Problem
Model training can produce exciting artifacts, but without lineage, gates, and replay context it is hard to trust which fighter is actually better.
Street FighterBots
A bot arena where training evidence becomes watchable competition.
Street FighterBots is the product layer around the SFB runtime: training fighters, preserving seed lineage, running tournaments, watching replays, and turning bot experiments into a usable arena.
From a technical product-management view, the important shift is from one-off experiments to an operator loop: launch training, validate candidates, enter tournaments, review evidence, and promote only when the gates prove it.
Product
SFB-Control gives the work a user-facing surface for jobs, seeds, tournaments, tactics, live streams, and replay review.
System
The platform splits execution across the Alienware training runtime and Raspberry Pi tournament/control plane, with runtime queues and evidence artifacts.
Outcome
The arena should make bot progress legible: what trained, what fought, what won, what regressed, and why a champion was or was not promoted.
Technical product direction
Make every promotion decision evidence-first and replayable.
The next product direction is a tighter loop across the phone app, control plane, training runtime, and tournament runner so Cristian can launch, monitor, compare, and watch the full bot lifecycle without reconstructing state from logs.