Adding new discussions (update workflow)

Nothing watches folders or Drive. You provide new material; then one command processes it. Every step skips finished work, so it is safe to run again.

What you provide

New material	Put it here
Recording file	[local path]
Drive link to a recording	a new line in [local path]
A recording date you know	sources/config.json → user_dates
Hand-corrected transcript	mwkb.py import <src-id> <file> --kind manual
New owner source-of-truth version	nothing: fetch_core.py picks it up from the Drive folder

Run it

From [local path] in PowerShell:

machine-world-knowledge\scripts\update.ps1            # everything below, in order
machine-world-knowledge\scripts\update.ps1 -NoAudio   # skip transcription (docs only)

Step	Command	Skips
1. Detect new files and links	mwkb.py scan, mwkb.py fetch, scan	known hashes
2. Find duplicates and cuts	mwkb.py overlap	pairs already compared
3. Decide scope of new sources	mwkb.py transcribe --triage …, classify, then you confirm with mwkb.py scope	—
4. Transcribe in-scope sources	mwkb.py transcribe --pending	done chunks, same-recording twins
5. Extract notes	mwkb.py notes <ids>	sources that have notes
6. Refresh owner's source of truth	fetch_core.py	unchanged versions
7. Rebuild pages, write change report	build_docs.py	hand-written text (kept)
8. Build site, check every internal link	build_site.py then mkdocs build --strict	—

Step 3 asks for your decision. The script stops and lists unreviewed sources rather than guessing.

What changes and where it is recorded

• Weekly page for the recording's ISO week: created or updated; status moves PARTIAL → COMPLETE when its last source is documented.
• Topic pages, decisions, roadmap, open questions, glossary, timeline, source index, home page.
• UPDATE-LOG.md: new sources, week status changes, pages touched.
• Text you wrote outside the GENERATED blocks, and the hand-written pages (overview, feature status, guides), are never changed. Re-check feature status when a new source-of-truth version arrives.

Contradictions and duplicates

• Duplicate audio (same bytes, re-encoded copies, clips cut from a session) is detected in steps 1–2 and transcribed once.
• Similar items within a week are merged in the generated pages, with every source cited.
• Conflicts between sources are not resolved automatically: add them to the hand-written table in open questions.