docs: 0.5.0 truth pass, TUI scrollbar clipping fix, and media hygiene

[DLANSAMA]

Draft until the plate tui GIF is recorded and wired into the README — everything else is done.

1. Docs truth pass

Not just stale numbers. Seven claims were actively false:

• CONTRIBUTING.md listed four "main gaps to A+/1.0" — coverage ~82 vs 92, domain→cli extraction, single-sourced TLS pin, remaining schemas. All four had closed (B.4, B.5, generated schemas) and coverage is 89.
• docs/quality-roadmap.md said schemas were "not complete for every command", contradicting docs/api.md and tests/contracts/test_schema_validation.py, which derive the list from build_parser() and fail if any command lacks one. Its own scoreboard row already said the opposite.
• The same file framed the TUI as branch work that "has not reached CI yet". It shipped in 0.5.0.
• docs/test-backlog.md gave the mutation floor as 40%; mutation.yml and run_mutation_baseline.sh both say 48. The same row asked for a mutmut re-run that had already happened and disproved its premise.
• docs/manual.md's Linux detection table listed only the legacy Flatpak app id and no Flatpak profile paths — a partial regression of a defect a previous pass fixed in the prose only, 50 lines above.
• docs/plans/interactive-mode-plan.md still read "Do not start Phase 1 until the 0.4.0 tag exists". plate go shipped in 0.4.0.
• README.md told every reader to pip install 'platecli[tui]' — including the pipx and uv-tool users the install section had just created, for whom that installs Textual into the wrong environment and leaves plate tui failing. docs/troubleshooting.md already had it right.

Plus stale figures re-measured and, importantly, re-sourced: numbers now cite the run or machine they came from, because local Linux and CI Linux legitimately differ and "fixing" one to match the other is how they went stale before.

2. TUI: the scrollbar was clipping which material was detected

On a terminal short enough that the prepare form overflows, #prepare-inputs grows a vertical scrollbar — and Textual scrollbars take real cells rather than overlaying. The widest radio label is the AMS-detected material, PLA — easy, rigid, most models (detected in AMS) at 49 cells, +4 for the toggle and padding = 53, against a column offering 54. One cell of slack, which the scrollbar ate: the label rendered as (detected in AMS with the paren shaved off.

So the screen silently truncated the one thing that tells you which filament it found. Column widened to 62, which leaves 54 even while scrolling.

Found by recording the TUI and looking at the frames. Every existing assertion passed — the label string was intact; only its rendering was cut. The new test asserts the RadioSet has room for its widest button while the scrollbar is up, and fails when the column goes back to 60.

3. Media hygiene

docs/job-hero.mp4 (2.3 MB) is the raw full-length capture; the trimmed cut job-hero-post.mp4 (427 KB) is already committed and is what the launch material uses. The raw file was untracked and unignored — a trap the post-audit gameplan called out by name — and duly got swept into #102, a test-decomposition PR, by a git add. Untracked from HEAD and ignored going forward.

It stays in history, which is fine: I extracted frames and checked them — no IP, serial, or home path.

Gates

ruff / ruff format / mypy (88 files) / bandit (Medium 0, High 0 by severity)   clean
pytest -m "not live"    1420 passed, 1 deselected — coverage 89.09% (floor 83)
syntax / cli_help / ci_workflow / python_compat smokes                          ok
check_layers            Layer boundaries OK (1 allowlisted debt edge)
gen_schemas --check     docs/schemas up to date (26 schemas)
test_docs_consistency   6 passed
uv build + package_contents_smoke                                               ok

sdist doc set verified by extraction: docs/releasing.md correctly absent, and both edited AGENTS.md tables render as contiguous blocks rather than being split by a paragraph — a failure mode this repo has actually shipped.

Known, not fixed here

.pre-commit-config.yaml pins ruff-pre-commit v0.15.20 while ci.yml pins ruff@0.16.0, even though the file's own header calls itself "a local mirror of the lint job… to match CI exactly". Bumping a pinned formatter can reformat code, so that wants its own PR rather than riding along in a docs pass.