diff --git a/local/docs/redbear-power-improvement-plan.md b/local/docs/redbear-power-improvement-plan.md
index 6adb91a529..e2348d3540 100644
--- a/local/docs/redbear-power-improvement-plan.md
+++ b/local/docs/redbear-power-improvement-plan.md
@@ -4044,6 +4044,105 @@ background thread.
 
 ---
 
+## 46. v1.22 Sort by IO (2026-06-20)
+
+Per the user's "v1.22 = Sort by IO (Recommended)" directive, v1.22
+adds per-process disk IO totals as a new sortable column in the Process
+tab.
+
+### 46.1 What was implemented
+
+**Two new fields on `ProcessInfo`**:
+- `io_read_kb: u64` — total bytes read by the process over its lifetime
+  (sourced from `/proc/[pid]/io:read_bytes`, normalized to KiB).
+- `io_write_kb: u64` — total bytes written by the process
+  (sourced from `/proc/[pid]/io:write_bytes`, normalized to KiB).
+
+**Two new helpers** in `process.rs`:
+- `read_io_bytes(pid: u32) -> u64` — reads `/proc/[pid]/io` and extracts
+  the `read_bytes:` field. Returns 0 if the file is missing or the
+  field is absent (process may have just exited, or `/proc/[pid]/io`
+  requires `CAP_SYS_PTRACE` for an owned UID).
+- `write_io_bytes(pid: u32) -> u64` — same pattern for `write_bytes:`.
+
+Both helpers are silent on failure — IO is a "best-effort" column, never
+a build or runtime blocker.
+
+**`io_total_kb()` method on `ProcessInfo`** — sums `io_read_kb` and
+`io_write_kb` for sorting and display.
+
+**`SortMode::Io` variant** added to the existing `SortMode` enum:
+```rust
+pub enum SortMode {
+    Rss, Cpu, Io, Pid, Name,
+}
+```
+
+**Cycle updated** — `o` now cycles `Rss → Cpu → Io → Pid → Name → Rss`.
+The pre-existing `sort_cycle` test was updated to assert the new order;
+the pre-existing `sort_by_*` tests continue to pass.
+
+**`SortMode::Io.sort()` implementation** — descending order by
+`io_total_kb()`, with `pid` as the tiebreaker (consistent with the
+other sort modes).
+
+**Render-side column** — the Process panel header now reads:
+```
+PID     STATE  PRIO  NI   THR  CPU%   IO          RSS         COMM
+```
+
+VIRT was replaced by IO — the panel width is constrained, and IO is
+the higher-information column for diagnosing "what is hammering the
+disk" workloads. RSS is preserved as the memory column.
+
+### 46.2 Test coverage
+
+Four new unit tests added in `process.rs` `io_sort_unit_tests`:
+- `io_total_sums_read_write` — basic field summation.
+- `io_total_saturates_on_underflow` — non-negative inputs only.
+- `sort_by_io_descending` — sort puts highest `io_total_kb()` first.
+- `sort_cycle_includes_io` — full cycle is `Rss → Cpu → Io → Pid → Name`.
+
+The pre-existing `sort_cycle` test was updated to reflect the new
+cycle. Total tests now: **80** (up from 76).
+
+### 46.3 Cross-compile + smoke test results
+
+| Target       | Size        | SHA256                                                            |
+|--------------|-------------|-------------------------------------------------------------------|
+| Linux host   | 3.0 MB      | (run from `target/release/redbear-power`)                         |
+| Redox x86_64 | 4,123,496 B | `65336a2f50b5e3a7100486b93ce2ab0443ccc1276d84619e5b6346a3a182adcb` |
+
+Linux host smoke test confirms the IO column renders with realistic
+values:
+```
+│PID     STATE  PRIO  NI   THR  CPU%   IO          RSS         COMM                                                    │
+│3317951 R     20   0    41   0.0    384.5 GiB   4.0 GiB     opencode                                                  │
+│105364  S     20   0    92   0.0    14.2 GiB    2.1 GiB     thunderbird                                               │
+│1857542 S     20   0    8    0.0    0.0 KiB     649.9 MiB   kscreenlocker_g                                           │
+```
+
+`opencode` processes dominate (heavy disk IO), `thunderbird` is moderate
+(mailbox indexing), `kscreenlocker_g` shows 0.0 KiB (no disk access).
+
+### 46.4 Why IO instead of VIRT in the panel
+
+Virtual size (VIRT) is the address-space reservation, not actual
+memory consumption. For diagnosing "this process is using all the
+disk" or "this process is paging", **IO bytes is the actionable signal**.
+Resident set (RSS) remains in the panel because it tracks actual
+physical memory usage, which VIRT does not.
+
+### 46.5 Permission caveat (documented for future maintainers)
+
+`/proc/[pid]/io` requires `CAP_SYS_PTRACE` to read another UID's IO
+counters on Linux. On Redox the proc scheme behavior is different and
+may or may not enforce this. The helpers silently return 0 on read
+failure so the column degrades gracefully — no panic, no missing-data
+marker needed (the 0 itself communicates "IO counter unavailable").
+
+---
+
 ## See Also
 
 - **`local/docs/RATATUI-APP-PATTERNS.md`** §13 — the canonical ratatui 0.30 best-practices update that this plan is derived from. Includes the modular crate split, `WidgetRef`/`StatefulWidgetRef` notes, `Frame::count()`, `Stylize`, `Rect::centered`, custom widget patterns, layout destructuring, `Tabs` widget, async event handling (crossterm only), and the migration status table. Use this as the implementation guide while this doc is the roadmap.