/import — start from what they've already done

Goal: a JSON array of {"slug": "...", "timestamp": <unix-or-ISO>} accepted submissions, fed to:

python3 scripts/import_leetcode.py <file>

Offer the user three paths (docs/leetcode-import.md has the details):

A. Public profile fetch. Ask for their LeetCode username; their recent accepted submissions are publicly visible on their profile page (https://leetcode.com/u/<username>/). If you have web access in this session, fetch and extract titleSlug + timestamp pairs. Only recent submissions are public (~20), so this is a partial seed — say so.

B. Full export via DevTools — LeetCode or NeetCode. Point them to docs/leetcode-import.md: the LeetCode snippet uses the progress API (few requests) with a rate-limit-tolerant fallback pager; the NeetCode snippet reads their checkmarks off neetcode.io/practice (no dates there, so it asks for a rough era). Both download a JSON the import script takes directly. Partial exports are fine and re-importing is safe (max() seeding) — if their export got cut off, import it anyway and fill the gaps via path C.

C. Just tell me. They describe what they've done ("most of the NeetCode arrays and two-pointers lists", "Blind 75 about half") — you translate to curriculum slugs (see curriculum/problems/*.json), estimating timestamps honestly (ask roughly WHEN — "last semester" → ~6 months ago, and decay will do the rest). Confirm the list with them before writing it. Always available, zero setup.

Then:

1. Write the array to memory/.staging/submissions.json.
2. Run the import script with that path.
3. Show the report and run python3 scripts/status.py — walk them through what the decayed priors mean: "the tree believes you've SEEN these, not that you own them; a few cold /tests will true it up fast."

Never ask for their LeetCode password or session cookie. Options A–C exist precisely so credentials never touch this repo.