The WOLFX Daily — 2026-04-28 (Tuesday)

By WOLFX Research

Scoreboard

• NAV by broker (post-IBKR-recovery): IBKR $1,009K (+$42K unrealized) + OANDA + Alpaca
• Active positions: 7 across all three brokers — long SLV / short SPY (Alpaca), long USD_CHF + EUR_CHF / short USD_CAD (OANDA), long ES 7 cts + NQ 4 cts (IBKR)
• First shadow trade just closed: Overnight Drift Reversal SPY entry $715.17 → exit $711.82 = -0.097 % net. One trade, statistical noise — first of ~30 needed for promotion math.
• Backtest gauntlet: 3 PASS / 15 rounds (unchanged — research is paused while we accumulate shadow data on the three already-validated strategies)

What happened: the 50-hour IBKR outage, postmortem

WOLFX lost dashboard visibility into the IBKR portfolio from Saturday morning through Tuesday morning — 50 hours of "$0 IBKR" on every monitor, while the actual broker account continued running ES + NQ futures untouched.

The root cause was an IBKR account configuration that became visible only when IB Gateway had to re-authenticate. Last spring we had set up two IBKR users: winklhawk for live trading on Railway, winklhawk2 as a separate sub-user for the desktop TWS client (so the two clients don't kick each other's session). What that setup did invisibly is provision a second paper trading account in the master tree. IBKR's auth then refused to auto-pick one — every login attempt got rejected with "the specified user has multiple Paper Trading users associated with it."

• The fix had two parts:
• Delete the second paper user (winklhawk2) at IBKR Account Management — that removed the orphan paper account
• Switch Railway's TWS_USERID from the live username winklhawk to the paper account's actual auto-generated username (in our case cxgcsa981, a format every IBKR paper account gets)

That second piece was the real lesson. The gnzsnz/ib-gateway:stable image — the standard IB Gateway Docker community uses — does NOT honor the TWS_USERID_PAPER env var. It reads only TWS_USERID, regardless of TRADING_MODE=paper. So when we set TWS_USERID=winklhawk and TRADING_MODE=paper, IBKR was being asked to authenticate the live identity at the paper login endpoint. With one paper account that was tolerated; with two, IBKR refused to disambiguate.

The actual paper username (cxgcsa981) lives behind IBKR Account Management → Settings → Paper Trading Account. Memory file reference_ibkr.md now records this so future Railway restarts pick up clean credentials without intervention.

Autonomous pipeline: first end-to-end run

Yesterday at 15:55 ET the V174 cron fired. It ran the Overnight Drift signal logic, found the prior 5-day intraday log return was -0.30 % (oversold filter triggered), and recorded an overnight_drift_shadow row to agent_state with enter=True. At 16:35 ET the V170 vix_carry cron fired and recorded an OPEN decision (contango +93.63 %, VIX 18.0, SPY 6-month return +4.96 %). At 16:45 ET the V179 auto-promote cron fired, read both shadow records, and logged HOLD decisions for all three strategies (insufficient sample size, expected at day 1).

This morning at 09:30 ET the Overnight Drift "trade" closed at the market open. The V178 shadow MTM endpoint marked it: SPY 715.17 → 711.82, net -0.097 % after the 1.5 bps cost model. Statistical noise — the backtest had 73 % win rate but daily-level outcomes vary widely. We need 20+ trades (40+ sessions accounting for the ~43 % fire rate) before the rolling-Sharpe math gives anything meaningful.

The pipeline is in place. Now it just runs.

What today and the rest of the week looks like

• 15:55 ET: V174 cron fires Overnight Drift signal #2. Filter input is whatever the prior 5 intraday returns sum to as of today's close.
• 16:35 ET: V170 vix_carry decides whether to keep yesterday's OPEN position alive or close it. Strategy is monthly rebalance, so absent a regime gate fail (VIX > 30 or SPY 6-month negative) it stays open.
• 16:45 ET: V179 auto-promote logs eligibility — still HOLD given trade count.
• Wednesday, April 30: First Trend rebalance day. The V170 trend cron will run signed 12-1 skip-month momentum on ES, NQ, RTY, YM, 6E, 6J and record the target positions per leg.

Sniper attribution drift

The V173c attribution endpoint now shows 58 trades across 6 strategies. sniper_mean_reversion was the workhorse last week (PF 3.10, +$1,539 across 26 trades). It continued trading through the IBKR-blind period and the latest snapshot has it at 35 trades, +$782, PF 1.52 — strategy degraded over the weekend. Still a worker, but the edge thinned. factor_trend and factor_breakout remain disabled at the code level (PF 0.11 and 0.00, locked-in losers). The strategy-kill admin endpoint shipped as V177 lets us flip any of the three validated shadow strategies on/off remotely once the data justifies it.

Public surface unchanged from Saturday

wolfx.trade continues to serve: landing with live pulse strip + Free/Pro pricing, dedicated /pricing page, /whitepapers index, /changelog, /swarm agent dashboard, three strategy whitepapers, the 15-round backtest gauntlet scoreboard at /reports/gauntlet-scoreboard.md, this and prior daily briefs. API endpoints /api/v1/signals/closed (24h-delayed, public), /api/v1/signals/live (Pro tier only), /api/v1/pulse, /api/v1/strategies/performance, /api/v1/strategies/shadow/performance, /api/v1/keys/generate. Stripe polling (V176) runs every 5 minutes — when the first Pro subscriber checks out via the payment link, the API key is auto-provisioned within 5 minutes. No Stripe webhook needed.

---

WOLFX publishes every signal and every realized fill. Past performance, including walk-forward backtest performance, is not predictive of future results. Trading involves substantial risk of loss.