Analysis and Detect¶
Run analysis on transcribed text, read the results under View, and scan for patterns with Detect.
First-time path: user guide. Limits: known limitations.
This notebook¶
UI: Workflow → Analyse
Choose a preset (Quick / Balanced / Thorough / Custom). First-time: keep Balanced.
Optionally enable an Ask-notebook question.
Run analysis. On success, open View → Overview.
Preset |
What you get |
|---|---|
Quick |
Fast lexical / structural modules — no LLM, no heavy modules, no detectors |
Balanced |
Adds similarity clustering and an LLM summary |
Thorough |
All suitable core modules plus every detector |
Custom |
Pick modules and detectors (starts from Balanced) |
Use a text Ollama model for LLM modules and detectors (Settings → Models, or pick one on Analyse / Batch). Deterministic modules work without it. Missing pieces show as plain messages such as “Needs a text model”.
Edit what each preset includes under Settings → Analysis. Mid-run settings changes apply to the next run only.
View¶
Inspect results under View: Overview / Themes / Mood / Summaries / People & Places. Mood includes Moments; Summaries includes Ask. A status strip shows whether results are current.
Overview / Mood: Compare with corpus / period against other notebooks
Mood → Moments and page-series charts: Jump to page → Reading
People & Places: This notebook | All notebooks (opt-in map geocoding)
Technical module details under Advanced (enable in Settings → Configuration → Overview)
Chart notes: visual compare.
Detect¶
UI: View → Detect — review findings (Accept / Reject, jump to source pages). Accept applies remaining page tags. Multi-page findings add per-page Accept / Reject; Accept remaining keeps the others.
Built-ins: poetry, to-do lists, other lists, quotations, beer labels,
first-person I counts, swear-word counts, names / people, plus custom
detectors. Count detectors show a per-page table instead of review cards.
Ad-hoc runs still launch from Detect → Run Detection.
./transcribe.sh cli detect "$TRANSCRIBE_PROJECTS_DIR/my-notebook" --detector poetry
./transcribe.sh cli detect "$TRANSCRIBE_PROJECTS_DIR/my-notebook" --list
Tag matching pages unions tags onto span pages; rejected findings are
skipped. Names / people tags each detected person name rather than a generic
names tag.
Manage prompts under Settings → Prompts; custom detectors under Settings → Detection; labels under Settings → Tags.
Batch Analyse¶
UI: Workflow → Analyse → Batch — pick notebooks, a preset, optional shared text model, Start batch analysis. Empty-text notebooks are skipped.
./transcribe.sh cli bulk-analyse pending|import-run|notebooks --preset balanced
./transcribe.sh cli bulk-analyse status|resume <analysis_batch_id>