diff --git a/app/mix.exs b/app/mix.exs
index c1bdc1a..b091dd1 100644
--- a/app/mix.exs
+++ b/app/mix.exs
@@ -114,9 +114,11 @@ defmodule Firehose.MixProject do
   # not the floor.
   #
   # Measured 2026-10-26 (PostgreSQL on localhost:5432, exclusions below):
-  # 86.17%, identical via `mix test --cover` and via
-  # `mix test --cover --export-coverage default` + `mix test.coverage`, and
-  # stable across seeds. Floor is the integer below that.
+  # 86.79% after the last exclusion revision (86.17% before excluding the
+  # embed_templates wrappers and Firehose.Application), identical via
+  # `mix test --cover` and via `mix test --cover --export-coverage default` +
+  # `mix test.coverage`, and stable across seeds. Floor is the integer below
+  # that — so a floor of 86 means "at least 86.00%".
   #
   # `test.coverage` deletes stale exports before sampling: `mix test.coverage`
   # imports *every* .coverdata it finds, so leftover exports from earlier runs
@@ -133,15 +135,31 @@ defmodule Firehose.MixProject do
   #       FirehoseWeb.ErrorJSON   Phoenix.Controller.status_message_from_template/1
   #   * FirehoseWeb              the `use FirehoseWeb` macro: imports and aliases
   #                              only, nothing that executes
+  #   * embed_templates wrappers FirehoseWeb.BlogHTML,
+  #                              FirehoseWeb.UserRegistrationHTML,
+  #                              FirehoseWeb.UserSettingsHTML: each is `use` +
+  #                              `embed_templates`, so every function in them is
+  #                              generated by the macro. Their TEMPLATES are
+  #                              exercised through the controller tests; the
+  #                              wrapper modules hold no hand-written code.
+  #                              NOTE: PageHTML and UserSessionHTML are
+  #                              deliberately NOT excluded — both carry private
+  #                              helpers with real logic (blog_label,
+  #                              post_base_path, local_mail_adapter?), so their
+  #                              gaps are behavior gaps, not macro noise.
+  #   * Firehose.Application     start/2 is proven by the suite booting at all (a
+  #                              broken supervision tree fails every test), and
+  #                              config_change/3 only runs on a release hot
+  #                              upgrade, which no test can reach.
   #   * Firehose.Release         production release tasks (`mix ecto.migrate` on
   #                              Dokku); the suite never runs them
-  #   * Firehose.Checks.NoConnShadowing
-  #                              lib_dev credo check, exercised by `mix credo`
-  #                              and not by tests
   #   * test scaffolding         Firehose.DataCase, FirehoseWeb.ConnCase,
   #                              Firehose.AccountsFixtures, Firehose.Test.FakeBlog:
   #                              fixtures measure the tests, not the app
   #
+  # `Firehose.Checks.NoConnShadowing` needs no entry: lib_dev/ is no longer in
+  # elixirc_paths (credo loads it from source), so it is not measurable here.
+  #
   # Everything else counts — CoreComponents, controllers, LiveViews included:
   # that is where the behavior-test work shows up as a higher floor.
   defp test_coverage do
@@ -151,11 +169,14 @@ defmodule Firehose.MixProject do
         Firehose.Repo,
         Firehose.Mailer,
         Firehose.Release,
+        Firehose.Application,
         FirehoseWeb,
         FirehoseWeb.Telemetry,
         FirehoseWeb.ErrorHTML,
         FirehoseWeb.ErrorJSON,
-        Firehose.Checks.NoConnShadowing,
+        FirehoseWeb.BlogHTML,
+        FirehoseWeb.UserRegistrationHTML,
+        FirehoseWeb.UserSettingsHTML,
         Firehose.DataCase,
         FirehoseWeb.ConnCase,
         Firehose.AccountsFixtures,