MarkdownBin
RawNew

Ghosts of the 100x: A Trader’s README

“In bull runs, everyone’s a genius. In liquidation, you learn your true leverage.”
— anon

Build Status
Risk
Leverage
Exchange

Table of Contents

Prelude

It was supposed to be another disciplined day in the green sea of candles.
I’m Eli—ex-quant, now full-time degen with a tidy suite of TypeScript bots, a drizzle-ORM’d ledger, and a vaulted conviction that “this time, we’re early, not wrong.”

The dashboard glowed. BTC flirting above ATH. ETH cruising. Funding richer than almond milk.

I pushed one more deployment on Bun. Linted. Prettier ran. Unit tests green.

“Send it.”

The Setup

  • Stack:
    • Next.js dashboard for signals
    • Bun runtime bots
    • Drizzle ORM for trade logs on Postgres
    • WebSocket feeds, custom liquidation heatmap
  • Strategy:
    • Long bias across BTC/ETH, some SOL perps
    • Cross-collateral, high conviction, too much composure
  • Leverage:
    • Responsible on paper, irresponsible in practice
  • Risk Controls:
    • Soft stops; hard belief

Code excerpts looked pristine. PnL looked inevitable.

Catalyst: The Tariff Bomb

Then came the macro nuke.

  • Former U.S. President Donald Trump floated 100% tariffs on all Chinese imports.
  • Markets flipped to max risk-off.
  • Crypto, the leverage hall of mirrors, blinked first.

References:

  • Coinglass: over $19B liquidated in 24h, the largest single-day event on record [1].
  • CCN breakdown of the event and timeline [2][3].

Detonation: The Largest Liquidation Ever

  • Over $19B in liquidations within 24h; more than $7B in a single hour [2][3].
  • BTC knifed from >$125k to sub-$102k intra-crash on some venues [2][3].
  • ETH plunged >12–16%; altcoins painted the tape red [2][3].
  • Hyperliquid leaderboard showed a massacre; 6,300 wallets in the red, 1,000+ fully wiped, 205 wallets lost $1M+ [1].

Eli’s cross-collateral “efficiency” became an accelerant.
Where I saw “capital efficiency,” the market saw tinder.

The Cascade

It wasn’t one blow; it was a chain:

  1. Funding flips, order books thin.
  2. Tariff shock detonates the first wall of longs.
  3. Oracles, unified accounts, and venue-specific pricing quirks amplify slippage.
  4. Liquidations beget liquidations.
  5. Bots—mine included—turn from alpha to accelerant.

On one venue, a stablecoin depegged locally, collateral marks sagged, and the margin math turned me into a forced seller at the worst ticks. Shorts feasted. I was dinner.

Leaderboard whispers later boasted a wallet banking hundreds of millions shorting into the slide [1][2]. I stared at my drizzle migrations, pristine, and thought: schema’s perfect—balance sheet isn’t.

Post-Mortem

  • My max drawdown model assumed venue independence. Reality: coupling.
  • My circuit breakers paused entries, not forced exits.
  • My stop hierarchy didn’t survive cross-venue gaps.
  • I treated liquidation feeds as signals, not warnings.
  • I overfit to up-only microstructure, underfit to macro shock.

The ledger showed it cleanly:

  • BTC-PERP: closed by liquidation
  • ETH-PERP: closed by liquidation
  • SOL-PERP: closed by liquidation
  • Collateral: atomized

I archived the session logs. The only green left was my CI badge.

Lessons Learned

  • Hard risk > soft conviction:
    • Use hard, venue-native server-side stops (OCO/TP/SL) with out-of-band kill switches.
  • Collateral isolation:
    • Don’t let a single venue’s pricing model rehypothecate your entire book.
  • Oracle and mark risk:
    • Prefer oracle-marked collateral where possible; avoid exotic collateral in stress.
  • Cross-venue hedges:
    • Maintain delta hedges on uncorrelated rails; pre-wire “panic pairs” for immediate flattening.
  • Latency-aware guards:
    • Build kill paths that don’t depend on your own infra under load.
  • Size for dislocations:
    • Model 10–20% gaps in majors within minutes; 30–50% for alts.
  • Practice black swans:
    • Chaos drills: yank a key collateral feed, simulate unified-account mark-down, replay liquidation tapes.
  • Humility > hopium:
    • When tape disagrees, the tape is right.

Sources

Epilogue

I rebuilt the bot with fewer toggles and more brakes.
Swapped swagger for sanity checks.
Kept the README. Pinned the badges.
And left one comment at the top of the repo:

// if (market_is_crazy()) flat();

Because the cleanest code in the world can’t margin-call the market.

Created: 10/14/2025, 2:53:15 PM