Workflow walkthrough media capture¶
Maintainer notes for regenerating screenshots and GIFs under docs/_static/workflows/.
Sample corpus¶
Item |
Value |
|---|---|
Fixture |
|
Provenance |
Synthetic authored text for docs; not a real recording |
Speakers |
|
Suggested names |
Maya (facilitator), Jordan (engineering), Sam (support) |
Import via Import Transcript. Do not commit managed library copies under data/.
Application state¶
Capture against a disposable data root when practical.
Walkthrough media |
Expected state |
|---|---|
Import / Run Analysis |
Fresh import; no requirement for prior runs |
Overview / Insights / Export / Charts |
Completed Balanced run on the fixture (re-run after naming if speaker cards should show names) |
Speaker Identification |
Fixture selected; speakers still diarized or mid-naming |
Local AI |
|
Groups |
At least one managed transcript; create via Groups expander |
Corrections |
Transcript VIEW with Correct mode; unique find text in a segment |
Rename Transcript |
Managed transcript selected on Rename page |
Speakers |
At least one longitudinal profile (link from Speaker ID) |
Export HTML preview |
Unpacked ZIP |
Viewport and chrome¶
Browser window 1440×900 (or device pixel ratio 1 with that CSS viewport).
Capture the Streamlit app pane only — no OS desktop, no browser URL bar if avoidable.
Prefer tight crops around the meaningful region (sidebar + main panel, or main panel alone).
Scrub usernames, absolute host paths, tokens, and personal transcript titles before committing.
Capture method¶
Product policy: do not add Playwright-driven Streamlit acceptance automation for docs media.
Practical approach used for the initial set:
Launch the web UI with a disposable data root (
TRANSCRIPTX_*_DIR) and a writtenschema_epoch.jsonmarker.Import
docs/workflows/fixtures/planning_review.json, name speakers, run Balanced analysis.Select the transcript and run in the sidebar VIEW pickers before opening Overview / Insights / Artifacts.
Capture stills with Chromium (Playwright one-off helper
scripts/docs_capture_workflow_media.pyor OS screenshots).Build short GIFs from a few stills (
convert+gifsicle) when a live interaction capture is awkward.Keep raw files under
.local/workflow_media/; promote curated assets todocs/_static/workflows/.
Do not couple production code to documentation capture.
Example compress commands¶
# PNG (lossy palette when text remains readable)
pngquant --quality=65-85 --ext .png --force docs/_static/workflows/*.png
# or lossless shrink
optipng -o2 docs/_static/workflows/*.png
# GIF
gifsicle -O3 --colors 128 -o out.gif in.gif
If those tools are unavailable, use an equivalent compressor that keeps UI text legible.
Asset checklist¶
Committed names expected by the walkthrough pages:
first-analysis-import.pngfirst-analysis-run-analysis.pngfirst-analysis-run-complete.giffirst-analysis-overview.pngspeaker-identification-page.png(legacyspeaker-trust-page.png)speaker-identification-naming.gif(legacyspeaker-trust-naming.gif)speaker-identification-transcript.png(legacyspeaker-trust-transcript.png)investigate-overview.pnginvestigate-highlights.pnginvestigate-evidence-jump.giflocal-ai-llm-setup.pnglocal-ai-overview-summary.pnglocal-ai-meeting-extracts.pngexport-artifacts-browse.pngexport-panel.pngexport-download.gifexport-html-preview.png
Embed images with GitHub-compatible Markdown so they render in the GitHub file view and in Sphinx/MyST:

Paths are relative to docs/workflows/. Alt text is required; captions are optional and only when they add information beyond surrounding prose.
The public landing hero is a copy of first-analysis-overview.png at website/images/overview.png. When you recapture Overview, copy it there as well.
Rebuild docs¶
pip install -e '.[docs]'
make docs
# open docs/_build/html/workflows/index.html