diff --git a/docs/python-split-recipes.md b/docs/python-split-recipes.md
index 1ebebbfb..02c20385 100644
--- a/docs/python-split-recipes.md
+++ b/docs/python-split-recipes.md
@@ -167,3 +167,41 @@ Coverage: `meta/pyproject.toml` sets `source = ["python_pkg"]`, so a new
 sibling module is measured automatically. Confirmed: new modules show up in
 the report at 100% rather than being silently skipped. Add no new branches
 (no `if TYPE_CHECKING` guards beyond what moved, no defensive `try/except`).
+
+## The seam-selection rule (settled over 8 packages)
+
+Generalised from `python_pkg`; full detail in `docs/python-split-recipes.md`.
+
+**A block may move if every MOD-relative patch that reaches it comes from one
+test module. If two test modules patch a shared collaborator, keep the
+collaborator and move its callers instead.** The caller resolves the name
+through its *own* module's globals, so importing it into the new module makes
+the patch land there.
+
+Corollaries that each cost a failed run to learn:
+
+- **`test_X.py` needing no edit is the signal you cut in the right place.**
+  It held on four of five source splits in `brother_printer`.
+- **Never blanket-`sed` a MOD.** Decide per test *by which function is under
+  test*. `test_challenge_part3.py` needed 13 of its strings repointed and 4
+  left alone.
+- **A mispointed patch does not always fail — it can HANG or do real work.**
+  `code_tutor` spun for two minutes running a real interactive flow;
+  `cups_queue` would have restarted the live CUPS daemon. Always run under
+  `timeout`, and assert the targets resolve *before* invoking pytest.
+- **When splitting a test file, copy the WHOLE preamble** (docstring, imports,
+  fixture builders) into the new file and let ruff's `F401` pass prune it.
+  Hand-writing the import block failed on every early attempt; the copy-then-
+  prune order removed 82 imports correctly in `wsg_grabber` with none missed.
+  It does **not** cover module-level `def`s and constants sitting *between*
+  sections — `_make_verifier`, `_QUESTIONS`, `_logger` and
+  `_SAME_FIELD_X_TOLERANCE_PX` were each left behind that way. Grep for them.
+- **Budget two cuts.** Nine of the splits landed 1–20 lines over after the
+  first seam. Pure-data blocks (prompt strings, skip tables, SVG fragments)
+  are the cheapest second cut and have no patch targets.
+- **Prefer a `conftest.py` fixture over a duplicated builder** — that is where
+  `_make_verifier` went, and it removed a shadowing copy of `_make_live_mock`.
+
+Two helper scripts are in the scratchpad, not committed (they are throwaway):
+they cut a module at a banner comment or at a named `def`/`class` and copy the
+preamble. Rewriting them takes about five minutes.
diff --git a/docs/shell-split-recipes.md b/docs/shell-split-recipes.md
new file mode 100644
index 00000000..dc7ed237
--- /dev/null
+++ b/docs/shell-split-recipes.md
@@ -0,0 +1,29 @@
+# Shell split recipes for the 250-line cap
+
+Split out of `refactor_claude_todo_resume.md` (which is itself capped).
+Applies to `linux_configuration/` and `phone_focus_mode/`.
+
+`linux_configuration` and `phone_focus_mode` have **no test suite and no
+coverage gate**, so none of the above transfers. Decision 6 governs:
+`bash -n` + `shellcheck` + `systemctl cat` path checks, and **never execute an
+enforcement script that mutates the live system or the phone.**
+
+- After each split, `bash -n` and `shellcheck` the entry script *and* every new
+  lib. Neither catches a lib that is never sourced.
+- So also: for each new `lib/*.sh`, grep the entry script for the `source`/`.`
+  line naming it, and confirm the path resolves from where the script actually
+  runs (`${BASH_SOURCE[0]%/*}` or absolute — never a bare relative `./lib/`).
+- **Two live units point at repo paths** and break if a file moves or is
+  renamed. `systemctl cat` them after touching either:
+  - `/etc/systemd/system/dns-blocklist-refresh.service` →
+    `scripts/single_use/features/setup_dns_blocker.sh refresh`
+  - `/etc/systemd/system/media-organizer.service` →
+    `scripts/single_use/utils/organize_downloads.sh`
+
+  Every other in-repo unit references an installed copy under `/usr/local/`,
+  so those are only at risk if the *installer* is what you split.
+- Prefer files that **nothing** references first — pure lint risk.
+- Watch **jscpd** on the first multi-lib commit: repeated `set -euo pipefail` +
+  source-guard headers across several new libs is exactly what pushes
+  duplication toward the 2% threshold, and it only fires on the real
+  `git commit`.
diff --git a/refactor_claude_todo_resume.md b/refactor_claude_todo_resume.md
index 789ab56e..dce5dbd0 100644
--- a/refactor_claude_todo_resume.md
+++ b/refactor_claude_todo_resume.md
@@ -6,27 +6,45 @@
 
 ## Where things stand
 
-**183 files** over 250 lines (was 198). Everything is committed and pushed to
-`main` at `9e688c5`; the tree is clean. The three unrelated edits an earlier
-handoff warned about (`README.md`, the two `analysis_options.yaml`) are gone.
+**142 files** over 250 lines (was 183). Everything is committed and pushed to
+`main` at `4b2cd0c`; the tree is clean.
 
-Done already (do not redo):
+**`python_pkg/` is DONE — 50 violations → 0.** Do not reopen it. Cleared in
+this order, one package per tranche, pushing after each: `brother_printer`
+(15→0, suite pinned at 402), `code_tutor` (13→0, 216→217), then `android_ui`,
+`app_icons`, `token_audit`, `focus_policy`, `random_jpg` (57/44/81/137/14),
+then `wsg_grabber`'s six test modules (277).
 
-- `~/utils`: `plans/` + `sessions/` exempted from the cap (`specs/` still capped)
-- `17b20e5` new `no-inline-python` pre-commit hook + its one violation
-- `004b4b0` `CLAUDE.md` 259 → 141
-- `72a5e28` + `73b85f5` `poker-stakes/` tracked, then 4 violations → 0
-- `77c6d9f`…`d86965f` **every `wsg_grabber` source file** — `catalog` 251→175,
-  `cli` 282→247, `review` 330→196, `net` 335→109, `store` 382→202,
-  `downloader` 420→238, `player` 373→235, `ui` 398→242
-- `e03fbd1` `wsg_grabber/tests/test_review.py` 297 → 112 + two new modules
-- `2832b78` `brother_printer/display.py` 427 → 342 — **partial, still a
-  violation.** Extracted `_supply.py` + `_severity.py` and deduped the
-  toner/drum bars. The last ~92 lines were blocked on `mock.patch` pinning,
-  which the decision below now unblocks.
+Earlier sessions had already done: the `~/utils` exemptions, the
+`no-inline-python` hook, `CLAUDE.md`, `poker-stakes/`, and every `wsg_grabber`
+**source** file.
 
 `git log --oneline --grep='cap'` is the authoritative list.
 
+## What is left
+
+| Directory | Count |
+|---|---|
+| `linux_configuration` | 84 |
+| `phone_focus_mode` | 13 (do LAST) |
+| `kcd2_dice_solver` | 12 |
+| `focus_owner` | 11 |
+| `billsplit` | 6 |
+| `reverse_survivors` / `meta` / `docs` | 4 each |
+| `.github` / `bucket_catch` | 2 each |
+
+Within `linux_configuration` the bulk is `scripts/single_use/` (39 across
+`features`/`utils`/`fixes`), then `scripts/periodic_background/` (10) and
+`tests/` (4).
+
+## The seam-selection rule
+
+Settled over eight `python_pkg` packages and written up in
+`docs/python-split-recipes.md` under "The seam-selection rule". In one line:
+**a block may move if every patch reaching it comes from one test module; if
+two modules patch a shared collaborator, keep the collaborator and move its
+callers.** Read the doc before splitting anything with tests.
+
 ## The live worklist
 
 Do not work from a list in this file — it goes stale. Run:
@@ -36,18 +54,14 @@ bash ~/utils/scripts/check_file_length.sh --all
 ```
 
 **Every count in this file is a snapshot and will be wrong by the time you read
-it. Run the script.** As of `ac02a23` the shape was `linux_configuration` 84,
-`python_pkg` 41, `phone_focus_mode` 13, `kcd2_dice_solver` 12, `focus_owner`
-11, `billsplit` 6, `reverse_survivors`/`meta`/`docs` 4 each,
-`.github`/`bucket_catch` 2 each — quoted only to tell you where the bulk sits,
-never to work from.
+it. Run the script.** The table above is a snapshot at `4b2cd0c`, quoted only
+to tell you where the bulk sits, never to work from.
 
-Within `python_pkg`: `brother_printer` 15, `code_tutor` 13, `wsg_grabber` 6
-(tests only), then singles.
-
-**Do the source files in a package before its tests.** Proven on `wsg_grabber`:
-the test files then split along seams the source already established, instead
-of you inventing a new boundary for them.
+**Do the source files in a package before its tests.** Proven on `wsg_grabber`
+and again on all seven `python_pkg` packages: the test files then split along
+seams the source already established, instead of you inventing a new boundary
+for them. On `brother_printer` this meant four of the five source splits needed
+**no test edit at all**.
 
 ## Decisions already made (do not re-ask)
 
@@ -68,6 +82,14 @@ of you inventing a new boundary for them.
    stalled on `brother_printer` because of it.
 9. Per-commit gate is `pre-commit run --files <changed>`; the real `git commit`
    is the full gate (it adds jscpd). Push per directory tranche, not per commit.
+10. **When a seam splits code covered by a per-file-ignore, fix the lint —
+    do not copy the ignore onto the new file** (user's call, 2026-08-16).
+    Done for `code_tutor/cli.py`: `S607` fixed with `shutil.which`, `PERF203`
+    by extracting the probe so the poll loop carries no `try`/`except`. Both
+    entries deleted from `meta/pyproject.toml`; the repo now has two fewer
+    suppressions. The `which()` fix added a branch, so a covering test was
+    written (216 → 217) rather than excluding it. Expect the same question on
+    any remaining per-file-ignore you split through — ask, don't assume.
 
 ## The constraint that will bite you
 
@@ -93,13 +115,19 @@ Plus `.github/workflows/file-length.yml` modelled on
 `.utils`), scoped to changed files. Then prove it fails: stage a deliberately
 251-line file, confirm `git commit` aborts, delete it.
 
-## Python split recipes
+## Language recipes
 
-`docs/python-split-recipes.md` holds the settled Python rules: the `__all__`
-re-export form, the `TC001` trap that lints clean and breaks at runtime, the
-mixin shape for splitting a class, what the tests pin in place, and the
-coverage behaviour of new sibling modules. **Read it before the first Python
-split** — each rule there cost a failed gate run or a revert to establish.
+- `docs/python-split-recipes.md` — the `__all__` re-export form, the `TC001`
+  trap that lints clean and breaks at runtime, the mixin shape for splitting a
+  class, what the tests pin in place, and the coverage behaviour of new sibling
+  modules. `python_pkg` is done, so this now matters only if a stray `.py`
+  turns up outside it.
+- `docs/shell-split-recipes.md` — **read this before the first
+  `linux_configuration` split.** What replaces the pass count when there is no
+  test suite, the two live systemd units that point at repo paths, and the
+  jscpd trap on the first multi-lib commit.
+
+Each rule in both cost a failed gate run or a revert to establish.
 
 ## Split recipes