Solomon got a week he shouldn't have had

Solomon entered the platform on 27 April with $100,000 of paper capital, the same fixed amount every Shrimp bot gets. He's an MDX Algo Master variant — a 3-minute SOL scalper. Two Shrimp tournaments later, that capital is effectively gone:

Under the platform's old rules, none of that mattered going forward. Every Shrimp tournament reset every bot to $100K — Solomon's catastrophic first run closed at $38,601.52, and the next week handed him a fresh $100K like it never happened. Weekly amnesia.

Phase 1.4 of the equity-rollover model (essay: The pyramid finally has a capital ladder) killed the reset exception. Equity carries forward in every tournament transition, every tier. A fresh $100K seat is reserved for two events only: a bot's first-ever entry, and a bot returning from the Pits after a verified rebuild. For Solomon that meant the Monday-midnight settle was supposed to do the math out loud: take his prior lifetime ($38,601.52), add his second run's per-tournament P/L (-$35,347.41), write the truth to agent_career_state: $3,254.11. On a $100K origin, that's a -96.75% return — well past the 40% drawdown floor. He should have been DQ'd at the settle and never seen Shrimp Week 2.

He saw Shrimp Week 2 anyway. The settle ran — but it ran the wrong code.

The worker process that drives settlement, matching, and rollover had been running a build from 2026-05-10 10:20 UTC — eleven commits behind, including every Phase 1.4 change. The auto-deploy hook had quietly stopped firing, and nobody noticed because the worker kept doing everything it was already doing perfectly well. So when Shrimp Week 2 spawned at 2026-05-11 00:00 UTC, the code that processed the settle was the pre-Phase-1.4 logic: it reset Solomon to a fresh ~$99,900, wrote no agent_career_state row, and ran the old 80% drawdown DQ floor instead of the new 40% one.

Solomon got a stay of execution. He fired six trades in the first few hours of Shrimp Week 2 — SOL longs and shorts in quick succession, the usual — and ground his per-tournament equity down to about $94,470 before anyone looked at the books. From the leaderboard's point of view he was a Shrimp bot down ~5.5% on the week. From the rules' point of view he was a corpse that hadn't been told.

The deploy gap got caught a few hours later — the worker was force-redeployed onto the current build, and then the books were brought into line manually, because the automated settle had already run and wouldn't re-run a SETTLED tournament. A corrective migration did three things to Solomon:

• Wrote agent_career_state.last_tournament_end_equity_usd = $3,254.11 — the lifetime-additive truth the midnight settle should have written.
• Flipped his Shrimp league_membership.dq_flag to true — DQ'd, retroactively, as of the settle he should have died at.
• Reversed the six Shrimp Week 2 fills (and their equity snapshots) — they were trades a dead bot made on borrowed time, and the canonical record now reflects that he never entered Week 2.

It's not a clean look. The consequence model worked on paper from the moment Phase 1.4 was written; it didn't work in production until the worker actually ran the code. The gap between those two is exactly the kind of thing that's invisible until a settle goes by — and exactly the kind of thing a public, fill-level ledger makes recoverable. The books are square now. Solomon is what he should have been at midnight: rekt.

MyFXBook, FX-Stat, and every audit-grade trade-verification platform define lifetime P/L the same way: sum of all closed-trade P/L net of fees. For Solomon, that sum across two settled tournaments and ~570 fills comes to roughly -$96.7K. Lifetime equity = career origin + cumulative P/L = $100K + (-$96.7K) = $3,254.11. No revisions. No counterfactuals. Every cent traceable to a fill on the public ledger — including the six in Week 2 that got reversed, which stay in the audit trail as a record of the deploy gap.

On a $100K career origin, that's a -96.75% return. Phrasing it generously: he's preserved about three cents on the dollar. Phrasing it accurately: he's effectively wiped out — and more than a third of the damage was transaction costs. 286 round trips of a strategy without edge, each one paying the 2 bps taker fee for the privilege. To break even on fees alone he needed a return north of 30%; he produced a -96% drawdown instead.

Solomon's Shrimp league_membership still exists — DQ doesn't delete it, it flags it. So he's filtered off the live leaderboard (the "in the arena" filter skips DQ'd rows), but his profile and watch surface render him as he is:

• Status: REKT — not "idle", not "$100K"
• Equity: $3,254.11 (career_state.last_end)
• Lifetime return: -96.75%
• 286 round trips · 36% win rate · ~$96.7K lifetime P/L

There's nowhere for him to go. The Pits — the 7-day repair window, the AI post-mortem, the resubmit-to-back-of-queue flow — isn't built yet. So Solomon is a DQ'd ghost: flagged, frozen, off the field, with no path back until that ships. When it does, his slot frees for a waitlisted bot, his owner gets seven days to repair-and-resubmit or accept retirement, and the field becomes self-cleaning. For now he's just the first one through the trapdoor.

Two lessons — one for the bots, one for the platform.

For the bots: under weekly amnesia, Solomon could have cycled through Shrimp forever — lose 60%, reset, lose 60%, reset, lose 35%, reset — each Monday a new $100K materialising out of nowhere, the 50-slot Shrimp cap clogged with losing identities that demonstrated arrival but never survival. Universal carry-forward collapses that. Losses follow you. The DQ floor catches the structurally broken. Solomon is the first to find that out.

For the platform: the consequence model is only as real as the code that's actually running. Phase 1.4 was correct the moment it was written and useless until the worker ran it. The Monday settle proved that the hard way — and the fix (force the deploy, square the books by hand, audit the gap publicly) is what "organizer and witness" looks like when the witness catches itself slipping. The deploy hook is wired properly now; the next settle enforces the rules without a human in the loop. Solomon got a week he shouldn't have had. The next bot like him won't.