chore(quality): complexity refactors, stricter gates and fixes from the quality pass

.github/workflows/theme-validation.yml

@@ -34,13 +34,13 @@ jobs:
         run: mypy src/
 
       - name: Shell check
-        run: shellcheck --shell=bash --severity=warning scripts/*.sh
+        run: git ls-files 'scripts/*.sh' 'lib/*.sh' 'DreamcoderShell/*.sh' 'DreamcoderShell/**/*.sh' | xargs shellcheck --shell=bash --severity=warning
 
       - name: Run tests
         run: python -m pytest tests/ -v --tb=short
 
       - name: Coverage
-        run: python -m pytest tests/ --cov=dreamcoder_theme --cov-fail-under=40
+        run: python -m pytest tests/ --cov=dreamcoder_theme --cov-fail-under=80
 
       - name: Run theme health check
         env:

.pre-commit-config.yaml

@@ -28,7 +28,7 @@ repos:
     hooks:
       - id: shellcheck
         args: ["--shell=bash", "--severity=warning"]
-        files: ^(scripts|lib)/
+        files: ^(scripts|lib|DreamcoderShell)/.*\.sh$
 
   - repo: local
     hooks:

DreamcoderShell/.config/shell/aliases/dreamcoder-modern.sh

@@ -64,7 +64,10 @@ extract() {
     echo "Usage: extract <archive> [output_dir]"
     return 1
   fi
-  local file="$1" dir="${2:-${file%.*}}"
+  # Two statements: in bash a single `local` expands every value before any is
+  # assigned, so dir would not see file (ShellCheck SC2318).
+  local file="$1"
+  local dir="${2:-${file%.*}}"
   case "$file" in
   *.tar.gz | *.tgz) tar -xzf "$file" -C "$(dirname "$file")" 2>/dev/null || tar -xzf "$file" ;;
   *.tar.bz2 | *.tbz2) tar -xjf "$file" ;;

docs/adr/0003-python-quality-strategy.md

@@ -33,7 +33,7 @@ We adopt a three-layer quality strategy for all Python code:
 ### Layer 3 — Coverage (pytest-cov)
 
 - **Metric**: Branch coverage via `pytest --cov=dreamcoder_theme`
-- **Threshold**: `fail_under = 40` (baseline; expected to increase over time)
+- **Threshold**: `fail_under = 80` (ratcheted from the original 40 baseline on 2026-09-29, when measured coverage was 83%; raise it again only when coverage rises — never lower it)
 - **Reporting**: `--cov-report=term-missing` shows uncovered lines
 
 All three layers run in CI on every push and pull request.
@@ -52,15 +52,15 @@ Positive:
 
 Negative:
 - Mypy strict mode requires explicit annotations throughout the codebase
-- Coverage fail-under of 40 is low — new code should aim for 80%+
+- The 80% floor sits just under the measured level (83%), so a large untested addition fails CI; raising it is a deliberate follow-up
 - Initial migration required fixing pre-existing issues
 
 ## Compliance
 
 - `ruff check src/ tests/` must pass
 - `ruff format --check src/ tests/` must pass
 - `mypy src/` must pass (strict mode)
-- `pytest --cov=dreamcoder_theme --cov-fail-under=40` must pass
+- `pytest --cov=dreamcoder_theme --cov-fail-under=80` must pass
 
 ## Alternatives Considered
 

docs/configuration/ml4w.md

@@ -89,8 +89,9 @@ The ML4W 2.16 delta is ported as follows:
 | Rewritten AZERTY detection (`fr`, `be`) | Ported; binds the AZERTY keysyms only on AZERTY layouts, since the profile owns the digit workspace binds |
 
 `tests/ml4w/keybindings_variant.bats` fails on any new collision between the
-variant and the profile. The `SUPER + SHIFT + arrows` overlap (variant resize,
-profile move) predates 2.16 and is listed there as a known exception.
+variant and the profile; there are no exceptions. The profile owns
+`SUPER + SHIFT + arrows` (move window), so the variant's keyboard resize sits on
+`SUPER + CTRL + arrows` instead of ML4W's upstream `SUPER + SHIFT + arrows`.
 
 ## hyprctl dispatch is broken on Hyprland 0.55+ — native dispatchers used
 

ml4w_assets/hypr/conf/keybindings/dreamcoder.lua

@@ -17,17 +17,17 @@ local mainMod = "SUPER" -- Sets "Windows" key as main modifier
 -- Only the emoji picker stays here.
 hl.bind(mainMod .. " + CTRL + E", hl.dsp.exec_cmd("~/.config/ml4w/settings/emojipicker.sh"), { description = "Open the emoji picker" })
 
--- Windows
 -- Profile owns: Q (kill), F (fullscreen), T (float), H/L/K/J + arrows (focus),
--- SHIFT+H/L/K/J + arrows (move window), Y (toggle split).
+-- SHIFT+H/L/K/J + arrows (move window), Y (toggle split). Keyboard resize sits on
+-- CTRL + arrows here (upstream ML4W uses SHIFT + arrows, which the profile owns).
 hl.bind(mainMod .. " + SHIFT + Q", hl.dsp.exec_cmd("hyprctl activewindow | grep pid | tr -d 'pid:' | xargs kill"), { description = "Quit active window and all open instances" })
 hl.bind(mainMod .. " + M", hl.dsp.window.fullscreen({ mode = "maximized", action = "toggle" }), { description = "Toggle Maximize Window" })
 hl.bind(mainMod .. " + SHIFT + T", hl.dsp.exec_cmd("~/.config/ml4w/scripts/ml4w-toggle-allfloat"), { description = "Toggle floating for all windows of workspace" })
 hl.bind(mainMod .. " + ALT + T", function() hl.dispatch(hl.dsp.window.float({ action = "toggle" })); hl.dispatch(hl.dsp.window.pin()) end, { description = "Toggle floating + pinned" })
-hl.bind(mainMod .. " + SHIFT + right", hl.dsp.window.resize({ x = 100, y = 0, relative = true }), { repeating = true }, { description = "Increase window width with keyboard" })
-hl.bind(mainMod .. " + SHIFT + left", hl.dsp.window.resize({ x = -100, y = 0, relative = true }), { repeating = true }, { description = "Reduce window width with keyboard" })
-hl.bind(mainMod .. " + SHIFT + down", hl.dsp.window.resize({ x = 0, y = 100, relative = true }), { repeating = true }, { description = "Increase window height with keyboard" })
-hl.bind(mainMod .. " + SHIFT + up", hl.dsp.window.resize({ x = 0, y = -100, relative = true }), { repeating = true }, { description = "Reduce window height with keyboard" })
+hl.bind(mainMod .. " + CTRL + right", hl.dsp.window.resize({ x = 100, y = 0, relative = true }), { repeating = true }, { description = "Increase window width with keyboard" })
+hl.bind(mainMod .. " + CTRL + left", hl.dsp.window.resize({ x = -100, y = 0, relative = true }), { repeating = true }, { description = "Reduce window width with keyboard" })
+hl.bind(mainMod .. " + CTRL + down", hl.dsp.window.resize({ x = 0, y = 100, relative = true }), { repeating = true }, { description = "Increase window height with keyboard" })
+hl.bind(mainMod .. " + CTRL + up", hl.dsp.window.resize({ x = 0, y = -100, relative = true }), { repeating = true }, { description = "Reduce window height with keyboard" })
 hl.bind(mainMod .. " + G", hl.dsp.group.toggle(), { description = "Toggle window group" })
 -- hl.bind(mainMod .. " + SHIFT + G", hl.dsp.group.active("f"), { description = "Switch to next group window" })
 hl.bind(mainMod .. " + ALT + left", hl.dsp.window.swap({ direction = "l" }), { description = "Swap tiled window left" })

odd/tasks/engineering-quality-pass.md

@@ -0,0 +1,93 @@
+# Engineering quality pass
+
+## Objective
+
+Raise maintainability of the theme engine and the repo gates using measured
+evidence, without changing behaviour: reduce the worst cyclomatic-complexity
+hotspots, ratchet the coverage gate to the real level, and document publication.
+
+## Problem (measured 2026-09-29)
+
+- `mypy --strict`: clean (56 files). Coverage: 83% (gate is 40%). ruff: clean.
+- radon cc worst offenders: `palette.validate_palette` E(34),
+  `design_system.evaluate_contract` D(26), `design_system._parse_renderer_output`
+  D(21), then C(13–19) in doctor, targets, audit, herdr renderer, repair, backups.
+- Largest modules: `sync.py` 1061 lines, `cli_handlers.py` 620, `herdr_switch.py` 552.
+- Coverage gate at 40% would let a 40-point regression through.
+
+## Why
+
+Central gates (contrast validation, design-system contract) with cyclomatic
+complexity above 20 are hard to review and easy to break; a coverage floor far below
+reality is not a gate.
+
+## Scope
+
+- In: behaviour-preserving extraction refactors of the three D/E functions (small
+  pure helpers, no new abstractions beyond need), characterization tests written
+  first where existing coverage of a branch is missing, coverage ratchet in
+  `pyproject.toml` and CI, PR slicing plan for the stacked branches.
+- Out: splitting `sync.py`/`cli_handlers.py` wholesale (separate design decision,
+  in-flight `hexagonal-architecture-v2` change exists); pushing or opening PRs
+  (user decision); changing any generated artifact byte.
+
+## Constraints
+
+- Public function signatures and outputs unchanged; every Light/Dark artifact
+  byte-identical (`./scripts/dreamcoder sync` repo-only must report 0 changes).
+- ruff, ruff format, `mypy --strict`, shellcheck stay clean.
+- Never stage the unrelated timer-rewritten / user working-tree files.
+- TDD: off (no project configuration). Runners: `python -m pytest tests/`,
+  `bats tests/shell/ tests/ml4w/`. Complexity check: `uvx radon cc src -s -n C`.
+- Delivery: work-unit commits on `refactor/remove-night-profile`.
+
+## Tasks
+
+- [x] Q0 — Resolve the SUPER+SHIFT+arrows overlap (variant resize vs profile move):
+  resize moved to SUPER+CTRL+arrows, collision test has no exceptions. Route: inline.
+- [x] Q1 — Refactor `validate_palette` (E→ at most B) preserving every error message and order. Route: delegated writer.
+  `validate_palette` E(34) → B(6); helpers all ≤ B(8). Commits: 4c4214b (refactor),
+  c4db159 (fail-fast/skip tests); characterization tests landed in 684026f (see note).
+- [x] Q2 — Refactor `evaluate_contract` and `_parse_renderer_output` (D→ at most B). Route: delegated writer.
+  `evaluate_contract` D(26) → A(1), `_parse_renderer_output` D(21) → A(2) via a per-target
+  parser table; helpers all ≤ B(7). Commit: d503d2e.
+- [x] Q3 — Ratchet coverage `fail_under` (pyproject + CI + ADR-0003) from 40 to 80 (measured 83–84%). Commit: 089f513. Route: inline.
+- [x] Q4 — PR slicing plan (nothing pushed; user decides): keep commit order, cut at
+  commit boundaries into chained PRs: (1) ML4W 2.16 compat f7c2b9b..339ab7e,
+  (2) stow layer + Herdr 0.9.1 3398856..a7c925a, (3) theme-mode correctness + listener
+  hook 702751f..8127674, (4) shell safety c9cf723..754d352, (5) Night removal
+  319ffdd..f7597b1, (6) quality pass 089f513 onwards. PRs 1, 2, 3 and 5 exceed the
+  ~400-line heuristic (tests, fixtures and generated deletions dominate); 684026f mixes
+  characterization tests into the extract() fix, to be re-cut when slicing. Route: inline.
+- [x] Q5 — Found while verifying Q1/Q2: `dreamcoder sync --help` ran a real sync.
+  `main(argv)` now parses arguments before any side effect (--help exits 0, unknown
+  options exit 2); tests inject argv. Route: inline.
+
+## Acceptance criteria
+
+- radon reports no function above C(15) in the three refactored areas; `validate_palette` at most B.
+- Full pytest + bats green; repo-only sync reports 0 changes; mypy/ruff clean.
+
+## Progress
+
+- Baseline metrics above. Q0 done.
+- Q1/Q2 evidence: characterization tests written first and green on the old code
+  (golden ordered error list over a frozen dark palette; exact findings for
+  MISSING_MODE, ROLE_RESOLUTION, RENDER_FAILURE, SEMANTIC_PROVENANCE_INVALID; parser
+  invalid-JSON/unknown-target). The three functions are now 100% line-covered.
+  Coverage total 83% → 84%. pytest exit 0; bats 147/147 ok; ruff, ruff format, mypy
+  (56 files) clean. Byte identity: sync run in an isolated worktree with a throwaway
+  HOME/XDG_* (`DREAMCODER_THEME_MODE=dark|light`, `DREAMCODER_WRITE_REPO=1`), sha256
+  of all 900 tracked non-src files plus 20 HOME outputs identical HEAD vs refactor in
+  both modes.
+- Note: a concurrent commit (684026f `fix(shell): ...`) swept the staged
+  characterization tests in with it; left as-is (no history rewrite). Later commits
+  use `git commit --only <paths>`.
+- Incident: `./scripts/dreamcoder sync --help` does not parse `--help` and ran a real
+  sync once (active light identity re-applied to live paths; it also regenerated the
+  timer-modified `DreamcoderPi/.../dreamcoder-dark.json` and
+  `DreamcoderWarp/.../Dreamcoder-Dark.yaml` back to their committed bytes).
+
+## Next step
+
+Feature complete; publication is the user's decision.

pyproject.toml

@@ -111,7 +111,7 @@ relative_files = true
 
 [tool.coverage.report]
 show_missing = true
-fail_under = 40
+fail_under = 80
 
 [dependency-groups]
 dev = ["mypy>=2.1.0"]