Paper-trading fleet · Volna family + BlackIce · 03/21/2026 – 05/26/2026

Volna Fleet Ledger

Everything Igor's trading bots did in 2026, why each one was stopped, and what the numbers actually were. No real money was ever traded. Every figure below is tagged with the kind of evidence it came from, because the three kinds disagree by orders of magnitude.

As of 09/25/2026 11:50 AM ET Shared Alpaca paper account · funded $100,000.00 on 03/23/2026 BlackIce paper ledger · $11,000.00 on 03/21/2026
Simulator Internal simulatorThe bot priced its own fills from Yahoo or Tradier quotes and kept its own ledger. No broker involved. This is where the seven-figure swings live. Treated as fiction.
Paper Alpaca paper accountReal order routing, real fills at real quotes, fake money. The only tier used for the P&L tables on this page. Sniper, Neutral and Today were paper from day one; Option only from 04/21.
Backtest Historical replayWalk-forward and stress runs on Polygon data, 2019–2026. Used for the go/no-go decisions. Backtests with in-sample tuning are marked.
Alpaca account now
$103,969.33
+$3,969.33 · +3.97% since funding
When the bots stopped · 05/26
$100,535.82
+0.54% after 9 weeks of trading
SPY, same window
+17.73%
03/23 close $652.07 → 09/25 $767.67
Bots running
0 of 8
Project on hold since 05/26/2026

Account equity vs SPY, indexed to 100 at 03/23/2026

Everything after the dashed line is drift on two leftover positions, not bot trading: 100 NVDA assigned to the Option bot on 06/26 and 10 QQQ left open by Sniper.

Alpaca paper accountSPY05/26 all bots stopped

Performance per bot, year to date Paper

Each fill on the shared Alpaca account is matched FIFO to the bot that opened the lot, so a position one bot opened and another closed is credited to the opener. The sum reconciles to the broker within $1.89.

Total P&L per bot, $

BotActiveFillsRealizedUnrealizedTotalOn ledger
VolnaOptionoptions, Alpaca from 04/2304/23–06/26196−891.00+2,886.50+1,995.50+2.00%
VolnaSniperintraday, 289 round trips03/27–05/221,458+292.95+393.98+686.93+0.69%
VolnaTodayone trade: SMH 25 sh04/10–04/202+625.020.00+625.02+0.63%
VolnaNeutral4 pair trades04/06–04/2066+589.350.00+589.35+0.59%
Passive sleeveMTUM + JEPI, Path X04/29–04/303+90.370.00+90.37+0.09%
Manual cleanup + fees26 cleanup fills, 270 TAF fees03/27–04/3026−19.730.00−19.73−0.02%
Alpaca account03/23–1,751+1,686.96+3,280.48+3,967.44+3.97%
BlackIcecrypto, own $11,000 ledger03/22–04/2970−404.650.00−404.65−3.68%
BlackIce funding-arbBinance + Hyperliquid paper04/29–05/262−0.31+0.30−0.010.00%
VolnaArb · VolnaCarrydry-run scaffoldsnever00.000.000.00—

Realized P&L by month Paper

Broker-verified, per opening bot. The account had no activity before 03/23, so year-to-date equals since inception.

BotMarAprMayJunJul–SepTotal realized
VolnaSniper−116.92+207.74+202.13——+292.95
VolnaNeutral—+589.35———+589.35
VolnaToday—+625.02———+625.02
VolnaOption—−330.00−983.00+422.00—−891.00
Passive sleeve—+90.37———+90.37
BlackIce (own ledger)−388.83−15.82———−404.65

How the effort was wound down

The bar was set on day one and never met. Each step below narrowed the fleet until nothing was left running.

  1. Goal
    ≥ 12% a year net of fees and slippage, on out-of-sample data. Sized for about $50,000, so roughly $6,000 a year. Backstop: under 8% net out-of-sample kills the strategy. Advisor kill-switch from 04/25: Sharpe ≥ 0.5 and ≥ 5 trades a week, or stop.
  2. 03/30
    BlackIce diagnosed as having no edge. Nine days in, −10% with a 21% win rate. The improvement plan's root cause: lagging hourly indicators, a seven-layer pipeline that adds complexity but no alpha, an ML model trained on 33 trades. Phase 1 was "stop the bleeding".
  3. 04/18
    Focus on two bots. VolnaToday, VolnaNeutral and BlackIce frozen with systemd disabled, "out of focus portfolio, code preserved for reference". Sniper and Option kept because they traded most and had a live-prep plan (first live trade pencilled for 05/05).
  4. 04/25
    The edge gate. A day of 26 fixes made the engineering "live-ready", and a corrected 3-month walk-forward then showed −$1,581.46 over 346 trades with 1 of 13 weeks profitable. Igor: a stable, well-tested bot that loses money is still a failure. Go-live paused pending proof of edge.
  5. 04/28
    Data sprint: 0 of 7 strategy variants passed out-of-sample. ORB, Gap Reversal, three PEIDS variants, PEAD and the multi-ETF iron condor all failed. Verdict: "HOLD ALL strategies… Right now, stop." Sniper stopped and disabled at 2:14 PM ET. Three advisor models (GPT-4o, Claude, Gemini) picked crypto funding-rate arbitrage as the one idea worth paper-testing and killed the momentum rebuild.
  6. 04/29
    Per-bot return table at $50,000 each: Binance funding-arb 21.5%, Hyperliquid funding-arb 44.2%, Sniper 1.91%, Option −0.27%. Same day: Binance.com is closed to US persons, so the Binance branch was frozen as a baseline and a Hyperliquid branch built beside it.
  7. 04/30
    Account cleanup and the Option kill. Igor: "do whatever will help us get the right decision." Passive MTUM/JEPI sold, every option leg closed, cash back to $101,175. Decisions: Option = kill, Sniper = keep for four weeks of observation, funding-arb = keep. Option was switched back on 05/05 after the drawdown that triggered the kill turned out to be a stale-peak bug, then halted itself again on 05/13.
  8. 05/26
    Project on hold. After seven-plus weeks the fleet stood at +0.40%, about 2.4% annualized, against the 12% bar. Infrastructure fixes had worked (no halts, no equity drift, broker-P&L sync) but produced no edge. Igor: "stop bots and all processes related to bots project, this effort on hold for now." All services disabled, crons commented out, open positions left to expire. Nothing has run since.
  9. 09/09
    AWS account closed. The EC2 hosts of the original dashboards went with it; volna, volnaoption, volnatoday and volnasniper subdomains no longer resolve.

The bots

For each: what it does, the phases it went through, every number that was reported for it, and the decision that stopped it.

VolnaSniper

v1.2.3 → v1.3 · config baseline v3.3.1 · Node.js
Stopped 05/26/2026 · on hold

Intraday momentum bot for US large caps. Four setups: opening-range breakout (ORB), VWAP pullback continuation, VWAP reclaim and EMA9 re-entry. Scales out half at 1R, trails the rest, flat by 3:55 PM ET. Universe SPY, QQQ, NVDA, TSLA, AMD, AAPL, later AMZN and GOOGL. Risk 0.25% per trade, max 3 positions, 10 trades a day, 1.5% daily loss halt, 5 consecutive losses halt. Data from Polygon 1-minute bars, execution on Alpaca paper.

Timeline

  • 03/27First paper fills on Alpaca. Paper from day one, never a simulator.
  • 03/30Moved from AWS EC2 to contabo (.92).
  • 04/18Bug sprint: only 1 of 4 strategies had been active (VWAP reclaim); ORB was a stub, VWAP pullback had broken arguments, EMA9 trail was dead code. All four wired up in v1.2.3.
  • 04/19–25The .92 copy was down; fills kept arriving from a second copy on the production box (.49), which explained the "ghost sniper". From 04/25 all bot work moved to .49 only.
  • 04/2527 findings fixed (Polygon bid/ask swap, paper fill-price drift, partial fills, kill switch only checked at startup). Walk-forward run the same day came out negative.
  • 04/28Stopped and disabled at 2:14 PM ET after the data sprint. "Sniper definitively dormant. Code preserved on disk."
  • 04/30Reversed to KEEP for a four-week observation on a clean account.
  • 05/22Last trade. Service disabled 05/26 with the rest of the fleet.

All the numbers

TierMetricValueDate
PaperBroker-verified, 03/27–05/22, 289 round trips+$292.95 realized, +$686.93 with QQQ09/25
PaperBot's own trade log, 289 trades, 54.3% win, profit factor 1.04+$206.2505/26
PaperBy setup: VWAP pullback 119 trades / ORB 53 / EMA9 26 / VWAP reclaim 4 / untagged 87+$90.67 / −$199.05 / +$25.78 / +$25.37 / +$263.4805/26
PaperGolden week 04/14–04/18, 28 trades, 57.1% win+$264.5004/25
BacktestSame week after the lunch-gate fix, 26 trades, 42.3% win−$87.7204/25
Backtest3-month walk-forward, 13 windows, 346 trades, 1 of 13 weeks profitable−$1,581.4604/25
BacktestStress: 2022 bear 651 trades / sideways Apr-2024 167 / Aug-2024 VIX 61−$5,460 / +$347 / −$77304/25
BacktestORB only, TP 2.0R: in-sample vs out-of-sample Q4-2025+$1,692 / −$1,98904/26
BacktestMulti-period, 1,277 trades over 1.59 years: 2025 +1.28%, 2026 YTD +2.35%CAGR 1.91%04/29
BacktestGap Reversal, the proposed repurpose: 117 trades, 31.6% win, Sharpe −3.57−$6,68104/28
Why it was stopped

No out-of-sample edge. Every in-sample win evaporated on fresh data; the one profitable regime was a bear market. The 04/28 data sprint killed it, the 04/30 cleanup gave it four more weeks, and the 05/26 hold ended those weeks with the whole fleet at +0.40%.

"Verdict: No edge under current config. All 4 strategies underperform OOS."Master summary, 04/28/2026

"Мой вывод: нет, структурно. Не код плохой — а класс стратегий мёртв для retail в 2026."Sniper pivot pitch, 04/28/2026

"Sniper is regime-dependent — barely positive in calm, profitable in stress. Sharpe ~0 in calm. Real money case is weak."Per-bot ROI review, 04/29/2026

To resume: explain why 2026 outperformed 2025 (were the R1–R5 fixes the reason?), tune the low-volatility and too-late filters, and run it only with a small allocation that scales up in a bear regime. The code and the 05/26 state are preserved on the production box.

VolnaOption

v1.0 → v1.4 · Node.js
Killed 04/30 · restarted 05/05 · stopped 05/26

Options income bot. Sells iron condors and straddles/strangles on liquid ETFs and mega caps when IV rank is high, with vertical spreads, butterflies, covered calls and the wheel as secondary sleeves. Watchlist SPY, QQQ, IWM, AAPL, MSFT, GOOGL, AMZN, TSLA, NVDA, META, AMD, GLD, TLT, later SLV, USO, UNG, DBA. $100,000 notional, max 15 positions, 50% margin cap, 10% drawdown kill switch. Chains from Tradier and Yahoo; an internal Black-Scholes simulator until 04/21, Alpaca multi-leg paper orders after.

Timeline

  • 03/24v1.0 starts on Yahoo Finance quotes, no Tradier token, no broker.
  • 03/25–27The seven-figure days. Simulator marks the book up to $284,425.55, then down to −$892,216.40 (day P&L −$1,181,915), then closes 03/27 at −$665,670.45. Four butterflies are "emergency closed" 1,347 times because closes never left the state. Total booked: −$1,947,008.00. Alpaca did not move.
  • 03/28Restart resets capital to $100,000. Butterfly disabled later after "399 trades, 1 win".
  • 04/1812 bugs fixed. Greeks had been null the whole time, so the rolling defense never ran. Simulator "+$12.8K in 17 days" cited as proof it works.
  • 04/21Wired to Alpaca paper options at 5:49 PM ET. Seven simulator positions carried over.
  • 04/23–30Kill switch trips five days running. Iron condors with a zero-width call wing found and fixed; risk had been understated 2–4x.
  • 04/288-year multi-ETF iron-condor backtest killed: CAGR −0.27%.
  • 04/30KILL. Service stopped, QQQ and TSLA condors closed by hand.
  • 05/05Turned back on ("fix and make it running") once the 04/30 drawdown proved to be a stale peak-equity value: real drawdown 0.46%, not 11.8%.
  • 05/13–15Internal ledger drifts $10,000–$40,000 from the broker; 15.6% drawdown halt on 05/13, 25–40% warnings on 05/15. Broker equity was flat around $100,300–$100,900 the whole time.
  • 05/26Stopped with the fleet. One straddle on NVDA left open; its short 195 put was assigned 06/26.

All the numbers

TierMetricValueDate
PaperBroker-verified option trading, 04/23–06/26, 196 fills−$891.0009/25
PaperAssigned NVDA 100 sh at $195, marked $223.865+$2,886.50 unrealized09/25
PaperBot's narrower tracker: 30 trades (24 IC, 5 straddle, 1 wheel), 7 wins+$798.9005/26
SimulatorBot's position ledger, 89 closed positions, contradicted by its own equity-sync log+$19,632.7005/26
Simulator"Paper: +$12.8K / 17 days" claim (3–5 GLD trades, simulator period)+$12,80004/18
SimulatorInternal account on the .92 copy at shutdown−$2,365.8004/25
SimulatorPeak / trough / worst day P&L, 03/25–03/27$284,425.55 / −$892,216.40 / −$1,181,91503/26
SimulatorEmergency closes 03/25–03/27, 1,347 events on 4 butterflies−$1,947,008.0003/27
SimulatorDay P&L reading with capital pinned at $100,000−$2,395,64503/30
BacktestA3 multi-ETF iron condor, 8 years, 206 trades, 62.1% win, max DD −10.6%CAGR −0.27% · Sharpe −0.1504/28
BacktestA1 long-vol calendar, 60 trades, Sharpe 0.75CAGR 0.98%04/28
BacktestButterfly sleeve399 trades, 1 win04/28
Why it was stopped

The strategy class has no return at this size. Defined-risk premium selling on liquid ETFs backtests flat over eight years, and the live paper run matched: five kill-switch days in a row, then a ledger that could not stay within $10,000 of the broker. The 04/30 kill was partly triggered by a bug, so it was reversed, but nothing that followed changed the verdict, and the 05/26 hold made it final.

"🛑 Option bot = KILL (not revise): 8-year backtest: −0.27% CAGR, Sharpe −0.15 / Live paper: kill-switch trip 5+ days running (23, 24, 28, 29, 30 April) / Pattern: emergency closes at '184% of max profit' suggests sizing/calc bug on naked legs / Future: don't revive until strategy redesigned from scratch."Master summary, 04/30/2026, executed on Igor's instruction "do whatever will help us to get the right decision"

"Verdict: KILLED. Defined-risk vol-selling on liquid ETFs does not deliver target returns."A3 backtest, 04/28/2026. The advisor had pitched 12–17%.

"Option bot's live paper performance (5 days hitting kill-switch) is consistent with backtest forecast (−0.27% CAGR) — strategy is genuinely broken in calm regime, not just bad luck."04/30/2026

To resume: only with a strategy redesigned from scratch. The multi-leg order path, risk checks and broker sync are built and can be reused; the alpha is not there. Separate paper accounts per bot were proposed and rejected by Igor: production will be one Level 3 account, and paper must mirror it.

VolnaNeutral

v1.0.1 → v1.5.1 · Node.js
Frozen 04/18/2026

Market-neutral statistical arbitrage. Trades the spread between cointegrated pairs on z-score: enter at ±2.25, exit at 0.2, stop at 3.5, half-life and volume filters, three-day earnings blackout. Universe of eight mega-cap pairs (NVDA/AMD, V/MA, HD/LOW, XOM/CVX, GOOGL/META, JPM/BAC, COST/WMT, MSFT/AAPL); the pairs that actually traded were QQQ/XLK, KO/PEP and GDX/GDXJ. Risk 0.40% per trade, max 4 pairs, 40% gross. Polygon hourly bars, Alpaca paper, health server on :3002.

Timeline

  • 03/25v1.0.1 starts in "LIVE PAPER" mode on AWS, migrated to contabo 03/30.
  • 04/06First pair opened: QQQ/XLK long spread.
  • 04/07Reused client order id blocks exits for 35 minutes ("client_order_id must be unique"); the retry loop sold 8 QQQ the bot no longer held, taken from Sniper's inventory.
  • 04/09–17Four pairs closed, all green.
  • 04/18Frozen, systemd disabled, in the focus-on-two-bots decision. Last heartbeat 10:35 AM ET.
  • 04/20Leftover XLK and KO legs closed by hand.

All the numbers

TierMetricValueDate
PaperBroker-verified, 04/06–04/20, 66 fills+$589.3509/25
PaperBot's own log: QQQ/XLK +$108.23 and +$264.56, KO/PEP +$53.86, GDX/GDXJ +$0.43, partials +$146.23+$573.3104/17
BacktestNo backtest results were recorded in the project history——
Why it was stopped

Not for losing. It was frozen on 04/18 as part of narrowing the fleet to the two bots with a live-trading plan. Four trades in two weeks is too few to judge, and the exit bug on 04/07 showed the execution layer still needed work. Nobody came back to it before the 05/26 hold.

"Decision: focus on 2 bots. VolnaToday, VolnaNeutral, BlackIce frozen (systemd disabled)."Portfolio decision, 04/18/2026

"Out of focus portfolio. Code preserved for reference. Not running."Master summary, 04/18/2026

To resume: fix the duplicate order-id exit path, add a backtest (none exists), and run it long enough to get 50+ pair trades. Code and state are on contabo; the static dashboard still loads but its API is dead.

VolnaToday

v1.0.0 → v1.5.1 · Node.js
Frozen 04/18/2026

Daily momentum breakout on sector ETFs: SPY, QQQ, IWM, XLK, XLF, XLE, XLI, XBI, SMH. Buys Donchian-channel breakouts confirmed by ADX above 25 and volume 1.5x its 20-day average, scanning at 10:15 AM and 3:50 PM ET, holds one to five days with a 1.5x ATR stop and 2.5x ATR target. Risk 0.5% per trade, max 5 positions, 10% drawdown halt, VIX pause above 35. Polygon daily bars, Alpaca paper, health on :3003.

Timeline

  • 03/25v1.0.0 starts; v1.5.1 followed six review rounds with GPT and Gemini (backtest Sharpe 1.07).
  • 04/10Only signal: long SMH 25 sh at $438.30 on a breakout.
  • 04/11–17Daily scans return "no signals": low volume, flat regime.
  • 04/18Frozen at 10:39 AM ET with SMH still open.
  • 04/20SMH closed by hand at about $463.30. The bot never booked it.

All the numbers

TierMetricValueDate
PaperBroker-verified, one trade, SMH 25 sh+$625.0209/25
PaperBot's own ledger (position was still open when it stopped)$0.0004/18
BacktestPre-launch backtest after six review roundsSharpe 1.0703/2026
Why it was stopped

Same 04/18 decision as VolnaNeutral: out of focus. One trade in three weeks means the strategy was never really tested; the +$625.02 is a single lucky semiconductor breakout, not evidence.

"Decision: focus on 2 bots. VolnaToday, VolnaNeutral, BlackIce frozen (systemd disabled)."Portfolio decision, 04/18/2026

To resume: it needs months, not weeks, to produce a sample; a daily breakout system on nine ETFs fires a few times a month. Code and state preserved on contabo.

BlackIce

v1.0 → v1.6 signals · funding-arb v0.2 · Python
Signals frozen 04/18 · funding-arb stopped 05/26

Crypto agent in two lives. Life one: a seven-layer signal pipeline (DCA, momentum, mean reversion, scalping, grid, portfolio rebalancing, ML scoring) on hourly candles for BTC, ETH, XRP, SOL, ADA, DOGE, AVAX, PEPE, priced from Coinbase, paper ledger of $11,000. Life two, from 04/28: delta-neutral funding-rate arbitrage, long spot and short perpetual, collecting funding when the annualized rate clears a threshold, with a Binance engine (via a SOCKS tunnel to an EU proxy) and a Hyperliquid engine built beside it.

Timeline

  • 03/21Paper ledger starts at $11,000 on AWS EC2 (eu-west-1).
  • 03/24v1.5 "final AI consensus build": backtest loss in a −30% BTC quarter cut from −17.6% (v1.0) to −0.7%.
  • 03/30Improvement plan: −10% after nine days, 21% win rate, "root cause: no edge". Phase 1: pause trading, keep monitoring. Migrated to contabo the same day.
  • 04/18Frozen with Today and Neutral.
  • 04/28Audit: 4 ghost DCA positions worth $2,968, 6 critical bugs (daily-loss check failing open, non-atomic close). Ghosts closed for −$104.40. Funding-arb skeleton deployed in paper mode that evening; enabled 9:23 PM with $1,000 and max 5 pairs.
  • 04/29"Don't re-enable" the signals bot. Binance found closed to US persons; branch frozen as baseline, Hyperliquid engine built the same afternoon. Manual test pair BTC opened and closed (−$0.31).
  • 05/11Hyperliquid opens its only pair, LINK, on funding.
  • 05/26Both engines and six crons stopped with the fleet. Site returns 410 Gone; the AWS instance is gone with the account.

All the numbers

TierMetricValueDate
PaperSignals phase, 70 trades 03/22–04/29, 22 wins (31.4%), profit factor 0.36−$404.6504/29
PaperBy strategy: rebalance 18 / momentum 10 / DCA 36 / mean reversion 3 / scalping 3−$282.81 / −$81.24 / −$23.37 / −$12.41 / −$4.8404/29
PaperBy coin: ETH / XRP / SOL / BTC / DOGE / PEPE / AVAX−$126.54 / −$107.63 / −$93.19 / −$39.77 / −$34.16 / −$3.75 / +$0.3904/29
PaperSnapshot at nine days: about $9,900, 7 wins / 26 losses−10%03/30
PaperFunding-arb Binance: 1 manual test pair; 0 automatic entries in 4 weeks (APR below threshold)−$0.3105/26
PaperFunding-arb Hyperliquid: 1 LINK pair, funding +$0.38, price −$0.08+$0.3005/26
BacktestSignals v1.0 → v1.5, 90 days to 03/24/2026 (BTC −30%)−17.6% → −0.7% · PF 0.62 → 0.8103/24
BacktestFunding-arb A2 v2, 2019–2026, 98 pairs, 89.8% win, max DD −0.99%; calm 2024–26 regime 30.94%; bear 2022 zero trades; 51 liquidations at a 30% bufferCAGR 30.94% calm04/28
BacktestSame strategy vs QQQ over 6.63 years: 2022 +0.9%, 2025 +0.8%, 2026 YTD 0.0%CAGR 12.68%04/28
BacktestConfig B sweep, calm regime, max 8 pairs (the shipped config)21.5% · Sharpe 1.2704/29
BacktestHyperliquid replay, only 2024-05 → 2026-04 of data44.2% · Sharpe 0.7704/29
BacktestCross-venue spread arbitrage, 658 trades, 9.7% winCAGR −54.7%04/29
BacktestForward expectation for funding-arb after degradation8–10% a year, below the 12% bar04/28
Why it was stopped

Life one had no edge and lost on every strategy but one coin. Life two was the fleet's best backtest, but its forward expectation sat under the 12% bar, the primary venue is illegal for a US person, the alternative venue had two years of history, and in four weeks of paper the market never offered a single automatic entry. The 05/26 hold caught it still waiting.

"Root Cause: No Edge. The bot has no statistical edge. Technical indicators (RSI, MACD, Bollinger Bands) on 1-hour candles are lagging — by the time they trigger, the move is priced in. The 7-layer pipeline adds complexity but doesn't add alpha. The ML model was trained on 33 trades — statistically meaningless."BlackIce improvement plan, 03/30/2026

"Don't re-enable today. Reasons: 1. Funding-arb is the bot needed… 2. BlackIce main has 6 strategies + 9 filters + ML scoring — surface area for bugs is huge 3. We haven't run it in 10 days; market regime shifted… 4. Reconcile pass not yet written 5. Mid-day deploy violates non-negotiable #8."Re-enable plan, 04/29/2026

"Forward, expect 8–10% live (post-degradation), which is below the 12% bar. So FA-only does NOT meet the goal forward-looking… funding-arb makes sense as a diversifier sleeve (10–30% of portfolio), not a replacement for QQQ."QQQ vs funding-arb study, 04/28/2026

"If both <10% → strategy itself doesn't clear bar, abandon, all capital to QQQ/Multi."Four-week decision tree for Binance vs Hyperliquid, 04/29/2026. The four weeks ended in the 05/26 hold with zero entries on either venue.

To resume the signals bot: a reconcile pass, atomic close paths and seven smoke tests were specified and never built. To take funding-arb live: a US tax attorney first, 100+ live trades before any phase two, Hyperliquid only. The Telegram bot token still needs rotating.

VolnaArb and VolnaCarry

v0.1.0 scaffolds · 04/09/2026 · Node.js
Never launched

Two crypto dry-run skeletons from one afternoon. VolnaArb watches Uniswap v3 on Arbitrum against Bybit spot for ETH and ARB gaps above 15 bps with capital pre-staked on both venues ($6,000 notional). VolnaCarry pairs perpetuals across Bybit and OKX to harvest the funding spread on BTC and ETH ($10,000 notional, 8% APR to open, 5% to close). LIVE=true is hard-blocked in both; no wallet or exchange clients were ever wired.

Why they were stopped

They never started. Both were registered as disabled the day they were written, and the funding-rate idea reappeared three weeks later as the BlackIce funding-arb engines with a real backtest behind it. Zero opportunities logged, zero trades.

Ideas that were tested and not built

Strategies from the 04/28 data sprint and the advisor rounds that never became a bot.

Y v3 momentum + Path X passiveKilled 04/28

Monthly 12-1 momentum on US large caps. Backtest 1996–2026: Sharpe 1.05 net, CAGR 18.7%, max DD −44.1%, but survivorship-corrected to Sharpe 0.4–0.6 and 10–13% CAGR, about 1–2% a year over SPY since 2010. All three advisors: doesn't beat the MTUM ETF after slippage, taxes and dev cost. Verdict "Y v3 NOT BUILT"; $24,865 of MTUM and JEPI bought instead on 04/28 and sold in the 04/30 cleanup for +$90.37.

Gap ReversalKilled 04/28

Proposed Sniper repurpose. 117 trades, 31.6% win, −$6,681, Sharpe −3.57 annualized, max DD −$6,863. "DO NOT ITERATE: further parameter tuning would be in-sample fitting, exactly the trap that killed ORB."

PEIDS and PEADHold

Post-earnings IV decay: v1 naked puts 102 trades, 82.4% win, −$3,402; every variant negative. Post-earnings drift: +$92,497 in-sample, −$6,474 over 208 trades out of sample. "Edge real, instrument unworkable."

A1 long-vol calendarsKilled 04/28

Claude's pick in advisor Round 5. 60 trades, Sharpe 0.75, CAGR 0.98%. Killed under the 8% line.

Related projects

Pieces of the same effort that never traded through a broker, plus the one thing in this family that is still live.

Volna, the original dashboardRetired

Single-file SPA v1.3.8: $100k paper engine with SMA, RSI, MACD, Bollinger and VWAP strategies on simulated prices, 25 market-intelligence benchmarks. Last commit 03/23/2026; its EC2 host went with the AWS account and volna.meridiam.co now hosts this page instead.

volna-traderSkeleton

TypeScript paper-trading scaffold from 07/03/2026: Alpaca adapter, risk kernel, SMA-crossover strategy, backtester and journal with unit tests. Two commits, never deployed, no fills.

volna-alert-watcherStopped 05/26

The only routine Telegram sender for the fleet. Read bot health and P&L from the Meridiam API and decided what was worth interrupting Igor for. Its last hourly snapshot, 11:00 AM ET on 05/26: Sniper $101,444.82, Option $100,543.72.

ImpulsLive · info only

Momentum screener and daily digest at impuls.meridiam.co. Deliberately places no orders; since 08/17 it hosts a per-user paper portfolio for humans. Not a bot, so it has no line in the ledger.

How the numbers were built

  • Source of truth is the broker. 1,217 orders and 2,027 account activities (1,751 fills, 270 fees, 3 option expirations, 1 assignment) were pulled from the Alpaca paper API on 09/25/2026. BlackIce comes from its SQLite trade table.
  • Attribution. Orders carry a bot prefix in their client id (vs- Sniper, vo-/vc- Option, volnaneutral-, volnatoday-, passive-). Fills were replayed FIFO per symbol; a closing fill credits the bot that opened the lot, whoever sent the close. Sum of parts $3,967.44 vs broker $3,969.33.
  • Shared account. All four Volna bots traded one $100,000 account, so percentages are on that pool, not on a per-bot allocation. Option positions ate Sniper's day-trading buying power in April, which is why the 04/30 cleanup sold the passive sleeve and closed every option leg.
  • Unrealized. NVDA 100 sh at $223.865 and QQQ 10 sh at $741.56, intraday 09/25. NVDA arrived through assignment of a short 195 put on 06/26, a month after the Option bot was switched off.
  • Simulator artifacts are excluded. Before 04/21 the Option bot only pretended to trade, and its pretend ledger produced the −$1.95M cascade on 03/25–03/27 and a −$2,395,645 day figure on 03/30 with capital pinned at $100,000. The Alpaca account did not move. The bots' own numbers appear in the tables above only so they can be compared with the broker.
  • Self-reported vs broker. Sniper's own log says +$206.25 over 289 round trips; the broker says +$292.95 realized. The Meridiam trade table shows −$619.50 for Sniper over 93 trades, but it only recorded trades from 04/29 after a date-format bug was fixed. Option's position ledger claims +$19,632.70; its own equity-sync log contradicts it by up to $39,806 on a single tick.
  • Sources. Bot state files and trade logs on both contabo boxes, the Meridiam MySQL bot tables, the BlackIce SQLite database, the Volna master session summary and its ten companion studies, and the memory notes from each working session.
Times in ET · dates MM/DD/YYYY · money to the centPaper trading throughout; no real capital was ever at risk