diff --git a/CHANGELOG.md b/CHANGELOG.md
new file mode 100644
index 0000000..c70ce85
--- /dev/null
+++ b/CHANGELOG.md
@@ -0,0 +1,51 @@
+# Changelog
+
+Notable changes per release. Versions follow [semantic versioning](https://semver.org/),
+with the caveat that 0.x means the API may still move between minor versions.
+
+## [0.1.2] — 2026-07-25
+
+### Changed
+
+- `convex::paginated_query` now splits an oversized page instead of resetting
+  pagination. A page's range is pinned by its query journal, so inserts into
+  that range make it grow; previously a page that outgrew the server's read
+  limits (`pageStatus: "SplitRequired"`) reset the whole session, dropping every
+  loaded page and snapping the list back to the first one. It is now split in
+  two at the server's `splitCursor`, with both halves bounded by explicit
+  `cursor`/`endCursor` and swapped in only once both have loaded, so loaded
+  pages stay put. Pages are also split on `SplitRecommended`, or once a page
+  grows past twice `initial_num_items`. This matches convex-js's
+  `usePaginatedQuery`.
+- A page the server reports as `SplitRequired` may be missing part of its range,
+  so `paginated_snapshot::results` now stops before it and the status returns to
+  `loading_more` (or `loading_first_page`) until a split repairs it. Previously
+  such a page was published, which could leave a gap in the middle of the list.
+  Also matches `usePaginatedQuery`.
+- An incomplete page the server gives no `splitCursor` for is left alone rather
+  than reset. Its subscription stays live, so a later result that fits the read
+  limits repairs it. A query too heavy to ever return a complete page leaves the
+  list loading, with no error — as in convex-js, which has no recovery path for
+  that case either.
+
+No public API change: `paginated_query`, `paginated_snapshot`, and
+`pagination_status` are unchanged.
+
+## [0.1.1] — 2026-07-14
+
+### Added
+
+- FetchContent consumption snippet in the README.
+
+## [0.1.0] — 2026-07-14
+
+First release. Realtime sync client over the Convex WebSocket protocol (live
+query subscriptions, ordered mutations with read-your-writes, actions, auth,
+reconnection with jittered backoff), the `convex::paginated_query` helper,
+`convex::http_client` for one-shot calls, file storage upload/download, and full
+Convex value fidelity including Int64, Bytes, and non-finite floats. The
+protocol core is sans-IO; a bundled IXWebSocket transport is optional.
+
+[0.1.2]: https://github.com/Potionify/convex-cpp/compare/v0.1.1...v0.1.2
+[0.1.1]: https://github.com/Potionify/convex-cpp/compare/v0.1.0...v0.1.1
+[0.1.0]: https://github.com/Potionify/convex-cpp/releases/tag/v0.1.0
diff --git a/README.md b/README.md
index e5354b8..b735b8d 100644
--- a/README.md
+++ b/README.md
@@ -61,7 +61,7 @@ include(FetchContent)
 FetchContent_Declare(
     convex
     GIT_REPOSITORY https://github.com/Potionify/convex-cpp.git
-    GIT_TAG v0.1.0
+    GIT_TAG v0.1.2
 )
 set(CONVEX_WITH_IXWEBSOCKET ON)  # bundled desktop transport; omit if you supply your own
 FetchContent_MakeAvailable(convex)
diff --git a/scripts/bump_version.cmake b/scripts/bump_version.cmake
index e890792..dad6766 100644
--- a/scripts/bump_version.cmake
+++ b/scripts/bump_version.cmake
@@ -1,5 +1,6 @@
-# Bumps the library version (include/convex/version.h), commits, and creates
-# the matching annotated tag. Run from anywhere inside the repo:
+# Bumps the library version (include/convex/version.h and the README's
+# FetchContent pin), commits, and creates the matching annotated tag. Run from
+# anywhere inside the repo:
 #
 #   cmake -DBUMP=patch -P scripts/bump_version.cmake    # 0.1.0 -> 0.1.1
 #   cmake -DBUMP=minor -P scripts/bump_version.cmake    # 0.1.0 -> 0.2.0
@@ -10,9 +11,18 @@
 # Or use the wrappers (no quoting needed): scripts/bump_version.bat patch
 # on Windows, scripts/bump_version.sh patch on Linux/macOS.
 #
-# Then publish (this triggers the release workflow):
+# Publishing. The release workflow triggers on the tag, and git does not push
+# tags by default — neither does a plain `git push` nor the Push button in a
+# GUI client. A tag left behind that way looks like a release that silently
+# never happened. So this script sets push.followTags in the repo's local
+# config, which makes every subsequent push carry annotated tags reachable
+# from the commits it sends. Publish with either:
 #
-#   git push origin main --tags
+#   cmake -DBUMP=patch -DPUSH=ON -P scripts/bump_version.cmake   # bump + push
+#   git push origin main                                        # after a bump
+#
+# Both send the tag. `git push origin main --tags` still works and is what to
+# use if an earlier tag was stranded locally.
 
 if(NOT DEFINED BUMP)
     message(FATAL_ERROR "Pass -DBUMP=major|minor|patch|x.y.z (the -D must come before -P)")
@@ -62,7 +72,24 @@ foreach(part MAJOR MINOR PATCH)
 endforeach()
 file(WRITE "${version_h}" "${content}")
 
-execute_process(COMMAND git -C "${repo_root}" add include/convex/version.h
+# The README's FetchContent snippet pins a tag. Left to a human it goes stale
+# silently, and a stale pin hands every new user an old library.
+set(readme "${repo_root}/README.md")
+file(READ "${readme}" readme_content)
+if(NOT readme_content MATCHES "GIT_TAG v[0-9]+\\.[0-9]+\\.[0-9]+")
+    message(FATAL_ERROR "No 'GIT_TAG vX.Y.Z' pin found in ${readme}; update it by hand "
+                        "or fix this script's pattern")
+endif()
+string(REGEX REPLACE "GIT_TAG v[0-9]+\\.[0-9]+\\.[0-9]+" "GIT_TAG v${new_version}"
+       readme_content "${readme_content}")
+file(WRITE "${readme}" "${readme_content}")
+
+# Make an ordinary push carry the tag. Without this the release workflow waits
+# on a tag that never arrives, and the failure is silent.
+execute_process(COMMAND git -C "${repo_root}" config --local push.followTags true
+                COMMAND_ERROR_IS_FATAL ANY)
+
+execute_process(COMMAND git -C "${repo_root}" add include/convex/version.h README.md
                 COMMAND_ERROR_IS_FATAL ANY)
 execute_process(COMMAND git -C "${repo_root}" commit -m "Release v${new_version}"
                 COMMAND_ERROR_IS_FATAL ANY)
@@ -70,4 +97,11 @@ execute_process(COMMAND git -C "${repo_root}" tag -a "v${new_version}" -m "v${ne
                 COMMAND_ERROR_IS_FATAL ANY)
 
 message(STATUS "${old_version} -> ${new_version}, committed and tagged v${new_version}")
-message(STATUS "Publish with: git push origin main --tags")
+
+if(PUSH)
+    execute_process(COMMAND git -C "${repo_root}" push origin HEAD "v${new_version}"
+                    COMMAND_ERROR_IS_FATAL ANY)
+    message(STATUS "Pushed the release commit and tag v${new_version}")
+else()
+    message(STATUS "Publish with: git push origin main   (push.followTags carries the tag)")
+endif()
diff --git a/src/paginated.cpp b/src/paginated.cpp
index 661a528..b23d676 100644
--- a/src/paginated.cpp
+++ b/src/paginated.cpp
@@ -234,7 +234,7 @@ struct paginated_impl : std::enable_shared_from_this<paginated_impl> {
         s.first.key = ++next_key;
         // The first half starts where the original page started — NOT at the
         // beginning of the query. convex-js sent a null cursor here, which
-        // duplicated every item before the page (get-convex/convex-js#54981).
+        // duplicated every item before the page (convex-js commit 7ceee3e).
         s.first.cursor = orig->cursor;
         s.first.end_cursor = split_cursor;
         s.second.key = ++next_key;
diff --git a/tests/paginated_test.cpp b/tests/paginated_test.cpp
index d1a751d..5a5b788 100644
--- a/tests/paginated_test.cpp
+++ b/tests/paginated_test.cpp
@@ -472,7 +472,7 @@ TEST(Paginated, IncompletePageHidesItselfButNotThePagesBeforeIt) {
 }
 
 TEST(Paginated, SplitFirstHalfStartsAtTheOriginalPagesCursor) {
-    // Regression guard for get-convex/convex-js#54981: convex-js sent a null
+    // Regression guard for convex-js commit 7ceee3e: convex-js sent a null
     // cursor for the first half, so splitting any page but the first re-read
     // the list from the beginning and duplicated everything before it.
     harness h;