[Jammy2211]

The arXiv inbox is strong lensing only. This adds its sibling, arxiv-interests.md: the day's ten most relevant papers that are not strong lensing — black holes, dark matter, galaxy formation, statistics — rendered under the inbox on the dashboard with the same five per-paper actions (📄 PDF, ➕ queue, 📥 intake, 📑 cite, ✖️ dismiss).

Paired with PyAutoMind's arxiv_interests.yml, which writes the papers. This side renders and acts on them; that side fills them in. Neither blocks the other — merged alone, this just shows an empty list saying no digest has run yet.

The one thing that differs from the inbox

It is a day-batched backlog, not a seven-day timer, and that is the point of it. Each nightly run appends one dated batch, the board shows the oldest un-cleared batch only, and one 🧹 clear button drops that whole day and reveals the next. So a fortnight away is a fortnight of batches to cycle through, one tap each, rather than a fortnight of lapsed recommendations.

Everything else is the inbox's contract, reused rather than restated: interests_actions.py imports inbox_actions for the line grammar, the identity/dedup rule, the freshness stamp and add_to_queue.

Topic routing

Every inbox paper routes to one reading-queue section; these span domains, so each line carries a [Topic] naming the section ➕ files it into (SMBHs, Dark Matter, Galaxy Formation / Evolution, Stats). A topic naming no real section falls back to the new ## Interests section rather than stranding the paper — the topic is a nightly prompt's guess, and a guess that misses must not cost a paper.

Changes

• scripts/interests_actions.py (new) — owns the format and the transitions: append / stamp / clear / add / dismiss. append dedupes against the interests list, the reading queue and arxiv-inbox.md, so a paper never lands on both lists, and caps a batch at 10 so a miscounting prompt cannot flood a day.
• arxiv-interests.md (new) — the file and its documented format.
• scripts/board.py — renders the tier between the inbox and the reading queue, with its own freshness stamp (one digest can break while the other keeps running, and the page must not report the healthy one's date for the broken one). _queue_issue_url is generalised over "suggestion tiers" rather than special-casing the inbox a second time.
• reading-queue.md — the ## Interests fallback section.
• .github/workflows/queue_actions.yml — interests-add / interests-dismiss / interests-clear; close-out comments name the tier the human actually tapped.
• .github/workflows/queue_filing.yml — 📥/📑 learn the third file: value and the [Topic] prefix to strip.
• .github/workflows/knowledge_board.yml — republish paths, plus the three new labels (a prefilled labels= silently drops a label the repo does not have, which makes a tapped button look like it worked and then do nothing).
• scripts/validate_structure.py — root allowlist.

Testing

make validate clean; pytest tests/ 168 passed. 34 new tests in tests/test_interests_actions.py cover the batch grouping, the clear transition and its idempotence, the three-way dedup, the topic fallback, and the CLI; 13 new in tests/test_board.py cover the render order, the clear button's body, the per-tier stamps and the topic routing.

Three pre-existing board tests moved with the fixture (which now writes an interests file): the two inbox-freshness helpers blank the other tier so they still assert about the inbox alone, and the PDF-button count goes 4 → 6.

Also exercised end to end against a clone of the live repo — append/stamp/add/dismiss/clear, and the cross-list dedup correctly skipped a lensing paper already sitting in arxiv-inbox.md.

---

Generated by Claude Code