatch is reported as an error instead of
+/// silently producing a misleading score.
+pub fn compare_frames(actual: &Image, reference: &Image) -> Result<FrameScore, String> {
+    if actual.width != reference.width || actual.height != reference.height {
+        return Err(format!("frame size mismatch: {}x{} vs {}x{}", actual.width, actual.height, reference.width, reference.height));
+    }
+    let mut sum = 0.0f64;
+    let mut max = 0.0f32;
+    for (a, b) in actual.data.iter().zip(&reference.data) {
+        for (&x, &y) in a.iter().zip(b) {
+            let error = (x - y).abs();
+            sum += f64::from(error);
+            max = max.max(error);
+        }
+    }
+    let compared_pixels = actual.data.len();
+    let samples = compared_pixels.saturating_mul(4).max(1);
+    Ok(FrameScore { compared_pixels, mean_absolute_error: sum / samples as f64, max_error: max })
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn reports_pixel_error_and_rejects_size_mismatch() {
+        let mut actual = Image::filled(2, 1, [0.0, 0.0, 0.0, 1.0]);
+        let reference = Image::filled(2, 1, [0.0, 0.0, 0.0, 1.0]);
+        actual.data[1][0] = 0.5;
+        let score = compare_frames(&actual, &reference).expect("same size");
+        assert_eq!(score.compared_pixels, 2);
+        assert!((score.mean_absolute_error - 0.0625).abs() < f64::EPSILON);
+        assert_eq!(score.max_error, 0.5);
+        assert!(compare_frames(&actual, &Image::new(1, 1)).is_err());
+    }
+}
diff --git a/crates/effects/src/lib.rs b/crates/effects/src/lib.rs
index 9c0fe95b..ea9dc011 100644
--- a/crates/effects/src/lib.rs
+++ b/crates/effects/src/lib.rs
@@ -21,6 +21,7 @@ mod color2;
 mod color3;
 mod color_fx;
 pub mod controls;
+pub mod fidelity;
 mod distort;
 mod distort2;
 mod distort3;
diff --git a/docs/fidelity/README.md b/docs/fidelity/README.md
index 682ea24e..ae88186b 100644
--- a/docs/fidelity/README.md
+++ b/docs/fidelity/README.md
@@ -52,6 +52,10 @@ the 168 live AE samples.
 
 ## Still unmeasured
 
+Rendered frame comparisons now have a deterministic `effectcraft_effects::fidelity::compare_frames`
+helper. Corpus runners can use it with locally generated host/reference frames and fail on either a
+mean or maximum channel error without committing proprietary renders.
+
 Mixed interpolation sides, non-unit durations, automatic/continuous Bezier, spatial speeds and paths,
 time remapping, expressions, actual rendered motion and other effects need separate corpus cases.
 The initial probe makes no claim about those features.