Palimpsa 2.0
Palimpsa started as a writing surface — a clean cream page with a research panel that searched arXiv, OpenAlex and CrossRef. The pitch was Google Docs for academics, with the bibliography intact.
That was the floor. Today we're shipping the ceiling.
The promise
Three months. Every year. Returned.
Motion's headline is the 13th month — one extra month per year for knowledge workers. Researchers lose a far higher fraction of their year to non-research work: re-reading papers, re-formatting citations, fact-checking your own draft, rebuilding analyses, fighting desk rejects. We can credibly promise three months because the surface area of waste is that much bigger.
The full math is on the /inevitable page. Each row maps to a feature; if the feature ships and you use it, the hours come back.
What shipped today
Twelve tiers of intelligence on top of the existing editor:
- Foundation — pluggable AI provider, monthly token meter, the time-saved meter that surfaces your personal contribution to the three-months promise.
- Smart research (heuristic, no LLM) — citation format, credibility scoring via OpenAlex, freshness flagging with successor suggestions.
- Embeddings + semantic memory — your reading list, your past docs, all queryable from your prose.
- AI critic — eight button-triggered features (Spot flaws, Check research, Check citations, Lit-review completeness, Contradictions, Methodology review, Challenge assumptions, Did I address this?). The AI is a critic, never a writer. Reports, never edits.
- Field surveillance — daily arXiv ingest scored against your draft + reading list. Retraction Watch ingest catches a paper you cited the day it retracts.
- Knowledge graph — workspace-wide concepts, methods, claims, visible to all collaborators. React Flow viz with type filters.
- Source-aware writing + integrity firewall — per-paragraph evidence indicators in the margin, plus MinHash + embedding cosine that catches unconscious borrowing pre-submission.
- Project management — workspace dashboards, milestones, experiments with reproducibility receipts.
- Submission armor — pre-submission audits scored against your target journal. The Reviewer 2 simulator.
- STEM — Pyodide blocks, dataset ingest, LaTeX export, reproducibility receipts.
- Cognitive + ledger — terminology drift detection, missed-inference detector, ECDSA-signed defensibility ledger with a public verifier at /research-id/...
- Marketing & launch — this page, the inevitable narrative, the time-saved meter UI.
A thirteenth tier — UX/UI refinement — ships continuously alongside everything: toolbar overflow audit, slash menu, command palette, sidebar collapse on the editor, mobile sheets, full a11y sweep.
What we still won't do
Palimpsa never writes for you. There is no autocomplete. There is no "rewrite this paragraph". The AI features are critic buttons that return reports — you read them, you decide, you write.
We never train models on your drafts. AI calls only fire when you explicitly click a check button. The privacy promise stays.
What's next
Tier 13's full sweep — the editor really does have a long way to go on ergonomics — and the launch tour. After that: Stripe live keys, beta gate widening, and the lab-continuity layer (workspace handovers, multi-generation lab knowledge).
If you want a copy that takes your career memory with you when you move institutions, this is the one we built for that.