docs: retire stale plan claims and fix mutation-baseline contradictions

[DLANSAMA]

Closes out the last two buckets of the doc audit (the number sync landed in #120, user-facing accuracy in #121/#122). No files deleted, no code changed — see the note at the bottom.

What I re-verified, and what changed

I re-checked the 2026-07-31 residual table in docs/plans/post-audit-gameplan.md against current main instead of trusting the audit snapshot. The table was wrong in both directions.

Closed, but still listed as open:

ID	Evidence
S5	utils.py:63-67 is now a 4-line shim delegating to jsonio.redact_url_credentials. Both spellings return https://192.168.1.5/a.stl for the audit's own bob:secret123@… repro — there is no weaker second implementation
A4	mqtt_port is wired end to end: context.py:89,137 → printer.py:339 → mqtt_tls.py:143 → mqtt_tls.py:162 client.connect(resolved_ip, _mqtt_port(printer), …)
T1	pyproject.toml:44 pins textual>=8.0,<9.0 — the <2.0 cap is gone (#115)
A3, Q1, S1, B1, R1, R3	Closed by #119, the camera opt-in work, and the TUI merge

Recorded as closed, but is not:

• Q4 — test: stop stubbing paho in sys.modules #116's commit message says "paho-mqtt is a hard runtime dependency. Tests import the real package." It removed all 81 sys.modules.setdefault("paho…") calls (0 remain), but tests/bambu_test_base.py:26-29 still unconditionally assigns MagicMock() to sys.modules["paho"|"paho.mqtt"|"paho.mqtt.client"] at import time and never restores them. Real paho is installed and importable. That's both an untested-dependency gap and the order-dependence hazard AGENTS.md forbids.

Re-confirmed open, with current evidence — P1, Q5, Q6, A1, A2, S3, R2, T2/WS4, W3.4. Notably _numeric_values("400abc") → [] while _numeric_values("400") → [400.0] (slicer/options.py:162-175), so a typo'd temperature override skips the bound check entirely. A1's target also moved to bambu_cli.cliparse.

Number staleness

Both gameplan headers led with pre-#119/#120 figures presented as current state. They now lead with measured values (1499 non-live, 90.99% Linux / 90.68% Windows over 8368 statements, floor 86, Python 3.10/3.12/3.14) and keep the old ones explicitly marked historical — including the a-plus W1 sizing, which was computed from the 89.4% baseline and is now noted as ~125 remaining hits rather than ~280.

mutation-baseline.md

Three internal contradictions:

1. The per-module table claimed it "reconciles to the 50.7% total, so they are measured rather than estimated." It doesn't. The seven rows are the entire only_mutate scope (pyproject.toml:159-167) yet sum to 2010 mutants (990 survived + 1020 killed) against a headline 2091 (1027 + 1061 + 3 timeout) — 81 unattributed. There is no eighth module to absorb them.
2. "Residual survivors here should be cosmetic" for slicer/output.py — the same file already disproves this 40 lines up at 21.8%, 269 survivors, the largest single pocket.
3. "re-run mutmut on slicer/output.py and update the per-module row" — already done 2026-08-04.

On the two plan docs grok said to delete

I kept interactive-mode-plan.md and tui-plan.md tracked. Both already carried "Implemented — kept as the design record, no longer a to-do list" banners, so they weren't masquerading as live plans, and two live pointers still reference them: session.py:8 cites the interactive plan as that module's design record, and quality-roadmap.md:641 points at the tui-plan cut note "if it is revisited." The one real problem — tui-plan.md kept imperative "all implementation work goes on feat/tui" instructions after its own banner — is folded into the historical framing.

Gates

• pytest -W error::ResourceWarning -m "not live" → 1499 passed, 1 deselected, 13 subtests passed, 8368 statements, 91.0%
• test_docs_consistency.py + test_docs_links.py → 25 passed
• Markdown only (4 files); no .py, workflow, or packaging changes, so ruff/mypy/bandit are untouched by this diff.