diff --git a/ABI.md b/ABI.md
index 80ef80f..37c4977 100644
--- a/ABI.md
+++ b/ABI.md
@@ -426,8 +426,8 @@ Canonical forms:
   full-change one-run delta is itself exactly 46093 bytes (= the keyframe size),
   and the inclusive `≥` makes two conformant encoders agree on shipping the
   canonical keyframe at that exact boundary. The wire cost is bounded at exactly
-  46093 bytes, never worse (+1.50× over v1's 30720-byte floor — the one honest
-  regression, paid only on full repaints / bootstrap / resume).
+  46093 bytes, never worse (+1.50× over v1's 30720-byte floor, the only size
+  regression in v2, paid only on full repaints / bootstrap / resume).
 
 **Validation (host-side, drop-not-fatal).** A validator SHALL check length
 consistency with `runCount`, every run in-bounds, runs ascending and
diff --git a/rust/README.md b/rust/README.md
index f2a642e..fdb92fa 100644
--- a/rust/README.md
+++ b/rust/README.md
@@ -109,7 +109,7 @@ table is the dictionary:
 | `acct.Store().Get/Set/Delete` | `r.kv_get/kv_set/kv_delete(&p, …)` (`Option` instead of `(v, ok)`) |
 | `f.SetGrapheme(...)` | `f.set_grapheme(...)` (≤3 code points per cell, refused otherwise) |
 
-## Toolchain notes (the honest list)
+## Toolchain notes
 
 - **Target**: `wasm32-wasip1`. The crate type that works is `cdylib` — it
   produces a WASI *reactor* (exports `memory` + `_initialize`; no `_start`),