kixelated · Nov 7, 2026 · Nov 7, 2026 · Nov 7, 2026 · Nov 7, 2026
diff --git a/BENCHMARKS.md b/BENCHMARKS.md
diff --git a/benchmarks/data/desktop.json b/benchmarks/data/desktop.json
diff --git a/benchmarks/data/kixtop.json b/benchmarks/data/kixtop.json
diff --git a/benchmarks/desktop-dark.svg b/benchmarks/desktop-dark.svg
diff --git a/benchmarks/desktop.svg b/benchmarks/desktop.svg
diff --git a/benchmarks/gaps.toml b/benchmarks/gaps.toml
@@ -23,11 +23,6 @@ operations = ["rsa2048-pkcs1-sha256-sign"]
 on = ["nixos"]
 reason = "Noise: other work kept 5 CPUs busy during this run, in which PSS signing, the same exponentiation, measured 0.95× graviola's time. Interleaved runs on `nixos` the same morning measured rind's PKCS#1 signing at 436 to 446 µs against graviola's 432 to 542 µs and ring's 459 to 568 µs. The next refresh rechecks it."
 
-[[gap]]
-operations = ["rustls-bulk/aes128-gcm", "xchacha20poly1305/1350B"]
-on = ["desktop"]
-reason = "Noise: other work kept 5 CPUs busy during this run. The run before it measured 0.99× ring's and 0.97× aws-lc-rs's time, and neither shows on another host. The next refresh rechecks both."
-
 [[gap]]
 operations = ["ed25519-verify"]
 on = ["ovh-iad", "ovh-sea"]
@@ -53,7 +48,6 @@ operations = [
   "aes128-gcm/32B",
   "aes128-gcm/1350B",
   "aes128-gcm-open/32B",
-  "aes128-gcm-open/1350B",
   "aes256-gcm/32B",
   "aes256-gcm/1350B",
   "aes256-gcm-open/32B",
@@ -71,9 +65,4 @@ operations = [
   "hkdf-sha256/expand",
 ]
 on = ["kixtop"]
-reason = "DIT: the M4 has it, and rind turns it on and off around every call that handles a secret, which costs about 35 to 55 ns a call there; the other libraries never set it. Keeping it per call was decided with the maintainer from [#387](https://github.com/moq-dev/rind/pull/387)'s findings. Holding `rind::cpu::Dit::hold()` ([#431](https://github.com/moq-dev/rind/pull/431)) across a batch of calls, as MoQ does, pays it once and closes every one of these gaps: rind-bench's `rind-dit-held` rows time that, and a run at [`ec4b720d`](https://github.com/moq-dev/rind/commit/ec4b720daadd41f8ab7f3669f262c678b1af8c89), noisy with other work on 1.1 CPUs, measured each within 10% of the fastest library, SHA-256 at 1350 B (1.10× ring's) and HMAC-SHA-256 at 64 B (1.10× RustCrypto's) at the edge."
-
-[[gap]]
-operations = ["xchacha20poly1305/32B"]
-on = ["kixtop"]
-reason = "Noise: a rerun of the row on the same commit measured 1.04× aws-lc-rs's time, as did the run before this one, on a commit that changed only x86_64 code. The next refresh rechecks it."
+reason = "DIT: the M4 has it, and rind turns it on and off around every call that handles a secret, which costs about 30 to 55 ns a call there; the other libraries never set it. Keeping it per call was decided with the maintainer from [#387](https://github.com/moq-dev/rind/pull/387)'s findings. Holding `rind::cpu::Dit::hold()` ([#431](https://github.com/moq-dev/rind/pull/431)) across a batch of calls, as MoQ does, pays it once and closes all but one of these gaps: the rind (DIT held) column times that, and on [`9e7f6a5a`](https://github.com/moq-dev/rind/commit/9e7f6a5a9ea59a0cf1251c24af6c5d4f21c2b0fd), the run this report shows, every one measured within 10% of the fastest library except HMAC-SHA-256 at 64 B, at 1.10× RustCrypto's (109.7 ns against 99.4 ns)."
diff --git a/benchmarks/headline-dark.svg b/benchmarks/headline-dark.svg
diff --git a/benchmarks/headline.svg b/benchmarks/headline.svg
diff --git a/benchmarks/kixtop-dark.svg b/benchmarks/kixtop-dark.svg
diff --git a/benchmarks/kixtop.svg b/benchmarks/kixtop.svg
diff --git a/benchmarks/overview-dark.svg b/benchmarks/overview-dark.svg
diff --git a/benchmarks/overview.svg b/benchmarks/overview.svg
diff --git a/quest/m1/README.md b/quest/m1/README.md
@@ -11,5 +11,4 @@ Bun's fastest asymmetric operations run on rind, ahead of Bun writing its own.
 - [Bun runs its fastest asymmetric crypto on rind](/quest/m1/bun.md) - Ed25519, X25519, P-384, and P-521 in Bun on rind, before Bun writes its own
 - [The launch post can say what building rind cost in AI tokens, time, and money](/quest/m1/ai-usage.md) - tokens, agent and maintainer hours, and API-price cost, before the post
 - [rind's launch blog post is published](/quest/m1/launch-post.md) - the Bun PR waits on it
-- [The bench report has a quiet Arrow Lake run and kixtop's build cost](/quest/m1/bench-apple-m4.md) - a fresh desktop column and M4 build costs before the launch post
 - [rind covers MoQ](/quest/m1/moq/README.md) - the primitives, CPU check, armv7 backend, and `rind-*` adapters MoQ needs, verified and released
diff --git a/quest/m1/bench-apple-m4.md b/quest/m1/bench-apple-m4.md