How the Engine Works

1. The state machine

The monitor's job is to answer one question: what is the probability of a ≥10% S&P 500 drawdown within six months, and what should the book hold given that probability? It answers with a small state machine, not a forecast.

Layer 1 (L1) — the early-warning quorum. Four legs watch for deterioration: (1) high-yield credit spreads widening fast (OAS +50bp in 20 sessions), (2) high-beta underperforming low-volatility stocks (below its 50-day average), (3) the 2s10s yield curve at or near its 12-month low, and (4) volatility spiking (VIX > 1.5× its 20-day average). When at least 2 of 4 legs are hot, the warning quorum is met.

The confirmation latch. A quorum alone is not enough — the state entry used here required a confirming credit event (a CCC-tier spread blowout of +200bp in 20 sessions), which latched the entry. This is deliberate: credit confirmation filters out the many L1 quorums that resolve harmlessly.

The four states. Stage 1 is the base portfolio. PHASE_2A (current state, entered 2026-10-14) rotates the book defensive: equity cut to a probability-matched cap, duration added in stages, cyclicals exited. PHASE_2B arms only when credit stress is severe (HY OAS ≥ 450bp) and the Fed has rolled over into easing — it is the maximum-duration state, and it cannot arm while the Fed is hiking. TAIL_OVERRIDE is the stagflation-tail backstop (10Y weekly close > 5.40% AND Brent > $120): it holds the Stage 1 book, suppresses all duration adds, and forces full gold.

The asymmetric exit. Getting in is fast; getting out is slow, on purpose. Exit from 2A requires both credit healing (HY OAS below its 50-day average for 3 consecutive sessions) and market internal repair (high-beta/low-volatility back above its 50-day for 3 sessions) — or a 60-session time stop (~mid-January 2027 for the current episode). A 10-session cooldown follows any exit before a new entry can trigger. This is why L1 decaying from 2/4 back to 1/4 does not exit the state: the legs are entry sensors, not exit sensors. The backtest evidence (see /backtest-v7/) shows exits on healed internals, not on quieted alarms.

2. Duration management — when the book adds and exits duration

Duration is the main 2A/2B position, and it is managed in stages, gated by a dedicated risk flag rather than by price targets alone.

Staged entry. In 2A, the long-duration sleeve (IDTL) deploys 15% immediately and a further 10% ("tranche 2") only while the Duration Risk Flag (DRF) is cold. In 2B the full 32% deploys, subject to the same flag.

The DRF. The flag watches for the one configuration that reliably destroys duration: an oil supply shock (Brent +12% in 20 sessions) combined with at least one transmission leg — yields-and-stocks falling together, breakeven inflation rising, or a debasement move (gold up with the dollar down). It sets after 3 confirming sessions and clears only after 15 quiet ones. While hot, prescribed duration is cut to one-third and tranche 2 is blocked; the freed sleeve goes to cash-equivalent carry.

Why pure debasement doesn't trigger it. The backtest showed the oil leg carries the entire signal: gold-up/dollar-down without a supply shock historically left duration unharmed. A flag that fired on every gold rally would have cut the book's best 2A asset repeatedly for nothing.

When duration exits. Three routes: the DRF gate (graduated, earliest), the state-exit rules above (the book returns to Stage 1's small 6% duration), or the mechanical TAIL backstop (10Y > 5.40% weekly close). There is also a take-profit ladder on the way down: 10Y at 4.25% trims IDTL to 15%; at 3.90% to 10%, rotating into inflation-linked and gold.

3. Gold — base weight, two channels back to full

Gold is held at a 10% base in all states (amended 2026-11-01, down from 15–18%). The amendment followed a straightforward finding: gold was in a confirmed bear market (−29% from its January 2026 peak, below both major averages, at 52-week lows) and the debasement configuration the position hedges — weak dollar, dovish Fed — was dormant. Raising gold into that tape (as the October amendment had) was the mistake this rule corrects.

Re-entry to full weight (15% Stage 1 / 18% 2A/2B), whichever fires first:

• Channel A — regime (primary). The debasement leg active for 3 consecutive sessions (gold ≥ +4%/20d with the dollar flat-to-down), or the DRF firing hot, or TAIL_OVERRIDE — the latter two force full gold immediately, because in those states gold is precisely the hedge the book needs.
• Channel B — trend (backstop). Gold's weekly close back above its 200-day average. This catches policy-pivot or momentum bulls that have no macro signature yet — the kind of run that produced the January 2026 peak.

Deactivation after a Channel A entry requires the debasement leg quiet for 15 sessions and price still below the 200-day — the same set/clear asymmetry as the DRF, designed to filter bear-market rallies where the dollar never cooperates. Restoration is always funded from carry, which remains the residual sink under every dial combination.

4. Equity sleeves — what the book owns and why

Equity is capped by rule, not by feel: equity = (1 − p) × 30%, where p is the framework-implied probability of a ≥10% drawdown within six months. At p ≈ 28% (the current up-tape reading), the cap is 21%.

Within the cap, the sleeves are chosen for the regime: COWZ (free-cash-flow value — quality at a discount, with an energy tilt that fits a supply-shock tape), XLV (healthcare — the classic defensive leader, overweighted in 2A and currently validating that choice at 52-week highs), and GRID (grid/electrification infrastructure — contract-backed capex tied to physical chokepoints, the "energizer" layer). AMLP and PDBC are zeroed in 2A/2B by design: midstream and broad commodities are Stage 1 yield/carry diversifiers, but in a risk-off state their equity beta and oil sensitivity are exposures the state exists to avoid — the oil hedge is already expressed through the DRF's response, not through holding cyclicals.

5. Position health — the "gold rule"

Adopted 2026-11-04, after the gold episode showed that nothing was watching position health between amendments. Two layers:

Layer 1 — amendment-time gate. Every weight change, however obvious, must include a one-paragraph pre-mortem per affected sleeve: price vs the 200-day, 20-day momentum, and whether the reason we hold it is active or dormant.

Layer 2 — the health card. Each sleeve is classified weekly: RED = weekly close more than 10% below its 200-day average and its regime driver dormant → mandatory review with pre-registered actions. One key failed = AMBER (watch); both intact = GREEN. The rule is role-aware: carry is exempt; IDTL's health is the DRF gate, not its price trend (duration below its 200-day is by design while the Fed hikes); gold's row is the two-channel gate above. Hysteresis mirrors the DRF — 3 sessions to set a RED, 15 quiet sessions to clear — and the human review is weekly, on Fridays, never daily: the research record here is unambiguous that higher-frequency discretion produces false starts, not better exits. Event overrides don't wait for Friday: DRF-hot, TAIL, the 10Y kill-switch, or the Anthropic IPO pricing.

6. Provenance and disclaimers

The v7 framework is frozen: trigger thresholds, state logic, exit conditions, and latches change only by explicit owner-approved amendment. Amendments to date: 2026-10-28 (probability-matched equity sizing, staged IDTL, the DRF overlay) and 2026-11-01 (gold base 10% with the two-channel gate). The validation record lives at /backtest-v7/; the drawdown re-entry research (L5b/L6c) at /backtest-retest/ is a separate, shadow-only workstream with no allocation role. Everything on this site is analytical research produced for one owner's book. It is not investment advice, and nothing here is an offer or solicitation.