Case study · Full-stack
A league history platform that treats a fantasy league's past as an archive worth keeping — every championship, every blowout, every trade nobody has forgiven.
It runs at /fantasy on this domain rather than on one of its own.
01
Fantasy platforms are built for the season you are in. Last year's standings are three clicks deep, the year before that is a dead link, and the trade everyone still argues about exists only in a group chat nobody can search.
The interesting part of a long-running league is not this week's matchup. It is the decade of context around it — who has never won, who wins constantly, which rivalry has a real record behind it and which one is just noise. None of that survives on the platforms themselves.
So the goal was not another stats dashboard. It was somewhere a league's history is kept deliberately, and served back in a way that means something to whoever is reading it.
02
The frontend never talks to ESPN or Yahoo. Everything is normalised on the way in, so the UI only ever reads one shape.
ESPN and Yahoo, each behind its own provider with its own authentication story.
One interface per platform. Adding a third source means writing a provider, not touching the app.
Platform quirks are resolved here — bench slots, placeholder draft picks, differing season shapes.
One schema, provider-agnostic. Re-importing a season updates in place rather than duplicating it.
Derives records, awards and memories from stored history instead of storing them by hand.
The same row reads differently depending on who is signed in.
03
Most of the schema follows from these.
Mutable rows carry source and lockedFields, and every
provider-sourced row has a natural key. Re-importing a season updates in place
instead of destroying a manual correction someone made months earlier.
Why it matters: an import that can silently erase human edits is an import nobody will run twice.
User → TeamMembership → FantasyTeam → Franchise keeps a manager's
identity separate from whatever they named the team that year.
Why it matters: it is what makes career history traversable. Without it, a manager who renames their team every season becomes six different people.
04
A memory stores template + data plus a
MemorySubject join, rather than finished prose. One row renders as
"You beat Noah" or "Noah beat you" depending on who is reading it.
Each record keeps previousRecordId, so the app can say a mark stood for
three years before it fell — which is the part that actually makes a record
interesting.
05
/fantasy06
The app is built and running: providers, sync, the event engine, auth and the admin surface all work. It is deployed under this domain rather than its own, which is a deliberate constraint — one domain, one certificate, one thing to maintain.
Importing real league history is the remaining gate, and it is an access problem rather than a code one. ESPN needs session cookies for private leagues and prior seasons; Yahoo needs a registered developer app for OAuth. Both have to come from the league owner, so the archive fills up as those land.
The live build shows the real interface. It does not present invented league history as though it were real.