Settings / profiles / models alignment (TranscriptX ↔ Transcribe)

TX source pin (read, not imported): glen-w/TranscriptX main at da742f937996d8fd2d4d40ecd9135af5126e4404 (2026-08-13). Transcribe UI stays on port 8510; TX binds 8501.

Theme in this note means Streamlit/CSS chrome, OCR preprocess_profile, and export typography profiles. It does not mean View → Themes (topics) or Overview wordcloud cards except existing ui.overview_cards copy.

What TranscriptX does

Concern	TX behaviour
Settings tabs	Configuration · Analysis · Storage · Watcher · Speakers · Interface · Models · Questions · Corrections (src/transcriptx/web/page_modules/settings.py)
System siblings	Settings, Tools, Profiles, Dashboard Builder, Diagnostics (navigation.py PAGE_SPECS, section settings)
Gating	Settings required_context is "none"; context bar hidden for the settings section
Configuration scopes	Default (read-only) / Project (config.json) / Draft override / Run override. Env TRANSCRIPTX_* wins
Profiles	Separate System page. ProfileManager targets: workflow, topic_modeling, semantic_similarity, acts, tag_extraction, qa_analysis, temporal_dynamics, vectorization, llm_models. Copy-into facade, then env
Models tab	Ollama text tags; live /api/tags; llm_models preset CRUD; “Set as project active”; Run Analysis “Custom (this run)”
Prompts	No hub. Templates in Python modules. Settings → Questions is a llm_custom_qa question library
Interface	Action menus + info tooltips + icon/text appearance; recovery preserves file; Replace writes .bak.
Theme	No .streamlit/config.toml; CSS in web/shell.py (#1f77b4, dark-theme nav). No Settings theme picker
Reset	Unsaved Reset (reload scope baseline) / Revert to defaults. test_config_reset.py is process-global test isolation
Language	Subject is transcript; “Project” is a config layer

What Transcribe does (aligned)

Concern	Transcribe behaviour
Settings tabs	Configuration · Analysis · Detection · Tags · Prompts · Interface · Models · Profiles · Export (settings_interface.py SETTINGS_TABS)
System	Settings · Diagnostics only. Profiles stays a tab, not a System page
Gating	Settings required_context is "none" (stay-don’t-bounce). Apply-OCR gated on a selected notebook; Re-apply declutter has its own notebook picker
Configuration	Workspace-only guided sections (folders, import, archive, Overview cards). No Default/Project/Run radio
Profiles	Activation-pointer + four targets (workflow, ocr, llm, export). Overlay at resolve; detach-on-edit; builtins virtual
Precedence	defaults → workspace → active profile overlay → env allowlist → project OCR allowlist. Project OCR wins env
Models tab	Workspace Ollama URL, text-model hint, token budgets, preprocess_profile seed, Apply-OCR. Live discovery stays on Transcribe / Analyse launchers
Prompts	Prompt Hub for ocr / cleanup / detection / custom; override version bump; dry-run. Analyse module prompts stay module-local
Export tab	Read-only effective export.* + active profile. Live editors on Workflow → Export. Typography: readable / compact / large_print
Theme	.streamlit/config.toml is port 8510 only (no [theme]). CSS in ui/shell.py already TX-shaped. No Settings theme picker
Reset	Archive settings.reset.{stamp}.json then factory defaults. Never silent overwrite (settings_corrupt, settings_schema_unsupported)
Language	notebook, British Analyse. GUI must not say project

Intentional divergences (keep)

1. No speaker/session-scoped settings — TX Speakers tab / speaker_profiles/. Transcribe has notebooks.

2. No TX run-override stack — no Draft/Run run_config_override.json. Jobs freeze EffectiveConfig; live OCR authority is project.json.

3. No copy-into-workspace profiles — overlay + detach-on-edit (contracted).

4. No TX module profile targets (topic_modeling, acts, …) — analysis depth is UI presets.

5. No Dashboard Builder / YAML layout clone — Overview cards checklist (ui.overview_cards) already shipped.

6. No TX Storage/Watcher/Questions/Corrections tabs — paths are env/Docker; Prompt Hub ≠ question library.

7. Profiles stay a Settings tab — System IA is Settings · Diagnostics.

8. Prompt Hub is Transcribe-native — TX has none. Do not CDN-host prompts. Do not move Analyse inline prompts into the Hub here.

9. Launcher vs Settings model split — live Ollama discovery stays on Transcribe/Analyse (model_info.py).

10. Visual theme control — ignore TX (already copied in shell.py). No light/dark picker.

11. Reset is archive-then-defaults — not TX unsaved editor Reset/Revert.

12. Env vs project OCR — TX env wins; Transcribe project OCR wins so a workspace/env URL cannot silently override an open notebook.

13. Port 8510 / no TX process coupling.

14. CDN-hosted prompt/theme assets — keep offline. TX’s optional Ollama-library scrape is not copied.

Do not

• Import TranscriptX at runtime or share a settings schema (docs/INTEGRATION_SEAM.md untouched).

• Reopen Analyse/View IA, Library, Open→Reading, Overview card ids, or frozen interface-menus action IDs.

• Bounce Settings → Home when no notebook is selected.

• Silently rewrite corrupt settings.

• Change ANALYSIS_CONFIG_VERSION / PRESET_POLICY_VERSION without a contract bump plan.