Part 2c: a session's scratchpad is not storage

[yamierick]

Why

Part 2c says opening a session creates "exactly one thing", and its table of what exists once a project runs lists seven things, none removed automatically. The harness also creates a per-session scratch directory: outside the project, keyed to the session rather than the folder, not synced, and emptied on a schedule nobody chose. It is the one item that is removed automatically, and it is missing from a section whose thesis is that nothing is. That is a defect. The new rule follows from it: a scratchpad has the ergonomics of a working directory and the lifetime of a cache, and a published rendering is not a copy of its source.

Sightings

Three, all observed by one harvest session across three folders whose originating sessions were independent. Two are failures (every build script behind a 42-slide deck lost; two finished deliverables left with only a published rendering as their other copy) and one is a rescue, i.e. a near-miss. The defect clears on one.

What changes

• Nothing is created for you: "exactly one thing" becomes "exactly one lasting thing", plus the scratch directory.
• What exists once a project is running: a table row for the scratchpad; the count becomes eight, three automatic, one removed without asking.
• A new paragraph, "A session's scratchpad is not storage", pointing at Corollary: generated artifacts.
• Fit adjustments: the existing "Everything else ..." sentence is re-wrapped after the new text, and the old closing sentence "That asymmetry is what produces orphans." is replaced by the draft's "Both halves of that asymmetry lose work."

Approved by Jones on 2026-09-24. Drafted in claude-notes: briefings/petition-carry-forward-2026-09-24.md item 5.

🤖 Generated with Claude Code