Where we landed
DIGEST ships as: pick your arXiv categories + a reader profile, get a tailored newsletter in your inbox at the hour you set. Free for casual reading (3/day, 1 category). Pro at €5/month for unlimited.
That's the product. Everything else was a detour.
What we dropped
The original v0 sketch was a shared daily feed — every user landing on /feed and scrolling through the day's papers. We built a lot of it: per-category feed views, topic subscriptions, sharing surfaces, a fan-out cron pipeline. It ran for ~2 months in early 2026.
We dropped it for two reasons. One technical: the production deploy chain broke in late March (a Sentry import without the dep installed, then a hard / → /feed redirect baked into the build), and digest.ltd became unreachable. Two conceptual: a shared feed is a different product. It's social, attention-driven, and competes with X / Mastodon for the same eyeballs. We didn't want to build that.
The custom-newsletter direction is what arXiv readers actually want — quiet, scoped, in your inbox at the time you read.
What we kept
The constraints were the easy decisions, looking back.
arXiv-only. We considered RSS + general websites + Substack imports. They'd have widened the audience but compromised the source-trust story. arXiv has a stable schema, permissive licenses, and a community that already trusts the source. Adding "arXiv + a random Substack" weakens the brand promise.
5 fixed profiles. Free-text custom profiles were tempting. We dropped them because evaluating summary quality across an unbounded profile space is hard, and the 5 fixed profiles cover the meaningfully-different reading modes. Custom profiles are post-launch if traction warrants.
No tracking, no analytics. We didn't add PostHog or GA "for v1 iteration speed". The honest privacy story is the whole brand promise. We'll measure traction in conversion rate from Stripe and qualitative email replies, not in time-on-page or click-tracking.
Daily 9 AM delivery default. Researchers we talked to all said "the same time every morning" beat "smart inbox prioritisation". Boring, predictable, easy to ignore on a busy day. Pro tier lets you tune the hour; the default is 9 AM in your timezone.
What's coming
The roadmap for the next few months:
- Recipe-builder UI — currently you pick categories during signup; we want a dashboard surface where you can run multiple "recipes" (source + angle + cadence) per account.
- Per-recipe email delivery — Pro tier upgrade. Tied to (1).
- Cross-reference Pro feature deeper integration — the
/how-to-usepage demos this with the Mixtral + Mistral + Switch Transformer triple. The real product surface is in the daily email. - Custom free-text profiles — gated behind quality-evaluation work; not before late 2026.
- Multi-language summaries — post-traction. We'd want native-fluent reviewers per language; not solo-buildable.
Why this changelog tone
Researchers respect honest shipping more than confident shipping. We dropped a direction, we built another, we shipped the second. That's the truth. Future changelog entries follow the same tone: what shipped, what didn't, what's next.
If something in the product looks off or there's a feature you'd actually pay for, email hello@digest.ltd.