docs(readme): refresh bundled LLM list to Qwen3.5/3.6

README LLM-formatting section now states four Qwen tiers (Qwen3.5 2B /
4B / 9B + Qwen3.6 27B) and the magnotia-llm crate row reflects the
four-tier registry. The whisper-ecosystem context doc gets the same
refresh and cites unsloth as the GGUF source.

Older roadmap and Phase-0 audit docs left untouched — they are dated
historical artefacts and rewriting them would muddy the audit trail.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

README.md

@@ -41,7 +41,7 @@ These are enforced in the codebase (where practical) and in the docs under [`doc
 
 ### LLM formatting (local only)
 - Local LLM runtime via [llama-cpp-2](https://github.com/utilityai/llama-cpp-rs) 0.1.144 with Vulkan.
-- Three Qwen3 tiers (1.7B, 4B-Instruct-2507, 14B) auto-selected by hardware probe.
+- Four Qwen tiers (Qwen3.5 2B / 4B / 9B + Qwen3.6 27B) auto-selected by hardware probe.
 - GBNF grammar-constrained output for task extraction (always-parseable JSON).
 - System prompt hardened against voice-delivered prompt injection.
 
@@ -167,7 +167,7 @@ magnotia/
 | **`magnotia-core`** | Shared types (`Segment`, `Transcript`, `Megabytes`, `ModelId`), constants, the `Engine` / `SpeedTier` / `AccuracyTier` enums, hardware probe (`sysinfo`-based), model registry (Whisper + Parakeet entries), hardware-aware recommendation scoring, `process_watch` for meeting detection. |
 | **`magnotia-audio`** | `cpal`-based microphone capture with device hotplug + error forwarding, VAD, `rubato` streaming resampler to 16 kHz mono, `symphonia` file decoding, `hound` WAV I/O. |
 | **`magnotia-transcription`** | `whisper-rs` backend (`WhisperRsBackend`) that owns a `WhisperContext` and supports `set_initial_prompt`. `LocalEngine` wraps both Whisper and Parakeet (via `transcribe-rs` ONNX) behind a common `Transcriber` trait. Streaming primitives (`VadChunker`, `LocalAgreement`, buffer trim) live in the `streaming/` module. Model manager handles downloads, paths, and disk checks. |
-| **`magnotia-llm`** | `llama-cpp-2` engine with Qwen3 model manager. Three high-level surfaces: `cleanup_text` (formatting), `decompose_task` (3–7 micro-steps, GBNF-constrained JSON array), `extract_tasks` (optional-array, GBNF-constrained). Resumable HTTP downloads with SHA-256 verify. |
+| **`magnotia-llm`** | `llama-cpp-2` engine with a four-tier Qwen3.5 / Qwen3.6 model manager. Three high-level surfaces: `cleanup_text` (formatting), `decompose_task` (3–7 micro-steps, GBNF-constrained JSON array), `extract_tasks` (optional-array, GBNF-constrained). Resumable HTTP downloads with SHA-256 verify. |
 | **`magnotia-ai-formatting`** | Post-processing pipeline: filler removal, British English conversion, anti-hallucination filter, smart paragraph breaks on long pauses, optional LLM cleanup. Also hosts the `llm_client::CLEANUP_PROMPT` constant (prompt-injection-hardened). |
 | **`magnotia-storage`** | SQLite via `sqlx` 0.8. Migrations, CRUD for transcripts / tasks / subtasks / profiles / profile terms / settings / error log, FTS5 search, file-storage paths. |
 | **`magnotia-hotkey`** | Linux `evdev` hotkey listener with device hotplug. Parses Tauri-style hotkey strings (`Ctrl+Shift+R`), emits Pressed / Released events. Works natively on Wayland (no X11 dependency). Checks `/dev/input/event*` access on startup; surfaces a clear "add yourself to the `input` group" error when missing. |

docs/whisper-ecosystem/magnotia-context.md

@@ -78,9 +78,9 @@ Magnotia uses `whisper-rs` 0.16 with the `vulkan` feature ([crates/transcription
 
 The `brief.md` references faster-whisper as a pattern source — treat it as one. Do not adopt it as a dependency.
 
-### LLM runtime: `llama-cpp-2` with Qwen3 tiers
+### LLM runtime: `llama-cpp-2` with Qwen3.5 / Qwen3.6 tiers
 
-`crates/llm/` owns the LLM runtime. Three tiers (Qwen3 1.7B, Qwen3 4B-Instruct-2507, Qwen3 14B) selected via hardware probe. Resumable HTTP downloads with SHA-256 verification. Do not add alternative runtimes (candle, mistral.rs, etc.) without strong evidence — the current stack is working.
+`crates/llm/` owns the LLM runtime. Four tiers (Qwen3.5 2B, Qwen3.5 4B, Qwen3.5 9B, Qwen3.6 27B) selected via hardware probe. All Q4_K_M GGUFs from unsloth's HF org. Resumable HTTP downloads with SHA-256 verification. Do not add alternative runtimes (candle, mistral.rs, etc.) without strong evidence — the current stack is working.
 
 ### GPU compatibility: Vulkan everywhere