diff --git a/.agents/skills/record-tui-gif/actop-demo.tape b/.agents/skills/record-tui-gif/actop-demo.tape
index 1c728b7..8d7760e 100644
--- a/.agents/skills/record-tui-gif/actop-demo.tape
+++ b/.agents/skills/record-tui-gif/actop-demo.tape
@@ -1,48 +1,67 @@
 # actop hero capture — the README/hero GIF.
-# One continuous, gapless take in the stack layout that tells the whole story:
-#   off-camera:  start actop -> l (stack layout)
-#   on-camera:   stack + braille dots  ->  g (block charts)  ->  p (process table)
-#                ->  t (cycle theme, several times) -> t back to dark
-# Layout switch is off-camera (grid would leave a gap at this height); the glyph,
-# process, and theme toggles are on-camera and stay full because stack fills the height.
-# Render from the repo root:  vhs .claude/skills/record-tui-gif/actop-demo.tape
+# One continuous, gapless take in the DEFAULT grid layout that tells the whole
+# v1.9.x story:
+#   off-camera:  start actop (grid is the default — no layout switch needed)
+#   on-camera:   grid + braille dots  ->  g (block charts)  ->  c (per-core panels)
+#                ->  p (process table + watt attribution)  ->  t (theme cycle)
+#
+# Why grid rather than stack: since v1.9.0 the dashboard has seven sections, and
+# grid is the only layout that shows the mirrored Network / Disk pair side by
+# side without a ~57-row column. Stack would need Height ~1240 here, which
+# leaves a large gap under Power once the process table is on.
+#
+# Width is load-bearing: `p` splits the dashboard against the process table, and
+# below ~96 cols the grid preset auto-degrades to stack. At 1400px (~165 cols)
+# the dashboard drops to ~80 cols and silently degrades mid-take; 1800px
+# (~212 cols) leaves it ~122 cols, so grid survives the whole recording.
+#
+# Render from the repo root:  vhs .agents/skills/record-tui-gif/actop-demo.tape
 # (or via record.sh, which drives a GPU workload so the gauges move).
 
 Output images/actop-demo.gif
 
-Require actop
+# Record the REPO build, not whatever `actop` happens to be on PATH. Typing a
+# bare `actop` picks up the Homebrew/pipx install, which silently recorded a
+# stale TUI (a v1.9.x hero was captured off a v1.8.0 binary before this was
+# caught — no Network/Disk sections at all). Require the venv entry point.
+Require .venv/bin/actop
 
 Set Shell "zsh"
 Set FontSize 16
-# 1400px @ FontSize 16 ≈ 165 cols — well above the ~96-col grid-degrade floor.
-Set Width 1400
-# Sized to the stack layout's natural column height so there's no bottom gap.
-Set Height 1050
+# 1800px @ FontSize 16 ≈ 212 cols — keeps the dashboard above the 96-col
+# grid-degrade floor even after `p` splits it against the process table.
+Set Width 1800
+# Sized to the tallest on-camera state (grid + per-core panels + process table
+# ≈ 42 dashboard rows + chrome), so nothing clips and no gap opens under Power.
+Set Height 930
 Set Theme "Dracula"
 Set Padding 20
 Set Framerate 24
 
-# --- Warm up + switch to stack layout OFF-camera ---
+# --- Warm up OFF-camera so the first visible frame already has live gauges ---
 Hide
-Type "actop"
+Type ".venv/bin/actop"
 Enter
-Sleep 6s
-Type "l"
-Sleep 1s
+Sleep 7s
 Show
 
-# --- 1) stack layout, classic braille-dot charts ---
-Sleep 5s
+# --- 1) default grid layout, classic braille-dot charts ---
+#     Seven sections: P-CPU | E-CPU, GPU·ANE | Memory, Network | Disk, Power.
+Sleep 6s
 
 # --- 2) toggle chart glyph dots -> modern solid blocks ---
 Type "g"
 Sleep 5s
 
-# --- 3) turn on the process table (needs a poll cycle to populate) ---
+# --- 3) reveal the per-core panels inside both cluster boxes ---
+Type "c"
+Sleep 5s
+
+# --- 4) turn on the process table (needs a poll cycle to populate) ---
 Type "p"
 Sleep 8s
 
-# --- 4) cycle through themes live ---
+# --- 5) cycle through themes live ---
 Type "t"
 Sleep 2s
 Type "t"
diff --git a/.github/workflows/publish-pypi.yml b/.github/workflows/publish-pypi.yml
index 25bbfd6..03a7644 100644
--- a/.github/workflows/publish-pypi.yml
+++ b/.github/workflows/publish-pypi.yml
@@ -1,8 +1,29 @@
 name: publish-pypi
 run-name: "publish-pypi | ${{ github.event_name == 'workflow_dispatch' && inputs.tag_name || github.ref_name }} | ${{ github.sha }}"
 
-# Publishes the actop package to PyPI on a version tag push using PyPI
-# Trusted Publishing (OIDC) — no API token is stored in the repo.
+# Publishes the actop package to PyPI using PyPI Trusted Publishing (OIDC) —
+# no API token is stored in the repo.
+#
+# MANUAL ONLY — this workflow does NOT run on tag push. Publishing is a
+# deliberate step taken AFTER the tag and its CI have landed, because the
+# automatic form was self-defeating: a tag push started the publish at the same
+# moment the tag appeared, so the run raced the release it was meant to ship and
+# never published against a settled tag. Making it a dispatch that takes the tag
+# as an input removes the ordering problem entirely — the tag already exists and
+# is named explicitly.
+#
+# Publish with (after `scripts/tag_release.sh` and green CI):
+#   gh workflow run publish-pypi.yml --ref vX.Y.Z -f tag_name=vX.Y.Z
+#
+# `--ref` must be the TAG, not a branch: the `pypi` environment's deployment
+# policy only admits refs matching `v*`, so a dispatch from main is rejected
+# before the job starts. Pointing it at the tag satisfies the policy and keeps
+# the "only a tagged commit ever reaches PyPI" guarantee that the old automatic
+# trigger provided.
+#
+# Local fallback, from a clean checkout of the tag (needs a PyPI API token —
+# OIDC is only available to the workflow above):
+#   python -m build && python -m twine upload --skip-existing dist/*
 #
 # One-time PyPI-side setup (do this once, signed in as the project owner):
 #   1. https://pypi.org/manage/account/publishing/  ->  "Add a new pending publisher"
@@ -15,9 +36,6 @@ run-name: "publish-pypi | ${{ github.event_name == 'workflow_dispatch' && inputs
 #   trusted publisher for the project.
 
 on:
-  push:
-    tags:
-      - "v*"
   workflow_dispatch:
     inputs:
       tag_name:
diff --git a/CHANGELOG.md b/CHANGELOG.md
index dbfdb0d..74f51ed 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,31 @@ This project follows a Keep a Changelog-style format and uses version tags for r
 
 ## [Unreleased]
 
+## [1.9.2] - 2026-08-12
+
+### Changed
+
+- **PyPI publishing left the tag flow — it is now a manual, post-tag step.**
+  `publish-pypi.yml` no longer triggers on `push: tags`; it is
+  `workflow_dispatch` only, taking the tag as an input:
+  `gh workflow run publish-pypi.yml --ref vX.Y.Z -f tag_name=vX.Y.Z`.
+  The automatic form was self-defeating — the tag push started the publish at
+  the moment the tag appeared, so the run raced the release it was meant to
+  ship rather than publishing a settled one. Dispatching against a tag that
+  already exists removes the ordering problem entirely, and because the `pypi`
+  environment admits only `v*` refs, the "only a tagged commit reaches PyPI"
+  guarantee is preserved rather than weakened. OIDC trusted publishing is
+  unchanged; no token is introduced.
+- **Hero GIF re-recorded for the v1.9.x dashboard** (`images/actop-demo.gif`),
+  and the recording tape now pins the binary it records. The tape typed a bare
+  `actop`, which resolved to whatever was installed on `PATH` — it had been
+  capturing a stale Homebrew build with no Network/Disk sections at all. It now
+  requires `.venv/bin/actop`. The capture moved to the default `grid` layout at
+  1800x930 (stack pushed the two new I/O boxes below the fold) and covers the
+  mirrored Network/Disk pair, per-core panels, both glyph modes, the process
+  table with watt attribution, and the theme cycle.
+- Cover page: net/disk feature card now describes the mirrored charts.
+
 ## [1.9.1] - 2026-08-12
 
 ### Fixed
diff --git a/CLAUDE.md b/CLAUDE.md
index 6d6e870..58d8ea4 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -48,7 +48,7 @@ This file is the single source of truth for repository guidelines, used by Claud
 
 ## Release Process
 
-`main` is **PR-only** (branch protection + `.githooks/pre-commit` redaction check and `.githooks/pre-push` guard; run `git config core.hooksPath .githooks` once). The Homebrew formula lives in the separate tap repo `binlecode/homebrew-actop` (not this repo); CI syncs it on tag and publishes to PyPI via OIDC. CI/CD mechanics are owned by the workflow files themselves (`.github/workflows/main-ci.yml`, `release-formula.yml`, `publish-pypi.yml`, `guard-release.yml`) — each carries its one-time setup inline. `guard-release.yml` is the post-push integrity gate: every tag push triggers a scan that verifies every `v*` tag has a corresponding GitHub Release object, catching the "silent link breakage" that caused the v0.8.7–v1.6.6 release-object gap.
+`main` is **PR-only** (branch protection + `.githooks/pre-commit` redaction check and `.githooks/pre-push` guard; run `git config core.hooksPath .githooks` once). The Homebrew formula lives in the separate tap repo `binlecode/homebrew-actop` (not this repo); CI syncs it on tag. **PyPI publishing is NOT part of the tag flow** — it is a manual step taken after the tag and its CI have landed (`gh workflow run publish-pypi.yml --ref vX.Y.Z -f tag_name=vX.Y.Z`). The old tag-triggered publish raced the tag that started it and shipped against an unsettled ref; a dispatch naming the tag explicitly has no such ordering problem. CI/CD mechanics are owned by the workflow files themselves (`.github/workflows/main-ci.yml`, `release-formula.yml`, `publish-pypi.yml`, `guard-release.yml`) — each carries its one-time setup inline. `guard-release.yml` is the post-push integrity gate: every tag push triggers a scan that verifies every `v*` tag has a corresponding GitHub Release object, catching the "silent link breakage" that caused the v0.8.7–v1.6.6 release-object gap.
 
 **RULE — cutting a release REQUIRES a GitHub Release, not just a tag.** A release is cut with `scripts/tag_release.sh`, which pushes the git tag **and** creates the GitHub Release for it via `gh release create` (release notes pulled from the matching `CHANGELOG.md` section). A bare `git tag` + push is **not** a release: it never surfaces on the `/releases` page or as `Latest`, and leaves the Releases history silently behind the code. Never cut a release by hand-tagging alone — if the `gh release create` step fails, finish it manually (see playbooks below) before calling the release cut done.
 
@@ -74,10 +74,15 @@ git checkout main && git pull --ff-only
 # 4. Cut the release — tag + GitHub Release (RULE: tag alone is not a release)
 scripts/tag_release.sh "X.Y.Z"
 
-# 5. Monitor CI — wait for release-formula and publish-pypi to pass
+# 5. Monitor CI — wait for release-formula and guard-release to pass
 gh run list -R binlecode/actop --limit 10
 
-# 6. Verify
+# 6. Publish to PyPI — MANUAL, only after the tag exists and CI is green.
+#    --ref must be the tag: the `pypi` environment admits only `v*` refs.
+gh workflow run publish-pypi.yml --ref "vX.Y.Z" -f tag_name="vX.Y.Z"
+gh run watch -R binlecode/actop "$(gh run list -R binlecode/actop --workflow publish-pypi.yml --limit 1 --json databaseId --jq '.[0].databaseId')"
+
+# 7. Verify
 pipx install actop && actop --version                # PyPI
 brew update && brew upgrade binlecode/actop/actop    # Homebrew (tap repo synced)
 ```
@@ -96,12 +101,12 @@ brew update && brew upgrade binlecode/actop/actop    # Homebrew (tap repo synced
   Never leave a release cut as a bare tag — verify it appears on `/releases` before moving on.
 - **`release-formula` fails with tag/version mismatch:** tag `vX.Y.Z` doesn't match `pyproject.toml` in the tag commit. Fix in a new PR, merge, then create a **new** tag (never reuse the old tag).
 - **`release-formula` fails at tap checkout/push:** missing or under-scoped `HOMEBREW_TAP_TOKEN`. Confirm the secret exists on `binlecode/actop` and the PAT has **Contents: Read/write** on `binlecode/homebrew-actop`. Re-run the failed job.
-- **`publish-pypi` fails with OIDC / trusted-publisher error:** verify the trusted publisher on pypi.org matches repo `binlecode/actop`, workflow `publish-pypi.yml`, environment `pypi`; confirm the run is from a `v*` tag (the `pypi` environment's deploy policy is tag-only). `skip-existing` makes re-runs safe. **Break-glass:** publish that version manually with the token-driven Flow B (`twine upload --skip-existing`), then restore OIDC for subsequent releases.
+- **`publish-pypi` fails with OIDC / trusted-publisher error:** verify the trusted publisher on pypi.org matches repo `binlecode/actop`, workflow `publish-pypi.yml`, environment `pypi`; confirm the dispatch used `--ref vX.Y.Z` (the `pypi` environment's deploy policy is tag-only, so a dispatch from `main` is rejected before the job starts). `skip-existing` makes re-runs safe. **Break-glass:** publish that version manually with the token-driven Flow B (`twine upload --skip-existing`), then restore OIDC for subsequent releases.
 - **Emergency rerun without resource refresh:** `workflow_dispatch` on `release-formula` with `refresh_resources=false` as a temporary workaround; follow up with a normal run (`refresh_resources=true`).
 
 ### PyPI publishing — two flows
 
-OIDC Trusted Publishing is the default (no stored secret; the `pypi` environment on `binlecode/actop` is tag-only, and the workflow file documents the one-time publisher registration). The token-driven flow is the fallback / bootstrap and the only place a long-lived PyPI credential appears:
+OIDC Trusted Publishing is the default (no stored secret; the `pypi` environment on `binlecode/actop` is tag-only, and the workflow file documents the one-time publisher registration). It is **dispatched manually against the tag**, never fired by the tag push. The token-driven flow is the fallback / bootstrap and the only place a long-lived PyPI credential appears:
 
 ```bash
 python -m build                       # produces dist/*.tar.gz and dist/*.whl
@@ -229,7 +234,7 @@ a mount point, not a fake); faking the data or the logic under test is not.
 
 ## Commit & Pull Request Guidelines
 - **Branch from `main`; PR strictly into `main`.** Every branch forks from `main` and targets `main`. **Never fork a feature branch off another feature branch** (no stacked PRs): if you need work that is still on an unmerged branch, wait for it to merge and re-branch from `main`. This holds especially for CI/CD and release changes — they land via a single PR into `main`, never a chained branch.
-- **Bump the version in every PR.** Each PR updates `pyproject.toml` version + moves `CHANGELOG.md` `[Unreleased]` into a new dated section, in the same PR: **patch bump by default, minor only for a milestone PR** (major reserved for breaking API/CLI changes). Tagging (`scripts/tag_release.sh`) is a separate step after merge and is what publishes — see the Release Process section above.
+- **Bump the version in every PR.** Each PR updates `pyproject.toml` version + moves `CHANGELOG.md` `[Unreleased]` into a new dated section, in the same PR: **patch bump by default, minor only for a milestone PR** (major reserved for breaking API/CLI changes). Tagging (`scripts/tag_release.sh`) is a separate step after merge; publishing to PyPI is a third, manual step after that — see the Release Process section above.
 - Use concise, imperative commit subjects (as seen in history), e.g. `Add support for M1 Ultra` or `actop/utils.py: add bandwidth of M2`.
 - Keep commits scoped to one logical change.
 - Before every commit and before every push, always run:
diff --git a/cover/actop-demo.gif b/cover/actop-demo.gif
index 9c5bbfa..9cd629c 100644
Binary files a/cover/actop-demo.gif and b/cover/actop-demo.gif differ
diff --git a/cover/index.html b/cover/index.html
index c756f62..bf29571 100644
--- a/cover/index.html
+++ b/cover/index.html
@@ -197,7 +197,7 @@ <h3><span class="ic">▮</span>Memory bandwidth</h3>
       </div>
       <div class="feature">
         <h3><span class="ic">▮</span>Net / disk I/O</h3>
-        <p>Live network ↓/↑ and disk read/write throughput in bytes/s, sourced from native getifaddrs and IOKit counters — no sudo, no subprocesses.</p>
+        <p>Network and disk throughput as mirrored in-over-out charts sharing a zero axis, sourced from native getifaddrs and IOKit counters — no sudo, no subprocesses.</p>
       </div>
       <div class="feature">
         <h3><span class="ic">▮</span>Thermals & throttling</h3>
diff --git a/images/actop-demo.gif b/images/actop-demo.gif
index 9c5bbfa..9cd629c 100644
Binary files a/images/actop-demo.gif and b/images/actop-demo.gif differ
diff --git a/pyproject.toml b/pyproject.toml
index aaeccc6..db8af14 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "actop"
-version = "1.9.1"
+version = "1.9.2"
 description = "Apple Silicon (M1–M4) power, GPU, ANE & memory-bandwidth monitor — sudoless TUI + Python API for profiling local LLM / MLX / CoreML inference"
 readme = "README.md"
 license = { text = "MIT" }
diff --git a/uv.lock b/uv.lock
index e4d1173..3e3934a 100644
--- a/uv.lock
+++ b/uv.lock
@@ -15,7 +15,7 @@ resolution-markers = [
 
 [[package]]
 name = "actop"
-version = "1.9.1"
+version = "1.9.2"
 source = { editable = "." }
 dependencies = [
     { name = "rich" },