diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 1af33c5..5ce9047 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -65,8 +65,9 @@ The broad areas that need help:
   compare by shape, dtype and contents with no tolerance, and datetimes and
   UUIDs are quarantined rather than normalised (issue #12). Reopening either
   means naming the real change the new policy would hide.
-- **Performance.** Recording currently costs ~8x. `sys.monitoring`
-  (Python 3.12+) should be much cheaper than wrapping functions.
+- **Performance.** Recording currently costs ~4x. Note that `sys.monitoring`
+  (Python 3.12+) is *not* the lever it looks like: profiling shows function
+  interception costs about 1% of the overhead, and serialisation the rest.
 - **Portability.** Windows is supported and tested in CI. Where `SIGALRM`
   does not exist the per-call timeout falls back to a watchdog thread, which
   cannot interrupt a call blocked inside C code.
diff --git a/README.md b/README.md
index fd07c06..55425cb 100644
--- a/README.md
+++ b/README.md
@@ -129,7 +129,7 @@ otherwise, but the gap is real.
 mode, size and a hash of the bytes). Network calls and database queries are
 still invisible.
 
-**Recording is slow.** Expect roughly an 8x slowdown on the recorded run.
+**Recording is slow.** Expect roughly a 4x slowdown on the recorded run.
 Tests that assert on wall-clock time may fail while recording.
 
 ## Measured behaviour
@@ -166,7 +166,7 @@ Start with the [good first issues][gfi]. The broad areas that need help:
 - **Side-effect capture** — intercept `requests`, `sqlalchemy`
 - **Comparators** for types that still need a policy — datetimes and UUIDs are
   settled: quarantined, not normalised (issue #12)
-- **Performance** — recording currently costs ~8x
+- **Performance** — recording currently costs ~4x
 - **Framework adapters** beyond pytest
 
 The most valuable bug report of all is a **false positive**: if `nodrift check`
diff --git a/pyproject.toml b/pyproject.toml
index e716199..298f412 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "nodrift"
-version = "0.1.3"
+version = "0.1.4"
 description = "Prove a refactor changed nothing — by replaying your tests' real inputs against both versions."
 readme = "README.md"
 requires-python = ">=3.9"
diff --git a/src/nodrift/__init__.py b/src/nodrift/__init__.py
index 147e552..a1bb2a8 100644
--- a/src/nodrift/__init__.py
+++ b/src/nodrift/__init__.py
@@ -7,7 +7,7 @@
 No model reviews the code. The verdict comes from execution.
 """
 
-__version__ = "0.1.3"
+__version__ = "0.1.4"
 
 from .compare import compare
 from .fingerprint import fingerprint
diff --git a/src/nodrift/cli.py b/src/nodrift/cli.py
index 4dfae6a..b8ab71d 100644
--- a/src/nodrift/cli.py
+++ b/src/nodrift/cli.py
@@ -191,6 +191,11 @@ def cmd_check(args: argparse.Namespace) -> int:
     return _print_report(report, args, abandoned)
 
 
+def _plural(count: int, word: str) -> str:
+    """"1 function", "2 functions" — the report is read by people."""
+    return f"{count} {word}" if count == 1 else f"{count} {word}s"
+
+
 def _print_not_covered(abandoned: list[str], verbose: bool) -> None:
     """Say which functions the verdict above does not speak for.
 
@@ -202,7 +207,7 @@ def _print_not_covered(abandoned: list[str], verbose: bool) -> None:
     """
     if not abandoned:
         return
-    print(f"  {len(abandoned)} function(s) not fully recorded "
+    print(f"  {_plural(len(abandoned), 'function')} not fully recorded "
           f"(inputs too large, or not picklable) — not covered by this check")
     if verbose:
         for name in abandoned:
@@ -224,8 +229,8 @@ def _print_report(report: dict, args: argparse.Namespace,
     quarantined = report.get("quarantined", 0)
 
     if quarantined:
-        print(f"  {quarantined} record(s) quarantined as nondeterministic "
-              f"(not compared)")
+        print(f"  {_plural(quarantined, 'record')} quarantined as "
+              f"nondeterministic (not compared)")
         # A bare count tells a user nothing about whether the function they
         # care about is among the ones that stopped being checked.
         targets = report.get("quarantined_targets") or []
@@ -241,12 +246,13 @@ def _print_report(report: dict, args: argparse.Namespace,
     _print_not_covered(abandoned, getattr(args, "verbose", False))
 
     if not differs:
-        print(f"\n  no behaviour change across {total} recorded inputs "
-              f"in {report['total_targets']} functions\n")
+        print(f"\n  no behaviour change across "
+              f"{_plural(total, 'recorded input')} in "
+              f"{_plural(report['total_targets'], 'function')}\n")
         return 0
 
     print(f"\n  {differs} of {total} recorded inputs behave differently "
-          f"({report['changed_targets']} functions)\n")
+          f"({_plural(report['changed_targets'], 'function')})\n")
     for target, counts in sorted(
         report["changed"].items(), key=lambda kv: -kv[1]["differs"]
     ):