I built a production fantasy-football platform in one week, then shipped it to iOS.
Here's how I made the calls.

01The problem

A live draft with 8 to 18 people has one failure mode: someone doesn't show, and the room sits dead on a clock. Every platform ships an autopick fallback, and every player has a story about how bad it is. I wanted one a commissioner could trust.

02The solution

The autopick encodes real roster judgment, not a best-available list: fill starting slots first, don't stash an injured player over a healthy one at equal value, keep the bench balanced. A genuine strategy function, unit-tested on its own.

But one missed pick isn't the failure mode; a person gone for the rest of the draft is. Miss two picks in a row and you're auto-enrolled: picks fire in about ten seconds, and the room keeps moving without anyone stepping in.

03Key tradeoffs

• The doctrine started too conservative: it wouldn't take a Questionable player over a healthy one even at equal value, and in a test draft that left a genuinely good player undrafted. I relaxed it so only real structural risk gets avoided.
• Enrollment needed an audit trail, not a flag. A ledger records the exact pick that triggered it, so undoing a pick rewinds only that pick, not the draft.
• Real commissioners surfaced a gap no tuning could fix: sometimes a human has to step in. I added overrides (force a pick, toggle autodraft for any owner) routed through the same functions the automatic system already uses, so there's no second code path to drift.

04What shipped

Bot tags on every auto-fired pick, a strike-based enrollment system, ~10-second auto-fire, and commissioner override powers for force-picks and per-owner autodraft toggling.

05Outcome & status

Live in production, verified with 8 state-machine tests plus a live end-to-end draft run. One bug worth owning: a boolean stored as raw 0/1 made the UI render the literal string "0" next to every hand-drafted pick in production. Caught it, fixed it at the type boundary, added a regression test.

01The problem

In a normal league, waivers are a once-a-week afterthought. Here they're the whole game: a full roster clears every week, all at once. Every platform ships the same flat FAAB auction, and it misses the real tension: a $20 bid placed early can beat a $95 bid placed late, because by round two the player's already gone. I wanted the mechanic to teach that, not a help article.

02The solution

Claims are ranked into priority groups: the whole league's round-one claims process together before any round-two claim is looked at, then round two, then round three. "Rank ruthlessly" isn't a tip, it's the rule the processor runs. That one mechanic became the feature's headline pitch.

03Key tradeoffs

• Every claim carries a fallback drop chain, so it never fails for lack of a roster spot. The chain was capped at a hardcoded number that silently truncated valid plans on larger rosters. Fixed by tying the cap to the real roster-capacity math instead of a magic number.
• Waivers run up to four times a week, not once, so a cleared roster doesn't sit dead in a queue waiting on the one weekly window.
• Every run posts a box score (who won which player, for how much, who got outbid) to the league feed. A deliberate call to turn a backend batch job into something people read and argue about.

04What shipped

Grouped, round-based claim processing; fallback drop chains; up to four processing runs a week; the Scramble Report box score in the league feed; mobile touch-reordering for priority groups.

05Outcome & status

Live in production, and the product's real differentiator: the one feature I rebuilt from scratch instead of reusing what every other app ships, because the flat-bid version couldn't tell the story the format needed.

01The problem

Public leagues have a cold-start problem: nobody joins an empty league, but a league needs people to stop being empty. I also drew a line before any code: no real money, on legal and liability grounds. That ruled out the usual fix, since a cash prize does the work of making an empty league feel worth joining.

02The solution

A "house" commissioner account (no human owner, never plays) seeds a standing pool of open leagues, and a driver runs each through its whole lifecycle with no person involved: fill, draft, run the week, finalize, reseed. It reuses the same commissioner code paths a real human already exercises, so there's no separate "fake league" to keep in sync.

03Key tradeoffs

• Two draft-timing modes, because "start immediately" is wrong for half the audience. Slow drafts start on fill, with a long per-pick clock that pauses overnight. Live drafts deliberately do not start on fill (2 of 6 seats online is a ghost town); they schedule a start 24 hours out in a friendly window, so a league that fills at 3 AM gets a Saturday-morning slot.
• The floor-of-three guard from earlier applies here too. I didn't duplicate the fix, I reused the same guard.
• Growth surfaces have to reach people already signed in, not just new visitors. So the "open leagues" module sits on the logged-in dashboard, not only a separate lobby page.

04What shipped

A self-driving house-league pool across three sizes and two endgame modes, two draft-timing strategies, a public lobby page, and a home-dashboard preview module.

05Outcome & status

Soft-launched: live and running real house leagues today, but deliberately not promoted past friends. Moderation and a policy for "what happens when a public league never fills" are still open, and I wrote both down as explicit questions rather than shipping around them.