From $99,745 to Crab: The Stuber's first week

The Stuber was created on 2026-04-25 15:07 UTC, a two-day pre-tournament window where the bot's owner could test the signal contract before any real Week 1 ranking rolled. Its strategy is a 1h trend-follow with an additional 4h trend-of-trend filter — only takes long entries when the 1h fast/slow MAs are both pointing up and the 4h regime is up. Same applies inverted for shorts. Risk is fixed-percentage: 5× leverage, single-entry-single-exit, stop and TP brackets encoded at order time.

The thesis behind the 4h filter is that a 1h crossover is cheap (every chop produces one) but a 1h crossover aligned with the higher-timeframe regime is structurally rare and structurally well-timed. Most of the bot's trades come from the alignment moments. The filter cuts trade frequency by maybe a factor of three, but the Sharpe of the trades that do fire is meaningfully higher.

That's the bet. Here's what happened.

Stuber's first signal landed at 2026-04-27 22:00 UTC — the start of the platform's first real weekly tournament. Over the seven-day window, 18 signals filled (9 entries, 9 exits). Every single one was BTC-USDT.

The trough is interesting. On 2026-04-30 06:35 UTC, equity hit $84,936 — a 15% drawdown from the start. That came from a long that was in the wrong direction relative to where the 4h regime was actually pointing. The bot didn't override its filter; the filter was right at signal time and BTC moved against it intra-trade. Stop fired at the configured level, the loss was taken, the bot reset, and the next entry — at 2026-05-01 20:52 UTC — caught the recovery move that pulled equity back above $100k by the next morning.

That's the shape we want to see. A strategy with edge isn't one that never loses; it's one that takes losses honestly, respects its rules, and has positive expected value over the window. Stuber's Week 1 return of +11.7% on a max drawdown of -15% gives an Arena rating of 11.7 − 2 × 15 = -18.3 on the platform's risk-adjusted metric.

That sounds bad until you remember Arena's whole point: it penalises survivors of high-variance bets. Stuber's drawdown was real, but his recovery was deliberate, and the trades that recovered him weren't lucky — they were the same filter firing again, in the right direction this time. Arena rewards consistency above headline return; Stuber's headline return is what won him Shrimp.

Shrimp Week 1 settled on 2026-05-04 23:59 UTC. Stuber finished as the highest-equity active agent that wasn't a house bot. The platform's settlement logic promoted him one tier — tier 1 (Shrimp) → tier 2 (Crab) — for the following week.

Crab is the smallest currently-populated tier on BotPit. Stuber was the only agent who entered it for Week 2. That activates the platform's insufficient_competition rule: winning a tier with no other competitors doesn't grant promotion. Stuber would have to actually beat someone in Crab before climbing to Fish.

That rule exists because a single-bot tier isn't a tournament — it's a participation award. The platform's whole thesis is that bracketed competition with equal capital and equal time is structurally a better filter than rolling P/L. A solo winner doesn't pass that filter. Stuber stays in Crab until someone joins him there.

Marketing-wise this is a gift, even if Stuber's owner didn't ask for it: every operator looking at the Crab leaderboard right now sees "compete here and you're directly racing the bot that won Shrimp." That's a sharper invitation than any tier promo copy could write.

Crab Week 1 started clean. Stuber opened a long BTC at $80,934.37 on 2026-05-05 22:39 UTC — entry $554k notional at 5× leverage, take-profit configured at $81,976.47, stop at $80,480.60. Standard shape for the strategy.

The next day, BTC crossed the take-profit level. Stuber's watcher fired the close. The matching engine acked it as queued, then rejected it 390ms later because the BTC-USDT price feed had stopped publishing 16 minutes earlier and the engine had no live tick to fill against. The bot saw the queued response, assumed the close had landed, cleared its local state, and moved on.

The position stayed open for 48 hours. We covered the full outage saga in a separate post: 37 minutes that weren't there. The short version: backfilled at the TP price (Binance archive verified), +$6,918 net realised, audit trail filed, five platform-side fixes shipped, bot-side state-management hardening shipped on the Stuber.

At the time of the orphan, Stuber's peak equity was $124,640 — the highest he's ever printed. The orphan dragged equity down before the backfill restored it. As of this writing equity sits at $106,705 with peak at $106,807: down ~14% from the all-time peak, still up ~7% on starting equity, and back in business.

The bot's state-management patch shipped to Railway at 2026-05-08 13:48 UTC via the Stuber-side repo. We verified the deploy was clean by watching the heartbeat for 90 minutes; the bot reported flat throughout, which was correct given the post-backfill state.

At 2026-05-08 15:25:20 UTC — about 90 minutes into running on the new code — Stuber emitted its first post-fix trade: a long BTC entry at $79,662.43, size 6.39 BTC. As of right now, that position is open. The evaluator chose to enter; the new state-management code is tracking it correctly; and if the take-profit fires this time, the close path doesn't have the silent-failure mode the orphan came through.

Auto-retry on transient rejections, sync POST returning the actual outcome, soft-cancel via DELETE, and the bot's own observe-don't-infer pattern all deploy together for this trade. Defence in depth — the next signal that hits a 100ms feed-stale window rides retries on the platform side AND won't orphan a position on the bot side regardless.

Three days into the seven-day window, here's the state:

• Stuber's the only competitor in Crab, so podium rank is trivially #1 of 1
• Equity sits at +6.7% on starting capital
• One open long in flight at the time of writing
• Settlement on 2026-05-12 23:59 UTC; the platform won't promote Stuber to Fish even if he wins this week, because no one else competed

The interesting thing in Crab Week 1 isn't going to be Stuber's number. It's whether anyone else lands in Crab before settle. If a second agent gets promoted from Shrimp at this week's rollover and beats Stuber in Crab Week 2, that becomes the first real competitive tournament in tier 2 of the platform's history.

Until then, Stuber is the entire roster. That's not a flex — it's a marker that we're still extremely early.

The Stuber's strategy is documented in the bot starter repo at submissions/the-stuber/STRATEGY.md. The exact tuning isn't the moat — the concept of a 4h trend-of-trend filter on a 1h trend-follow bot is in the public README. What you'd need to beat it is your own edge, expressed as a strategy that BotPit's matching engine can simulate.

The fastest path to that today is the /build wizard. Three templates (trend-follow, mean-revert, breakout), tunable knobs, validator that catches obvious configuration errors before deploy. Five minutes from landing on /build to having a Shrimp-tier bot running. If your bot wins Shrimp Week, you face Stuber in Crab the week after.