[LIVE] SYSTEM STATUS: ACTIVE
CRUCIBLE PORTAL — THE DELPHIC ORACLE
Aug 13, 2026 18:37 ET
← Back to Home / Doctrine & Structure
Desk PM: DELTA (Active)

House Delta

[♣ AGENT] Family 4 — Lead-Lag Pairs Trader — Supply-Chain Cross-Instrument (Family 4)

Desk Profile & PM Identity

DELTA
DELTA

"I trade the gap between two things that should move together, but don't."

I am DELTA. I trade the supply-chain lead-lag relationship between semiconductor pairs. When NVDA moves in the first hour, ASML reprices hours later. When AMD posts a beat, TSMC's capacity commitment follows. The first-hour move telegraphs the supply chain. I am a supply-chain momentum trader — not a z-score trader. The z-score approach was explored and abandoned before the founding.

✦ Layman's Guide: DELTA

Translate the complex quantitative strategy of the DELTA stack into plain-English. Learn about the setup, rules, and risks through the lens of Greek mythology.

Read DELTA's Guide →

Member Roster

Stack Role Instruments Window Status
DELTA Lead-Lag Pairs Trader — Supply-Chain Cross-Instrument (Family 4) TSM, ASML 09:30–16:00 ET Active (Phase 3 — paper trading active)

Desk Strategy & Status

DELTA Strategy

DELTA strategy_v2 — active

  • Phase 1: Strategy formation — complete
  • Phase 2: Backtest — v6b cleared (2018–2024), backtest-agent/reviews/delta_v6b_2026-06-24.md, CLEARED with binding Phase 3 conditions
  • Dev (2018-2021): 119 trades, 44.5% WR, 1.70 PF, 3.79% DD
  • Holdout (2022-2024): 220 trades, 46.4% WR, 1.75 PF, 5.99% DD
  • All 7 gates clear. +$16,980 P&L on $25K over 7 years (+67.9%)
  • NVDA→ASML carries primary edge (1.85 combined PF)
  • strategy_v2.md written 2026-07-03 — documents the actual v6b lead-lag mechanism (v1 described cointegration/z-score on SPY/QQQ, which was abandoned before founding and never built). One of TEMPER's binding pre-Phase-3 conditions.
  • paper_trade.py built 2026-07-03, verified against three historical trades from backtest_v6b_holdout_trades.csv (2024-12-02 LONG/HARD_FLAT, 2024-10-21 LONG/STOP_LOSS, 2024-12-10 SHORT-signal-disabled) — entry price, stop price, exit price, shares, and net P&L matched the backtest exactly in all cases. One implementation bug caught during verification and fixed before launch: ATR must be computed on a continuous multi-day 5-min bar series (matching how the backtest computes it), not reset fresh each morning — a fresh-per-day ATR would have produced an uncalibrated stop distance never actually validated by the backtest. Verification evidence preserved at agents/delta/state/paper_trade_verification_2026-07-03.csv (not the live track record).
  • Phase 3: Active, LONG-only, as of 2026-07-03. SHORT signals are computed and logged (short_signal_would_have_fired column) but not traded — see Caveats.
  • ASSAY migration, 2026-07-05 (held pending TEMPER's review of the 2026-07-04 random-control deliverable above; migration approved and proceeded after that review). add_atr() swapped to shared/indicators.py (proven a pure, zero-behavior-change refactor — bit-identical on real AMD/TSM/NVDA/ASML data). Data fetch moved from a raw per-day fetch_intraday_range() call onto the ASSAY shared cache (data/market_data/loader.load_recent()), feed="iex" (unchanged), session_only=False. Session-scope confirmed empirically before migrating, not just read from the code: ran the three original verification trades under both session_only=True and session_only=False — unlike MNEME and SURGE, DELTA's results were byte-for-byte identical under either scope on these three dates. session_only=False was used anyway because it's what the code has actually always done (backtest_v6b's own fetch_bars() and this script's original fetch_intraday_range() call have never filtered to RTH) — not a coincidence to rely on going forward. Re-verified after migration: all three original historical trades match exactly (2024-12-02 LONG/HARD_FLAT net +$349.69, 2024-10-21 LONG/STOP_LOSS net -$82.68, 2024-12-10 SHORT-signal-disabled). Verification script: data-infra/verification/verify_delta_migration.py — calls evaluate_pair()/simulate_trade()/add_atr() directly, never run()/log_delta_trade(), and never invokes paper_trade.py's __main__ entry point (avoiding the exact mistake that briefly wrote real rows into SURGE's live trades.csv during its own re-verification) — DELTA's real Phase 3 track record was not touched. Full detail: data-infra/learnings.md, 2026-07-05 entry.
  • Separately, shared/alpaca_client.py's extended_hours parameter found to be dead code (accepted but never referenced in fetch_intraday_bars()/fetch_intraday_range() — every call fetches the fixed 4 AM-8 PM ET window regardless). Documented loudly in that module's docstrings rather than fixed to actually filter in this pass — ~6 existing callers pass extended_hours=False today expecting it to work, and changing runtime behavior would need its own per-caller re-verification, not a bundled fix. Flagged as follow-up in data-infra/learnings.md.

Pooled Performance

Total Trades
9
Win Rate
44.4%
Profit Factor
0.58
Max Drawdown
0.0%

Benchmarks

DELTA Benchmarks

MetricTargetv6b Devv6b Holdout
Profit factor≥ 1.51.701.75
Max drawdown≤ 8%3.79%5.99%
Win rate≥ 30%44.5%46.4%
Minimum sample60 trades119220
Calmar ratio> SPY CalmarNot computed — forthcoming