A daily puzzle can be shipped without a database, an account system or a backend. That choice is what makes the first round free.
A daily puzzle has one unusual property as a product: its state changes once a day, for everyone, in the same order. That makes it a perfect candidate for a static-first build, where the day's puzzle is generated ahead of time and served as a plain file, with all the interactive rules running in the player's browser.
The advantage is not hipster simplicity. A static page is loadable in one request, works before any script hydrates, has no session to create, and cannot leak a player's data because it never receives any. The daily reset becomes a build step instead of a cron job that can fail at midnight. For a game that people play for ten minutes on a train, those properties translate directly into the experience: the board appears almost immediately and the first play is free of decisions.
The costs are equally real. Anything that must be shared across players has to be precomputed, so leaderboards and streaks require a backend later. Verification of a submitted route cannot be trusted to the client, so anything competitive needs a server. And the archive must be generated on a schedule, which means a failed build is a failed day, so the pipeline needs a health check that actually pages someone.
The interesting part of the design is the interface between the static part and the interactive part. The route acceptance rules live in the client because the puzzle is not competitive. The word lists and the route data live in the daily payload. The solver runs in the browser because the graph is small enough to ship. Every one of those choices is defensible precisely because nothing of value is being protected.
Next time you play, notice that the board is usable before anything feels like it has loaded. That is the design paying for itself. the unlimited mode benefits even more, because random practice starts are generated client-side and need no state at all.