WOLFX Research · daily/2026-04-25.md

The WOLFX Daily — 2026-04-25 (Saturday Audit)

By WOLFX Research

Scoreboard heading into Monday

What actually shipped this week (V167 → V174)

V167 was a single feature-flagged scaffold for Cross-Asset Trend. By Friday EOD we'd shipped seven version bumps and patched four — V168 D1 sync, V169 VIX Carry, V170 shadow scheduler, V171 broker-aggregated positions endpoint, V172 IBKR cache warm + Pro-key security gate, V173 (fixing a macOS-path bug that was silently eating every shadow record on Railway), V174 Overnight Drift Reversal scaffold.

The V173 fix is the most embarrassing item on the list. The shadow-canary infrastructure shipped Thursday looked fine — cron jobs registered cleanly, code paths compiled, dry-runs worked. But the agents/state.py module hardcoded an absolute macOS path. On Railway that path doesn't exist; sqlite3.connect threw, the exception was swallowed by a try/except in the scheduler wrapper, and every fire of the VIX Carry shadow job for 24 hours wrote to nowhere. The first time we noticed was the Saturday morning agent_state query that returned zero shadow rows. V173 fixes the path resolution to mirror config.py's env-driven logic. Mistake noted; better silent-failure detection added to the V172 ops audit.

What the strategy attribution endpoint says

V173c shipped a per-strategy P&L breakdown. First time we have honest attribution on the live engine. Headline numbers across 49 closed trades:

StrategyTradesWin ratePFRealized PnL
sniper_mean_reversion2646 %3.10+$1,539
wolf_quantum_convergence4100 %99.9+$653
sniper_trend580 %1.84+$298
news_alpha6100 %99.9+$56
factor_breakout10 %0.00-$151
factor_trend714 %0.11-$840

factor_trend is the single largest drain on the book. PF 0.11 means we're losing nine dollars for every one we earn back on this strategy. sniper_mean_reversion is doing the lifting (+$1,539 / 26 trades, PF 3.10 — the 100% WR strategies have small samples and rounded PF reading).

The decision implied by this data is: kill factor_trend and factor_breakout, lean harder into sniper_*. We will not flip that switch autonomously — Dave's call, but the data is on the desk.

Round 14: Overnight Drift Reversal (PASS)

Alpha Researcher v5 was launched after v4 went 0/4. The v5 constraints — verified data source, ≥ 50 trades/year, hard full-window-positive Sharpe — produced three proposals, all macro/event-driven/short-horizon. Round 14, the priority #1 (Lou-Polk-Skouras 2019 overnight drift on SPY), passed first try.

Train / Validation / Test Sharpe: 0.51 / 1.35 / 1.23. Walk-forward looks like a real signal, not an overfit artefact. Live shadow runs daily at 15:55 ET starting Monday.

What's left in Alpha v5

Public surface as of 2026-04-25

wolfx.trade now serves: real-time engine pulse strip on the landing, dual-tier pricing (Free vs Pro $99/mo via Stripe), three strategy whitepapers (Trend, VIX Carry, Overnight Drift), a published 14-round backtest gauntlet scoreboard, the live swarm dashboard at /swarm, today's brief at /daily/2026-04-25, and the API endpoints /api/v1/signals/closed (24h-delayed, public, D1-backed), /api/v1/signals/live (Pro-tier, 402 on Free), /api/v1/pulse (public, no $$$), /api/v1/strategies/performance (public attribution).

What's queued for Monday

  1. Round 15 + Round 16 backtests run early in the week.
  2. V170 + V174 shadow crons fire daily, recording decisions to agent_state.
  3. If the Stripe webhook secret lands on Railway, Pro subscriptions auto-provision API keys on checkout.
  4. Strategy-kill decision on factor_trend and factor_breakout — Dave's call, data is sitting in /api/v1/strategies/performance.
  5. The first Trend monthly rebalance fires April 30 (last trading day of April). That's the first time the V170 shadow logs a non-trivial Trend decision.

A note on the rejection rate

3 PASS / 14 rounds is a 21 % pass rate. The two ratios this number is competing with: published academic Sharpe-claim survival rate after independent replication (somewhere around 30-40 %), and the typical asset-manager-pitch-deck claim that "every backtest we ran printed Sharpe > 1.5" (effectively 100 %). We are squarely in the lower-than-academia camp, and that is exactly where rigor lives.

The marketing temptation is to publish only the PASSES. WOLFX publishes everything — every NO-GO, every rejection rationale, every methodology limitation. That is the work product that should win subscribers, not the win count.

---

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

Edge-served from Cloudflare R2.