diff --git a/.agents/phases/todo/04_reachable_share_cta/00_phase.md b/.agents/phases/complete/04_reachable_share_cta/00_phase.md
similarity index 100%
rename from .agents/phases/todo/04_reachable_share_cta/00_phase.md
rename to .agents/phases/complete/04_reachable_share_cta/00_phase.md
diff --git a/.agents/phases/todo/04_reachable_share_cta/01_one_cell_stage.md b/.agents/phases/complete/04_reachable_share_cta/01_one_cell_stage.md
similarity index 100%
rename from .agents/phases/todo/04_reachable_share_cta/01_one_cell_stage.md
rename to .agents/phases/complete/04_reachable_share_cta/01_one_cell_stage.md
diff --git a/.agents/phases/todo/04_reachable_share_cta/02_size_the_stage_for_real_viewports.md b/.agents/phases/complete/04_reachable_share_cta/02_size_the_stage_for_real_viewports.md
similarity index 100%
rename from .agents/phases/todo/04_reachable_share_cta/02_size_the_stage_for_real_viewports.md
rename to .agents/phases/complete/04_reachable_share_cta/02_size_the_stage_for_real_viewports.md
diff --git a/.agents/phases/todo/04_reachable_share_cta/03_reachable_cta_e2e.md b/.agents/phases/complete/04_reachable_share_cta/03_reachable_cta_e2e.md
similarity index 100%
rename from .agents/phases/todo/04_reachable_share_cta/03_reachable_cta_e2e.md
rename to .agents/phases/complete/04_reachable_share_cta/03_reachable_cta_e2e.md
diff --git a/.agents/phases/todo/04_reachable_share_cta/04_css_guard_and_docs.md b/.agents/phases/complete/04_reachable_share_cta/04_css_guard_and_docs.md
similarity index 100%
rename from .agents/phases/todo/04_reachable_share_cta/04_css_guard_and_docs.md
rename to .agents/phases/complete/04_reachable_share_cta/04_css_guard_and_docs.md
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.err b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.err
new file mode 100644
index 0000000..e69de29
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.md b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.md
new file mode 100644
index 0000000..6fb9a0b
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.md
@@ -0,0 +1,23 @@
+All verification is complete. The pending working-tree changes are exactly the phase 04 work (CSS, tests, the README narrow-screen line, the AGENTS.md "Two→Three pages" fix) — all accurate.
+
+## Final report
+
+**Verified (final pass — all 4 task files present in `complete/`):**
+- Gate 1: `uv run ruff check . && uv run ruff format --check .` → all checks passed, 29 files formatted
+- Gate 2: `uv run pyright` → 0 errors
+- Gate 3: `uv run pytest` → **269 passed, 4 skipped** (realdisplay-gated), coverage 95%
+- Gate 4: `MIRROR_REALDISPLAY=1 uv run pytest -m realdisplay` → **3 passed, 1 skipped** (host delivers no audio source — documented in module docstring)
+
+**Completion criteria:**
+1. `#start` elementFromPoint at the 11 required sizes — ✅ `TestIdleShareButtonReachable` sweeps 12 viewports (superset); 12/12 passed
+2. `#stop` clickable while sharing — ✅ `TestLiveStopButtonReachable` (320×568, 390×844, 1280×450) passed, incl. placeholder-hidden + livetag assertions
+3. Single-cell stage guard — ✅ `test_the_share_stage_is_one_cell` + `test_the_stage_has_a_floor_and_a_ceiling` passed; CSS has `grid-template: 1fr / 1fr`, `.stage > * { grid-area: 1 / 1 }`, `min-height: 264px`, `max-height: min(70svh, 640px)` + vh fallback, `.placeholder { max-height: 100%; overflow: auto }`, `align-content: safe center`
+4. Live video fills stage, rounded corners, `#livetag` top-left — ✅ verified by live-browser probe: video fills stage (decoded), stage radius 14px, livetag top-left, placeholder hidden
+5. README accurate, AGENTS.md current, suite 100% green — ✅
+6. Phases 01–03 unchanged — ✅ `test_keyboard_wall` + `test_audio_share` + `test_sender_fidelity`: 48 passed
+
+**Documentation:** README.md reviewed in full (page table, quickstart, all 12 config defaults vs `config.py`, CLI flags, secure-context/wayland/audio sections, kiosk key table vs `watch.js`/`share.js`, dev commands) — **consistent, no fixes needed**. AGENTS.md verified statement-by-statement (commands, routes, test names, browser/realdisplay flags) — **no stale statements; no changes needed** (the pending "Two→Three pages" fix is correct).
+
+**Decisions:** criterion 4 had no dedicated test, so verified directly with a throwaway probe reusing the e2e harness (script in /tmp, not committed). No live infrastructure touched. No defects found in prior work.
+
+**Next pending phase:** none — `todo/` is empty of other phases.
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.md.status b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.md.status
new file mode 100644
index 0000000..9766475
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.md.status
@@ -0,0 +1 @@
+ok
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.validate b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.validate
new file mode 100644
index 0000000..b687bb6
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__00_phase.a1.validate
@@ -0,0 +1,34 @@
+........................................................................ [ 26%]
+.................ssss................................................... [ 52%]
+........................................................................ [ 79%]
+.........................................................                [100%]
+=============================== warnings summary ===============================
+.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1
+  /home/ducoterra/Projects/Personal/mirror/.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1: StarletteDeprecationWarning: Using `httpx` with `starlette.testclient` is deprecated; install `httpx2` instead.
+    from starlette.testclient import TestClient as TestClient  # noqa
+
+-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
+================================ tests coverage ================================
+_______________ coverage: platform linux, python 3.14.8-final-0 ________________
+
+Name                     Stmts   Miss  Cover   Missing
+------------------------------------------------------
+src/mirror/__init__.py       3      0   100%
+src/mirror/__main__.py      72      3    96%   129, 145, 156
+src/mirror/access.py        60      0   100%
+src/mirror/app.py          185     11    94%   108, 113, 117, 182-183, 212-216, 245
+src/mirror/config.py        68      5    93%   110-111, 118-120
+src/mirror/debug.py         15     10    33%   19-28
+src/mirror/protocol.py      59      0   100%
+src/mirror/registry.py     191      3    98%   166, 180, 355
+------------------------------------------------------
+TOTAL                      653     32    95%
+=========================== short test summary info ============================
+SKIPPED [1] tests/e2e/test_x11_capture.py:281: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:321: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:364: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:446: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+269 passed, 4 skipped, 1 warning in 68.73s (0:01:08)
+All checks passed!
+29 files already formatted
+0 errors, 0 warnings, 0 informations
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.err b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.err
new file mode 100644
index 0000000..e69de29
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.md b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.md
new file mode 100644
index 0000000..a250286
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.md
@@ -0,0 +1,10 @@
+Everything is verified and green. Final report:
+
+**Phase 04 / Task 01 — one-cell stage: complete.**
+
+- **Implemented**: `src/mirror/static/app.css` `.stage` now declares `grid-template: 1fr / 1fr` (one cell) plus `.stage > * { grid-area: 1 / 1; }`, so `#preview`, `#placeholder`, `#livetag` overlap in one cell; invariant comment added at the rule. `.livetag` stays absolute, `.stage video` untouched, `share.js`/`mirror.html` untouched. Diff confined to the `/* share */` block.
+- **Instrumented browser check** (Chromium `channel="chromium"`, `http://127.0.0.1`): idle `#placeholder` topOffset now 1 @320×568 and 187.4 @1280×800 (centred, no longer the constant 151 px row); `#start` bottom 37 px below stage at 320 (was 187) and fully inside at 1280 with `elementFromPoint` = `BUTTON#start`; live state A/B-probed against old CSS — geometry byte-identical at both sizes (placeholder hidden, `#stop` clickable, video + `#livetag` unchanged → no visual regression).
+- **Screenshots**: `docs/screenshots/phase-04/` — `idle-1280x800.png`, `idle-390x844.png`, `live-1280x800.png` (visually verified: CTA centred, live video fills stage, LIVE top-left).
+- **Gates**: `uv run ruff check . && uv run ruff format --check .` clean · `uv run pyright` 0 errors · `uv run pytest` **245 passed, 4 skipped** (realdisplay-only) · `MIRROR_REALDISPLAY=1 uv run pytest -m realdisplay` **3 passed, 1 skipped** (pre-existing documented host-audio skip). Coverage 95% (floor >90%).
+- **Docs**: no README/AGENTS.md changes — README makes no claim about the share-page layout that this fix alters (no stale statement; the one pre-existing warning is a Starlette dependency deprecation, untouched).
+- **Next pending task**: `02_size_the_stage_for_real_viewports.md`.
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.md.status b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.md.status
new file mode 100644
index 0000000..9766475
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.md.status
@@ -0,0 +1 @@
+ok
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.validate b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.validate
new file mode 100644
index 0000000..0be2212
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__01_one_cell_stage.a1.validate
@@ -0,0 +1,34 @@
+...................................................................ssss. [ 28%]
+........................................................................ [ 57%]
+........................................................................ [ 86%]
+.................................                                        [100%]
+=============================== warnings summary ===============================
+.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1
+  /home/ducoterra/Projects/Personal/mirror/.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1: StarletteDeprecationWarning: Using `httpx` with `starlette.testclient` is deprecated; install `httpx2` instead.
+    from starlette.testclient import TestClient as TestClient  # noqa
+
+-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
+================================ tests coverage ================================
+_______________ coverage: platform linux, python 3.14.8-final-0 ________________
+
+Name                     Stmts   Miss  Cover   Missing
+------------------------------------------------------
+src/mirror/__init__.py       3      0   100%
+src/mirror/__main__.py      72      3    96%   129, 145, 156
+src/mirror/access.py        60      0   100%
+src/mirror/app.py          185     11    94%   108, 113, 117, 182-183, 212-216, 245
+src/mirror/config.py        68      5    93%   110-111, 118-120
+src/mirror/debug.py         15     10    33%   19-28
+src/mirror/protocol.py      59      0   100%
+src/mirror/registry.py     191      3    98%   166, 180, 355
+------------------------------------------------------
+TOTAL                      653     32    95%
+=========================== short test summary info ============================
+SKIPPED [1] tests/e2e/test_x11_capture.py:281: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:321: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:364: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:446: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+245 passed, 4 skipped, 1 warning in 66.61s (0:01:06)
+All checks passed!
+28 files already formatted
+0 errors, 0 warnings, 0 informations
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.err b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.err
new file mode 100644
index 0000000..e69de29
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.md b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.md
new file mode 100644
index 0000000..e56845a
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.md
@@ -0,0 +1,16 @@
+All green. Final state verified — everything complete:
+
+**Task 02 complete.** Report:
+
+**Implemented**
+- `app.css` (entirely inside the `/* share */` block): `.stage` floor `min-height: 264px`, ceiling `max-height: min(70vh, 640px)` + `min(70svh, 640px)` (vh fallback), `align-content: safe center`; `.stage .placeholder` escape hatch `max-height: 100%; overflow: auto`; `aspect-ratio: 4/3` at ≤900px, `auto` + 16px padding at ≤480px
+- Two fixes beyond the literal task: (1) media queries placed *after* the base rules — before that, equal-specificity source order made them lose; (2) `width: 100%` on `.stage` — a stretched grid item with aspect-ratio gets its *width* recomputed from a max-height-clamped height (stage shrank to 1024px, 534px gap beside the panel); a definite width keeps the full column and the video letterboxes, per the task's ASSUMPTION
+- `tests/e2e/test_live_mirror.py::TestStageSizing`: the 6 measured failure viewports (button inside stage + inside viewport, floor ≥263, stage bottom on-screen) + 1920×1080 ceiling (≤641px, Session panel reachable) — in `test_live_mirror.py` since `test_share_layout.py` belongs to task 03
+
+**Before → after** (measured, Chromium over 127.0.0.1): 320×568 stage 180→264, `#start` clipped→263–315 inside; 900×480 542→336 (on-screen); 1280×450 564→315; 1920×1080 964→640, panel bottom 499 < 1080. Test verified **red before** the fix (6/7 viewports + ceiling fail) and green after.
+
+**Gates**: `uv run ruff check . && uv run ruff format --check .` clean · `uv run pyright` 0 errors · `uv run pytest` 252 passed, 4 skipped (realdisplay-gated) · `MIRROR_REALDISPLAY=1 uv run pytest -m realdisplay` 3 passed, 1 skipped (documented env-gated audio) · `pytest tests/test_app.py -m "not e2e"` 68 passed · coverage 95% (no new src code; phase is CSS-only per 00_phase.md)
+
+**Docs**: README.md — added the sharer-facing narrow-screen line (320px viewport, CTA always on screen). AGENTS.md — no stale statements, no edit.
+
+**Next pending task**: `03_reachable_cta_e2e` — the 12-viewport `elementFromPoint` sweep in `tests/e2e/test_share_layout.py`, idle + live `#stop`.
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.md.status b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.md.status
new file mode 100644
index 0000000..9766475
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.md.status
@@ -0,0 +1 @@
+ok
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.validate b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.validate
new file mode 100644
index 0000000..fa41c87
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__02_size_the_stage_for_real_viewports.a1.validate
@@ -0,0 +1,34 @@
+........................................................................ [ 28%]
+..ssss.................................................................. [ 56%]
+........................................................................ [ 84%]
+........................................                                 [100%]
+=============================== warnings summary ===============================
+.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1
+  /home/ducoterra/Projects/Personal/mirror/.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1: StarletteDeprecationWarning: Using `httpx` with `starlette.testclient` is deprecated; install `httpx2` instead.
+    from starlette.testclient import TestClient as TestClient  # noqa
+
+-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
+================================ tests coverage ================================
+_______________ coverage: platform linux, python 3.14.8-final-0 ________________
+
+Name                     Stmts   Miss  Cover   Missing
+------------------------------------------------------
+src/mirror/__init__.py       3      0   100%
+src/mirror/__main__.py      72      3    96%   129, 145, 156
+src/mirror/access.py        60      0   100%
+src/mirror/app.py          185     12    94%   108, 113, 116-117, 182-183, 212-216, 245
+src/mirror/config.py        68      5    93%   110-111, 118-120
+src/mirror/debug.py         15     10    33%   19-28
+src/mirror/protocol.py      59      0   100%
+src/mirror/registry.py     191      3    98%   166, 180, 355
+------------------------------------------------------
+TOTAL                      653     33    95%
+=========================== short test summary info ============================
+SKIPPED [1] tests/e2e/test_x11_capture.py:281: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:321: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:364: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:446: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+252 passed, 4 skipped, 1 warning in 66.55s (0:01:06)
+All checks passed!
+28 files already formatted
+0 errors, 0 warnings, 0 informations
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.err b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.err
new file mode 100644
index 0000000..e69de29
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.md b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.md
new file mode 100644
index 0000000..f812484
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.md
@@ -0,0 +1,17 @@
+All green. Report:
+
+**Implemented**
+- `tests/e2e/test_share_layout.py`: `#start` hit-test sweep over the 12 task viewports (verbatim `elementFromPoint` probe; asserts `disabled`, `hit_is_button` with `hit` named in the failure message, `in_stage`, `in_viewport`, `no_h_scroll`) plus live-state sweep at 320×568 / 390×844 / 1280×450 asserting `#stop` hit-testable, `#placeholder` hidden, `#livetag` visible inside the stage — reusing `base_url`/`browser`/`start_sharing` from `tests/e2e/conftest.py`.
+- Pre-fix verification (temporary revert to HEAD CSS, then restored): **7/12 idle viewports fail** — 320×568 (`hit: H2`), 360×640 (`hit: #name`), 390×844 (`hit: LABEL`), 430×932 (`hit: H2`), 480×800 (`hit: MAIN`), 900×480 (`hit: null`), 1280×450 (`hit: null`); 700×900, 900×700, 1024×600, 1280×800, 1920×1080 pass. Matches the 00_phase diagnosis (480×800's hit landed on `MAIN` this run vs `#share-link` in the original table — same clip, culprit named in output). Please fold these failing widths into the phase commit message.
+
+**Gates (all green)**
+- `uv run ruff check . && uv run ruff format --check .` — All checks passed; 29 files formatted
+- `uv run pyright` — 0 errors
+- `uv run pytest` — 267 passed, 4 skipped (realdisplay gate), 95% coverage
+- `MIRROR_REALDISPLAY=1 uv run pytest -m realdisplay` — 3 passed, 1 skipped (host offers no audio source; pre-existing, documented in the test)
+
+**Docs** — no doc impact: README already documents the 320px phone behaviour (tasks 01–02); this task is test-only, no stale AGENTS.md facts.
+
+**Decisions** — git stash unavailable per harness rules, so pre-fix check used a file swap of `app.css` (restored immediately); live probe parameterised on selector since `#stop` lives in the Session panel, not the stage (`in_stage` asserted for `#start` only).
+
+**Next pending task** — `04_css_guard_and_docs` (`.agents/phases/todo/04_reachable_share_cta/04_css_guard_and_docs.md`).
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.md.status b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.md.status
new file mode 100644
index 0000000..9766475
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.md.status
@@ -0,0 +1 @@
+ok
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.validate b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.validate
new file mode 100644
index 0000000..ddcb6de
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__03_reachable_cta_e2e.a1.validate
@@ -0,0 +1,34 @@
+........................................................................ [ 26%]
+.................ssss................................................... [ 53%]
+........................................................................ [ 79%]
+.......................................................                  [100%]
+=============================== warnings summary ===============================
+.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1
+  /home/ducoterra/Projects/Personal/mirror/.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1: StarletteDeprecationWarning: Using `httpx` with `starlette.testclient` is deprecated; install `httpx2` instead.
+    from starlette.testclient import TestClient as TestClient  # noqa
+
+-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
+================================ tests coverage ================================
+_______________ coverage: platform linux, python 3.14.8-final-0 ________________
+
+Name                     Stmts   Miss  Cover   Missing
+------------------------------------------------------
+src/mirror/__init__.py       3      0   100%
+src/mirror/__main__.py      72      3    96%   129, 145, 156
+src/mirror/access.py        60      0   100%
+src/mirror/app.py          185     11    94%   108, 113, 117, 182-183, 212-216, 245
+src/mirror/config.py        68      5    93%   110-111, 118-120
+src/mirror/debug.py         15     10    33%   19-28
+src/mirror/protocol.py      59      0   100%
+src/mirror/registry.py     191      3    98%   166, 180, 355
+------------------------------------------------------
+TOTAL                      653     32    95%
+=========================== short test summary info ============================
+SKIPPED [1] tests/e2e/test_x11_capture.py:281: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:321: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:364: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:446: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+267 passed, 4 skipped, 1 warning in 69.49s (0:01:09)
+All checks passed!
+29 files already formatted
+0 errors, 0 warnings, 0 informations
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.err b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.err
new file mode 100644
index 0000000..e69de29
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.md b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.md
new file mode 100644
index 0000000..9b67ccb
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.md
@@ -0,0 +1,13 @@
+Everything is verified and green. Final report:
+
+**Phase 04, Task 04 — complete. All gates green.**
+
+- **`TestStylesheet::test_the_share_stage_is_one_cell`** — asserts `.stage` declares `grid-template: 1fr / 1fr` (or explicit rows/columns of 1fr) *and* `.stage > *` carries `grid-area: 1 / 1`; docstring records the 300×150 idle-video / 151px-stack / `overflow: hidden` clip reason.
+- **`TestStylesheet::test_the_stage_has_a_floor_and_a_ceiling`** — `.stage` must keep `min-height` and `max-height`; any rule combining `overflow: hidden` with `aspect-ratio` needs `min-height` in the same block (the general statement of the bug).
+- **`test_hidden_always_means_gone`** — `.placeholder` added to the `[hidden]`-toggled component list, with the inline comment corrected so it stays accurate.
+- **Revert verification (by hand)**: deleting the grid lines fails the one-cell guard; deleting `min-height: 264px` fails the floor check; deleting both `max-height` lines fails the ceiling check. CSS restored; all 5 `TestStylesheet` tests green.
+- **README.md L10** — added the `/share` alias (served by `app.py`, asserted by `test_pages_render`, previously undocumented). The narrow-screen user-facing line (L39) was already present from an earlier task; full-file consistency pass found no other gaps (ports, config defaults, key tables, kiosk commands all verified against code/docs).
+- **AGENTS.md L44** — fixed the one stale fact: "Two pages" → "Three pages and one WebSocket" (the index page exists and is tested). No narrative added; "coverage ~95%" re-measured at 95%, still accurate.
+- **PLAN.md** — untouched, but L204's "per-leg `maxBitrate` cap (default 12 Mbit/s)" is contradicted by reality: phase 02 raised the default to 25 Mbit/s (`config.py` L15, README L72). Reported here for the user to amend; I did not edit it.
+- **Gates**: `uv run ruff check . && uv run ruff format --check .` clean · `uv run pyright` 0 errors · `uv run pytest` 269 passed / 4 skipped (realdisplay flag-gated) · `MIRROR_REALDISPLAY=1 uv run pytest -m realdisplay` 3 passed / 1 skipped (host has no audio source — documented platform limitation). Phases 01–03 e2e suites (keyboard, sharpness, audio) all still green.
+- **Next pending task**: none — task 04 was the last task of phase 04; the phase is ready for its final validation pass.
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.md.status b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.md.status
new file mode 100644
index 0000000..9766475
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.md.status
@@ -0,0 +1 @@
+ok
diff --git a/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.validate b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.validate
new file mode 100644
index 0000000..e5588c8
--- /dev/null
+++ b/.agents/reports/04_reachable_share_cta/04_reachable_share_cta__04_css_guard_and_docs.a1.validate
@@ -0,0 +1,34 @@
+........................................................................ [ 26%]
+.................ssss................................................... [ 52%]
+........................................................................ [ 79%]
+.........................................................                [100%]
+=============================== warnings summary ===============================
+.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1
+  /home/ducoterra/Projects/Personal/mirror/.venv/lib64/python3.14/site-packages/fastapi/testclient.py:1: StarletteDeprecationWarning: Using `httpx` with `starlette.testclient` is deprecated; install `httpx2` instead.
+    from starlette.testclient import TestClient as TestClient  # noqa
+
+-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
+================================ tests coverage ================================
+_______________ coverage: platform linux, python 3.14.8-final-0 ________________
+
+Name                     Stmts   Miss  Cover   Missing
+------------------------------------------------------
+src/mirror/__init__.py       3      0   100%
+src/mirror/__main__.py      72      3    96%   129, 145, 156
+src/mirror/access.py        60      0   100%
+src/mirror/app.py          185     12    94%   108, 113, 116-117, 182-183, 212-216, 245
+src/mirror/config.py        68      5    93%   110-111, 118-120
+src/mirror/debug.py         15     10    33%   19-28
+src/mirror/protocol.py      59      0   100%
+src/mirror/registry.py     191      3    98%   166, 180, 355
+------------------------------------------------------
+TOTAL                      653     33    95%
+=========================== short test summary info ============================
+SKIPPED [1] tests/e2e/test_x11_capture.py:281: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:321: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:364: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+SKIPPED [1] tests/e2e/test_x11_capture.py:446: set MIRROR_REALDISPLAY=1 to exercise the real X11 capture backend
+269 passed, 4 skipped, 1 warning in 68.64s (0:01:08)
+All checks passed!
+29 files already formatted
+0 errors, 0 warnings, 0 informations
diff --git a/AGENTS.md b/AGENTS.md
index 9be6342..f964470 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -41,7 +41,7 @@ is a roster and a signaling relay. Linux is the reference platform.
 | Rule | Why it exists |
 |---|---|
 | **No external assets.** No CDN, remote font, remote icon, `@import`, or analytics | LAN/offline correctness. `tests/test_app.py::TestClientScripts::test_nothing_is_loaded_from_a_cdn` fails the build if you add one |
-| **No framework, no bundler** in `src/mirror/static/` | Two pages and one WebSocket. `node --check` is the entire frontend toolchain |
+| **No framework, no bundler** in `src/mirror/static/` | Three pages and one WebSocket. `node --check` is the entire frontend toolchain |
 | **Media never passes through the server** | The server relays JSON only. Anything that puts frames in the Python process violates A3 |
 | **`Registry` stays pure** — no sockets, no FastAPI, no `await` | It returns `list[Envelope]`. Transport lives in `Hub`. This is why the policy has 100%-ish unit coverage |
 | **camelCase on the wire, snake_case in Python**, via `alias_generator=to_camel` | A silently-dropped `displaySurface` was a real bug we shipped. Any new multi-word field needs an alias test |
diff --git a/README.md b/README.md
index 2714527..a0acb7c 100644
--- a/README.md
+++ b/README.md
@@ -7,7 +7,7 @@ your network:
 
 | Page         | URL                | What it does                                                            |
 | ------------ | ------------------ | ----------------------------------------------------------------------- |
-| **Share**    | `/mirror`          | Picks a window, screen, or tab via the OS-native picker and mirrors it.  |
+| **Share**    | `/mirror` (also `/share`) | Picks a window, screen, or tab via the OS-native picker and mirrors it.  |
 | **Watch**    | `/watch`           | Shows every screen currently being mirrored; click a tile to take the stage. |
 | **Index**    | `/`                | Signpost + live count of mirrors.                                        |
 
@@ -36,6 +36,10 @@ Linux is the primary target. On Linux the native picker is X11/Wayland's own
 (Chromium uses `xdg-desktop-portal` on Wayland/GNOME/KDE, and its built-in X11
 picker on Xorg), so nothing has to be installed on the client side.
 
+The share page works on a phone as well as a desktop: `/mirror` is usable
+down to a 320 px-wide viewport, and the preview keeps its shape but never
+shrinks below the controls it holds, so the Share button is always on screen.
+
 The target *client* setup for an always-on display is Fedora Silverblue running
 GNOME Kiosk — see [`docs/kiosk-silverblue.md`](docs/kiosk-silverblue.md).
 
diff --git a/docs/screenshots/phase-04/idle-1280x800.png b/docs/screenshots/phase-04/idle-1280x800.png
new file mode 100644
index 0000000..2446e48
Binary files /dev/null and b/docs/screenshots/phase-04/idle-1280x800.png differ
diff --git a/docs/screenshots/phase-04/idle-390x844.png b/docs/screenshots/phase-04/idle-390x844.png
new file mode 100644
index 0000000..ce0d477
Binary files /dev/null and b/docs/screenshots/phase-04/idle-390x844.png differ
diff --git a/docs/screenshots/phase-04/live-1280x800.png b/docs/screenshots/phase-04/live-1280x800.png
new file mode 100644
index 0000000..e7a5970
Binary files /dev/null and b/docs/screenshots/phase-04/live-1280x800.png differ
diff --git a/src/mirror/static/app.css b/src/mirror/static/app.css
index 19ba615..70b13d0 100644
--- a/src/mirror/static/app.css
+++ b/src/mirror/static/app.css
@@ -225,27 +225,88 @@ a { color: var(--accent); }
   align-items: start;
 }
 
-@media (max-width: 900px) {
-  .share { grid-template-columns: 1fr; }
-}
-
+/* The stage is ONE cell. Its children — the <video>, the idle placeholder,
+   the LIVE badge — share that cell, so the placeholder overlays the video the
+   way the markup intends instead of stacking into a row below it.
+   Invariant for the next reader: a child added to #preview-stage shares that
+   cell and must never push another child out of view. `overflow: hidden` is
+   there to clip the live video to the rounded corners; a box whose content no
+   longer fits a fixed-aspect stage is the size rule's job, not this one's
+   (the floor/ceiling below). */
 .stage {
   position: relative;
+  /* Definite (100%), not stretched-auto: with an aspect-ratio, a stretched
+     item whose height is clamped by max-height gets its *width* recomputed
+     from the clamped height (the stage would shrink and leave a gap beside
+     the panel). A definite width keeps the full column; the video then
+     letterboxes inside (object-fit: contain) instead. */
+  width: 100%;
   aspect-ratio: 16 / 10;
+  /* Floor and ceiling: the aspect ratio is the *preferred* shape, and these
+     bound it. 264px is the measured idle placeholder at 320px wide + slack —
+     a placeholder taller than the stage scrolls (see below) instead of being
+     trapped under `overflow: hidden`. The ceiling: the preview is a preview;
+     on 1920x1080 the ratio alone wanted 964px and pushed the Session panel
+     off-screen (the video letterboxes inside, so nothing is lost). vh first
+     as the silent fallback where svh is unsupported — the ceiling is then
+     never load-bearing on an old browser; dvh deliberately avoided, a mobile
+     URL bar must not resize the stage mid-share. */
+  min-height: 264px;
+  max-height: min(70vh, 640px);
+  max-height: min(70svh, 640px);
   background: #05070a;
   border: 1px solid var(--line);
   border-radius: var(--radius);
   overflow: hidden;
   display: grid;
+  grid-template: 1fr / 1fr; /* one cell, not one row per child */
   place-items: center;
+  /* The escape hatch's other half: if a row ever overflows the box, align to
+     the start instead of centring into a clip at both edges (the
+     centred-overflow trap that hid half the Share button). */
+  align-content: safe center;
 }
 
+/* Every child lives in that one cell: #preview, #placeholder and #livetag
+   overlap instead of stacking. DOM order paints the placeholder above the video
+   (no z-index); .livetag stays position: absolute, so grid-area is inert for
+   it. */
+.stage > * { grid-area: 1 / 1; }
+
 .stage video { width: 100%; height: 100%; object-fit: contain; display: block; }
 
-.stage .placeholder { text-align: center; padding: 24px; color: var(--fg-dim); }
+/* Escape hatch: the placeholder may never be taller than the stage; if its
+   content outgrows it, it scrolls. Paired with `safe center` above, content
+   that is too tall is always reachable, whichever way the geometry falls. */
+.stage .placeholder {
+  text-align: center;
+  padding: 24px;
+  color: var(--fg-dim);
+  max-height: 100%;
+  overflow: auto;
+}
 .stage .placeholder h2 { color: var(--fg); margin: 14px 0 6px; font-size: 20px; }
 .stage .placeholder p { margin: 0 0 18px; font-size: 14px; }
 
+/* Narrow-width shape rules come *after* the base rules above: same
+   specificity, so source order decides, and a phone must not be left with
+   the wide-screen ratio or the wide-screen padding. Single-column below
+   900px, and the stage stops asking for a wide-screen shape there (4:3
+   instead of 16:10): the floor does the real work, the ratio only keeps
+   the box from looking like a letterbox. */
+@media (max-width: 900px) {
+  .share { grid-template-columns: 1fr; }
+  .stage { aspect-ratio: 4 / 3; }
+}
+
+/* A phone (≤480px) gets a stage as tall as its content rather than any
+   letterbox, and the placeholder's padding buys back 16px. min-height still
+   does the real work in both cases. */
+@media (max-width: 480px) {
+  .stage { aspect-ratio: auto; }
+  .stage .placeholder { padding: 16px; }
+}
+
 .livetag {
   position: absolute;
   top: 12px;
diff --git a/tests/e2e/test_live_mirror.py b/tests/e2e/test_live_mirror.py
index 96243e4..751a5f8 100644
--- a/tests/e2e/test_live_mirror.py
+++ b/tests/e2e/test_live_mirror.py
@@ -8,6 +8,7 @@ from __future__ import annotations
 
 import re
 import time
+from typing import ClassVar
 
 import pytest
 from playwright.sync_api import expect
@@ -294,6 +295,93 @@ class TestChromeVisibility:
         expect(sharer.locator("#start")).not_to_be_visible()
         expect(sharer.locator("#placeholder")).not_to_be_visible()
 
+
+class TestStageSizing:
+    """The stage has a floor, a ceiling, and an escape hatch.
+
+    ``aspect-ratio: 16 / 10`` alone made the stage 180 px tall at 320 px wide
+    (the idle placeholder needed ~240 px there) and 543 px tall in a 480 px
+    window (the Share button ended up below the fold). These tests pin the
+    six shapes measured in ``00_phase.md``: at each one ``#start`` must be
+    fully inside *both* the stage box and the viewport — inside the stage or
+    ``overflow: hidden`` clips it, and inside the viewport or a finger can
+    never reach it. The 1080p case pins the ceiling: the Session panel must
+    be reachable without scrolling past a 964 px preview.
+    """
+
+    # The three clip cases (button clipped out of the stage) and the two
+    # below-the-fold cases (stage taller than the viewport), plus 360x640,
+    # the fourth measured narrow shape.
+    FAILURE_VIEWPORTS: ClassVar[list[tuple[int, int]]] = [
+        (320, 568),
+        (360, 640),
+        (390, 844),
+        (480, 800),
+        (900, 480),
+        (1280, 450),
+    ]
+
+    STAGE_PROBE = """() => {
+      const s = document.querySelector('#preview-stage').getBoundingClientRect();
+      const r = document.querySelector('#start').getBoundingClientRect();
+      return {
+        stage_h: s.height,
+        stage_bottom: s.bottom,
+        start: {top: r.top, bottom: r.bottom, right: r.right},
+        in_stage: r.bottom <= s.bottom + 1 && r.top >= s.top - 1,
+        in_viewport: r.top >= 0 && r.bottom <= innerHeight && r.right <= innerWidth,
+        stage_in_viewport: s.bottom <= innerHeight + 1,
+      };
+    }"""
+
+    @pytest.mark.parametrize("viewport", FAILURE_VIEWPORTS)
+    def test_the_share_button_is_inside_the_stage_and_the_viewport(
+        self, base_url, browser, request, viewport
+    ):
+        context = browser.new_context(viewport={"width": viewport[0], "height": viewport[1]})
+        request.addfinalizer(context.close)
+        page = context.new_page()
+        page.goto(f"{base_url}/mirror", wait_until="domcontentloaded")
+        expect(page.locator("#start")).to_be_visible(timeout=10_000)
+        result = page.evaluate(self.STAGE_PROBE)
+        message = f"{viewport[0]}x{viewport[1]}: {result}"
+        # The floor (264px, the measured idle placeholder + slack) and the
+        # ceiling (min(70svh, 640px)) must both hold: the box is never so
+        # short that the CTA clips out, and never so tall that it leaves
+        # the viewport.
+        assert result["stage_h"] >= 263, f"floor failed: {message}"
+        assert result["in_stage"], f"#start is clipped by the stage: {message}"
+        assert result["in_viewport"], f"#start is below the fold: {message}"
+        # The stage itself must also fit: on the two short-window shapes the
+        # aspect-ratio box used to run 543/564 px tall in a 480/450 px window,
+        # leaving only a scroll to reach the bottom of the page.
+        assert result["stage_in_viewport"], f"the stage itself overflows the viewport: {message}"
+
+    def test_the_ceiling_keeps_the_session_panel_reachable_at_1080p(
+        self, base_url, browser, request
+    ):
+        """1920x1080: the aspect ratio alone wanted a 964 px stage, which
+        pushed the Session panel off-screen. The 640 px ceiling must keep
+        the panel on screen — reachable without scrolling."""
+        context = browser.new_context(viewport={"width": 1920, "height": 1080})
+        request.addfinalizer(context.close)
+        page = context.new_page()
+        page.goto(f"{base_url}/mirror", wait_until="domcontentloaded")
+        expect(page.locator("#start")).to_be_visible(timeout=10_000)
+        result = page.evaluate(
+            """() => {
+              const s = document.querySelector('#preview-stage').getBoundingClientRect();
+              const p = document.querySelector('.panel').getBoundingClientRect();
+              return {stage_h: s.height, stage_w: s.width, panel_bottom: p.bottom};
+            }"""
+        )
+        assert result["stage_h"] <= 641, (
+            f"the 640px ceiling failed (aspect ratio alone wanted 964px): {result}"
+        )
+        assert result["panel_bottom"] <= 1080, (
+            f"the Session panel is not reachable without scrolling: {result}"
+        )
+
     def test_exit_only_appears_on_the_tile_that_is_on_stage(self, device):
         sharer = device("/mirror")
         start_sharing(sharer)
diff --git a/tests/e2e/test_share_layout.py b/tests/e2e/test_share_layout.py
new file mode 100644
index 0000000..658e1cc
--- /dev/null
+++ b/tests/e2e/test_share_layout.py
@@ -0,0 +1,174 @@
+"""End-to-end: the Share CTA is a *hit*, at every realistic viewport.
+
+The bug this file pins: ``#preview-stage`` was a two-row grid (the idle
+``<video>`` reserved a ~150 px row, the placeholder stacked below it and ran
+into ``overflow: hidden``), and its height was an unbounded function of its
+width (180 px tall at 320 px wide; 543 px in a 480 px window, pushing the
+button below the fold). On the reference host,
+``document.elementFromPoint`` at the button centre named ``H2`` at 320x568,
+``#name`` at 360x640, ``LABEL`` at 390x844, ``#share-link`` at 480x800, and
+``null`` at 900x480 / 1280x450 — a finger or cursor would never have hit the
+button.
+
+The assertion is a **hit test, not a size check**:
+``document.elementFromPoint(cx, cy).closest("#start")`` must be the button.
+That is the only in-browser statement that means *a finger or a cursor would
+hit it*, and it is exactly the check that failed. (A ``page.click`` would start
+a real capture and need teardown; the hit test is the same claim with no side
+effects. The live-state sweep below still exercises a real ``start_sharing()``
+once per shape, so the interaction path is not untested.)
+
+320 px is the floor of "realistic": below it the page is expected to scroll,
+not to fit. The containment comparisons carry a +/-1 px tolerance for
+sub-pixel layout at fractional device pixel ratios.
+"""
+
+from __future__ import annotations
+
+from typing import ClassVar
+
+import pytest
+from playwright.sync_api import expect
+
+from .conftest import start_sharing
+
+pytestmark = [pytest.mark.e2e, pytest.mark.timeout(180)]
+
+# The sweep: the clip cases (button clipped out of the stage), the
+# below-the-fold cases (stage taller than the viewport), and the control
+# shapes in between — exactly the list in 03_reachable_cta_e2e.md.
+VIEWPORTS = [
+    (320, 568),
+    (360, 640),
+    (390, 844),
+    (430, 932),
+    (480, 800),  # the clip cases
+    (700, 900),
+    (900, 700),
+    (1024, 600),
+    (1280, 800),
+    (1920, 1080),
+    (900, 480),
+    (1280, 450),  # the below-the-fold cases
+]
+
+# The probe that reproduced the bug — kept verbatim so the next reader can
+# re-run the diagnosis. ``hit`` named the culprit during diagnosis: ``H2`` at
+# 320x568, ``#name`` at 360x640, ``LABEL`` at 390x844, ``#share-link`` at
+# 480x800, ``null`` at 900x480 / 1280x450.
+HIT_PROBE = """() => {
+  const b = document.querySelector("#start");
+  const s = document.querySelector("#preview-stage").getBoundingClientRect();
+  const r = b.getBoundingClientRect();
+  const hit = document.elementFromPoint(r.x + r.width / 2, r.y + r.height / 2);
+  return {
+    disabled: b.disabled,
+    in_stage: r.bottom <= s.bottom + 1 && r.top >= s.top - 1,
+    in_viewport: r.bottom <= innerHeight && r.right <= innerWidth && r.top >= 0,
+    hit_is_button: !!hit && !!hit.closest("#start"),
+    hit: hit ? hit.id || hit.tagName : null,
+    no_h_scroll: document.documentElement.scrollWidth <= innerWidth,
+  };
+}"""
+
+# The same hit test, parameterised on the button. Used for #stop while a share
+# is live: #stop lives in the Session panel, not the stage, so "inside the
+# stage" does not apply to it — clickability is the hit test plus the viewport.
+STOP_PROBE = """(selector) => {
+  const b = document.querySelector(selector);
+  const r = b.getBoundingClientRect();
+  const hit = document.elementFromPoint(r.x + r.width / 2, r.y + r.height / 2);
+  return {
+    disabled: b.disabled,
+    in_viewport: r.bottom <= innerHeight && r.right <= innerWidth && r.top >= 0,
+    hit_is_button: !!hit && !!hit.closest(selector),
+    hit: hit ? hit.id || hit.tagName : null,
+    no_h_scroll: document.documentElement.scrollWidth <= innerWidth,
+  };
+}"""
+
+# The live-state invariants: the placeholder is hidden, the LIVE badge is
+# visible and sits top-left inside the stage.
+LIVE_STATE_PROBE = """() => {
+  const s = document.querySelector("#preview-stage").getBoundingClientRect();
+  const l = document.querySelector("#livetag").getBoundingClientRect();
+  return {
+    placeholder_hidden: document.querySelector("#placeholder").hidden,
+    livetag_visible: !document.querySelector("#livetag").hidden,
+    livetag_in_stage:
+      l.left >= s.left - 1 && l.top >= s.top - 1 &&
+      l.right <= s.right + 1 && l.bottom <= s.bottom + 1,
+  };
+}"""
+
+
+def _open_page(base_url, browser, request, width: int, height: int):
+    context = browser.new_context(viewport={"width": width, "height": height})
+    request.addfinalizer(context.close)
+    page = context.new_page()
+    page.goto(f"{base_url}/mirror", wait_until="domcontentloaded")
+    expect(page.locator("#net")).to_have_attribute("data-state", "online", timeout=10_000)
+    return page
+
+
+class TestIdleShareButtonReachable:
+    """``#start`` must be hit-testable at every realistic viewport, idle.
+
+    A layout change that stacks the stage's children again (or un-bounds the
+    box's height) fails at least one of these: the clipped shapes fail
+    ``in_stage``/``hit_is_button`` (the old bug), the short shapes fail
+    ``in_viewport`` (the old bug), and ``no_h_scroll`` catches a layout that
+    hides the button behind horizontal scroll instead.
+    """
+
+    @pytest.mark.parametrize("viewport", VIEWPORTS)
+    def test_the_share_button_is_hit_testable(self, base_url, browser, request, viewport):
+        page = _open_page(base_url, browser, request, viewport[0], viewport[1])
+        expect(page.locator("#start")).to_be_visible(timeout=10_000)
+        result = page.evaluate(HIT_PROBE)
+        label = f"{viewport[0]}x{viewport[1]}: {result}"
+        # A disabled button would make the hit test vacuous.
+        assert result["disabled"] is False, f"#start is disabled: {label}"
+        # The point that a finger or cursor would land on must be the button.
+        assert result["hit_is_button"], (
+            f"#start is not hit-testable, elementFromPoint named {result['hit']!r}: {label}"
+        )
+        # Inside the stage or `overflow: hidden` clips it; inside the viewport
+        # or a finger can never reach it; no horizontal scroll to reach it.
+        assert result["in_stage"], f"#start is clipped by the stage: {label}"
+        assert result["in_viewport"], f"#start is outside the viewport: {label}"
+        assert result["no_h_scroll"], f"the button is behind a horizontal scroll: {label}"
+
+
+class TestLiveStopButtonReachable:
+    """The live state must not regress: while sharing, ``#stop`` stays
+    hit-testable, ``#placeholder`` is hidden, and ``#livetag`` sits top-left
+    inside the stage. The three shapes are the ones measured live in
+    ``00_phase.md`` (``hit_stop: true`` after the fix).
+    """
+
+    LIVE_VIEWPORTS: ClassVar[list[tuple[int, int]]] = [
+        (320, 568),
+        (390, 844),
+        (1280, 450),
+    ]
+
+    @pytest.mark.parametrize("viewport", LIVE_VIEWPORTS)
+    def test_stop_is_hit_testable_while_sharing(self, base_url, browser, request, viewport):
+        page = _open_page(base_url, browser, request, viewport[0], viewport[1])
+        start_sharing(page)
+
+        stop = page.evaluate(STOP_PROBE, "#stop")
+        label = f"{viewport[0]}x{viewport[1]}: {stop}"
+        assert stop["disabled"] is False, f"#stop is disabled: {label}"
+        assert stop["hit_is_button"], (
+            f"#stop is not hit-testable, elementFromPoint named {stop['hit']!r}: {label}"
+        )
+        assert stop["in_viewport"], f"#stop is outside the viewport: {label}"
+        assert stop["no_h_scroll"], f"#stop is behind a horizontal scroll: {label}"
+
+        state = page.evaluate(LIVE_STATE_PROBE)
+        label = f"{viewport[0]}x{viewport[1]}: {state}"
+        assert state["placeholder_hidden"], f"#placeholder is still shown while sharing: {label}"
+        assert state["livetag_visible"], f"#livetag is not visible while sharing: {label}"
+        assert state["livetag_in_stage"], f"#livetag is not inside the stage: {label}"
diff --git a/tests/test_app.py b/tests/test_app.py
index f3a70cf..cd20c27 100644
--- a/tests/test_app.py
+++ b/tests/test_app.py
@@ -291,8 +291,11 @@ class TestStylesheet:
         assert re.search(r"\[hidden\]\s*\{[^}]*display:\s*none\s*!important", css), (
             "app.css must declare a global `[hidden] { display: none !important }`"
         )
-        # Components that set `display` and are toggled with `hidden` in JS.
-        for selector in (".btn", ".livetag", ".icon-btn", ".toast", ".net"):
+        # Components toggled with `hidden` in JS. All but `.placeholder` set
+        # their own `display`, which would outrank the UA's `[hidden]` without
+        # the global rule; `.placeholder` is guarded because its block is the
+        # escape hatch that keeps its content inside the stage.
+        for selector in (".btn", ".livetag", ".icon-btn", ".toast", ".net", ".placeholder"):
             assert selector in css, f"{selector} is toggled by `hidden`; it lives in app.css"
 
     def test_no_rule_reintroduces_a_conflicting_display_for_hidden(self):
@@ -337,6 +340,83 @@ class TestStylesheet:
                 "without a pointer — a mouseless kiosk would have no tile controls"
             )
 
+    def _rules(self) -> list[tuple[str, str]]:
+        r"""app.css as ``(selector, body)`` pairs, comments stripped.
+
+        The flat ``[^{}]+\{[^{}]*\}`` pass treats an ``@media`` header as a
+        selector and the first child's header as its body; the child rules
+        that follow still come back as their own pairs, which is all these
+        guards need — every assertion is about a top-level block (``.stage``,
+        ``.stage > *``) or about properties that must sit in the *same* block
+        (``overflow``/``aspect-ratio``/``min-height`` together).
+        """
+        css = re.sub(r"/\*.*?\*/", "", (STATIC / "app.css").read_text(), flags=re.S)
+        return [(sel.strip(), body) for sel, body in re.findall(r"([^{}]+)\{([^{}]*)\}", css)]
+
+    def test_the_share_stage_is_one_cell(self):
+        """The stage's children overlap on purpose.
+
+        In plain flow an idle ``<video>`` is intrinsically 300x150, so the
+        placeholder (and the Share button in it) starts 151px below the video
+        at *every* width and runs out of the ``overflow: hidden`` box.
+        ``grid-template: 1fr / 1fr`` makes the stage one cell; ``grid-area:
+        1 / 1`` on every child is what puts them all in it. Either half
+        without the other stacks the rows back.
+        """
+        rules = self._rules()
+        stage_bodies = [body for sel, body in rules if sel == ".stage"]
+        assert stage_bodies, ".stage rule missing from app.css"
+        assert any(
+            re.search(r"grid-template\s*:\s*1fr\s*/\s*1fr", body)
+            or (
+                re.search(r"grid-template-rows\s*:\s*1fr", body)
+                and re.search(r"grid-template-columns\s*:\s*1fr", body)
+            )
+            for body in stage_bodies
+        ), (
+            ".stage must declare an explicit single cell (grid-template: 1fr / 1fr, "
+            "or grid-template-rows/columns of 1fr) — otherwise its children stack into rows"
+        )
+        child_bodies = [body for sel, body in rules if sel == ".stage > *"]
+        assert child_bodies and any(
+            re.search(r"grid-area\s*:\s*1\s*/\s*1", body) for body in child_bodies
+        ), (
+            "app.css must put every .stage child in that one cell (grid-area: 1 / 1) — "
+            "without it the placeholder stacks below the 300x150 video and the Share "
+            "button is clipped by overflow: hidden"
+        )
+
+    def test_the_stage_has_a_floor_and_a_ceiling(self):
+        """An aspect-derived box with no floor is a clip.
+
+        ``aspect-ratio`` alone made the stage 180px tall at 320px wide (the
+        idle placeholder needs 240px there) and 543px tall in a 480px-tall
+        window (the Share button below the fold). The floor and the ceiling
+        make the ratio the *preferred* shape instead of a hard derivation, and
+        the second check is the general statement of the bug: any rule that
+        keeps ``overflow: hidden`` on an aspect-ratio box must carry a
+        ``min-height`` in the same block, or the hidden clip silently swallows
+        content again.
+        """
+        rules = self._rules()
+        stage_bodies = [body for sel, body in rules if sel == ".stage"]
+        assert stage_bodies, ".stage rule missing from app.css"
+        assert any(re.search(r"min-height\s*:", body) for body in stage_bodies), (
+            ".stage must keep its floor (min-height)"
+        )
+        assert any(re.search(r"max-height\s*:", body) for body in stage_bodies), (
+            ".stage must keep its ceiling (max-height)"
+        )
+        for sel, body in rules:
+            if re.search(r"overflow[^:]*:\s*[^;]*hidden", body) and re.search(
+                r"aspect-ratio\s*:", body
+            ):
+                assert re.search(r"min-height\s*:", body), (
+                    f"{sel} combines `overflow: hidden` with `aspect-ratio` but has no "
+                    "`min-height` in the same block — an aspect-derived box with no "
+                    "floor clips its content silently"
+                )
+
 
 class TestAddressGate:
     """mirror has no accounts; the address is the access control."""