Skip to content

YC (S26) | Open Computer History | Continuously record your company computer work, map your workflows, help you find work worth automating, and power your agents' context

fix(redact): keep benchmark prose out of doctests

Louis Beaumont682ca1b87340
diff --git a/crates/screenpipe-redact/src/adapters/rfdetr.rs b/crates/screenpipe-redact/src/adapters/rfdetr.rs
index 4ce7988c3..cadce1ae5 100644
--- a/crates/screenpipe-redact/src/adapters/rfdetr.rs
+++ b/crates/screenpipe-redact/src/adapters/rfdetr.rs
@@ -25,8 +25,10 @@
//! [email protected] on a 150-state real-app suite of planted PII that no
//! model here trained on:
//!
-//! v27 (previous): recall 92.9 %, secrets 74/76, 1.04 strays/shot
-//! v34 (this): recall 95.3 %, secrets 75/76, 1.03 strays/shot
+//! ```text
+//! v27 (previous): recall 92.9 %, secrets 74/76, 1.04 strays/shot
+//! v34 (this): recall 95.3 %, secrets 75/76, 1.03 strays/shot
+//! ```
//! Both: 0 decoy false fires. For scale, an RFDETR-**Large** trained on
//! the same data scores 95.7 % — i.e. +0.4 over this nano model, so
//! accuracy here is data-limited, not capacity-limited.