Methodology & About

The product

Free

• Sentindex — flagship tracked NAV. Strategic Core + Strata blend, regime-dynamic weighting, daily composition + equity curve.
• Strata — thematic portfolios, each a Living Engine of sub-theses with conviction. Headline thesis, sub-thesis breakdown, current holdings, conviction log.
• The Regime Letter — bi-weekly Substack. What changed in the engine, what just flipped, what's in the book. Free, no daily-email spam.

Paid (in development)

• Full holdings with attribution — every position, every weight, daily per-position contribution to NAV, refreshed nightly.
• Sentindex CSV download — daily composition snapshot.
• Conviction-change alerts — push notifications when a sub-thesis conviction is updated or a portfolio is killed.
• Paid Substack articles — same author, same framework, deeper cuts.

The thesis

The 1991–2020 generic-access global regime — single-pole credit, free trade as default, marginal gold, compute as a global commodity — has been replaced. We're now in a multi-polar, specific-access regime along structural axes that interact in non-stationary ways. No single composite captures it. A panel of substrates, read literally, does.

The methodology is what we sell. The current substrate set is the worked example. When the regime rotates, the substrates rotate with it — pre-registered retirement criteria, all logged publicly.

The substrate model in one sentence

A substrate is a single observable macroeconomic variable with a pre-registered classification rule. Each substrate, every refresh, produces one of a small set of states (e.g. EXPANDING, RISING, TIGHTENING, CRISIS). Those states roll up into country-level regime reads. Those regime reads drive a global trade book. The trade book drives the Sentinel Index NAV.

substrate observation
    ↓  percentile + slope + level computation
substrate state (per pre-registered rule)
    ↓  consensus across substrates in the same country × dimension
country regime read (cycle, inflation, monetary, political)
    ↓  composite engine (BCI, CCMI, Recession Ensemble, Energy, Dollar, Global BCI, GPR)
two-layer portfolio
    ├─ Strategic Core — slot-based ETF sizing from composites
    └─ Strata — thematic portfolios sized basket × conviction
    ↓  regime-dynamic Core/Strata weight blend
Sentindex — flagship tracked NAV

The whole pipeline is deterministic. Same data in, same output. No weights are estimated from in-sample performance. No look-ahead. Every substrate, every rule, every classifier threshold is documented in the Substrate Browser.

Part 1 — Substrates

Definition

A substrate has four required components:

1. A data source — FRED series, IMF cache, yfinance ticker, Caldara-Iacoviello GPR, defined provider, traceable URL.
2. A pre-registered rule — how raw values map to a small discrete state vocabulary.
3. An asset implication — which retail-tradeable instrument the substrate informs, in what direction, on what horizon.
4. A frequency and release calendar — so we know when its state can change.

A substrate is not a model. It's not a composite. It's not a latent factor. It's one published macro variable, read literally, with a published rule on top.

State vocabularies

Each substrate emits a state from a small, dimension-specific vocabulary:

• Cycle: ACCELERATING / EXPANDING / DECELERATING / CONTRACTING
• Inflation: DISINFLATING / STABLE / RISING / SURGE
• Monetary: EASING / NEUTRAL_DOVISH / NEUTRAL_HAWKISH / TIGHTENING
• Political: STABLE / WATCHING / ELEVATED / CRISIS
• Credit: COMPLACENT / NORMAL / WIDENING / PANIC
• FX: STRENGTHENING / STABLE / WEAKENING
• Fiscal: SURPLUS / BALANCED / DEFICIT / DOMINANCE / CONCERN
• External: SURPLUS / BALANCED / DEFICIT
• Commodities / flows / supply-chain: percentile-banded with category-specific labels

Vocabularies are intentionally small. The point is to read the regime, not curve-fit it.

Computation per substrate

For each observation, three numbers feed the classifier:

1. Current value — the latest observation.
2. 3-month slope — direction of change.
3. Long-window percentile — where the current level sits relative to its own history. 5-year percentile default; some long-history substrates use 30-year.

Each substrate's pre-registered rule consumes some subset of those three numbers and emits a state. The rule lives in the substrate's JSON file — no buried hyperparameters, no estimated weights.

Part 2 — Country regime reads

Each of 16 countries gets four dimensional reads: Cycle / Inflation / Monetary / Political. The country regime is computed two ways depending on substrate density:

• Plurality-with-confidence when the country has 3+ substrates on the dimension — the dimensional state is the modal substrate state; confidence is the share holding that state.
• Single-substrate carry when only one substrate exists — that substrate's state IS the dimensional state, with confidence labeled as derived-from-one.

The regime map renders the 16 × 4 grid live. Cells with no registered substrate render as a dash; we don't bluff coverage.

Part 3 — The active trade book

The trade book is generated by four rule families, each reading from substrate state directly:

1. Substrate firing — a single substrate flips into a stressed state (RISING / SURGE / CRISIS / etc.) and its asset implication fires directly. Most of the book.
2. Country composite — two or more dimensions of the same country fire stressed states together. Triggers a country-level long or short.
3. Cross-country pair — two countries diverge meaningfully (e.g. India ACCELERATING vs Mexico CONTRACTING with paired inflation states). Triggers a long-short pair.
4. Risk-off composite — N countries fire simultaneously on inflation SURGE or political CRISIS. Triggers a market-wide short or vol position.

Each thesis is tagged with: trigger substrate(s), conviction (HIGH / MEDIUM / LOW), direction (long / short / pair / watch), horizon (days), and the mechanism — why the thesis should work. The full active book is paid.

Part 4 — The two-layer portfolio

The Sentindex is constructed from two independent layers that combine under a regime-dynamic weight rule.

Layer A — Strategic Core

A broad-ETF book sized by the composite engine outputs. Each composite drives one or two slots:

Slot ceilings sum > 100% then scale proportionally to a 100% gross cap. Monthly rebalance. Walk-forward backtested 2014-01 → 2026-04 (147 months): annual +11.24%, Sharpe 1.35, max drawdown −11.0%, alpha +229 bp/yr vs 60/40.

Layer B — Strata (Living Engine thematic portfolios)

Each Strata portfolio is its own micro-product:

• A headline thesis (the "why")
• 3-5 sub-theses, each with a basket of names and a conviction (0.0–1.0)
• Sub-thesis weight = conviction / Σ conviction within the portfolio
• Equal-weight within each basket
• Conviction updates re-shift weights without rebalancing the basket structure

A portfolio's included_in_sentindex flag determines whether it contributes. Conviction → 0 kills a sub-thesis without deleting the spec; conviction → 0 across all sub-theses kills the portfolio. Kill criteria are pre-registered in each portfolio's spec.

Combination — regime-dynamic Core/Strata split

The Core/Strata weight split shifts by BCI phase:

Within the Strata bucket, portfolios are weighted by their average sub-thesis conviction. The Sentindex daily return = core_weight × core_return + strata_weight × Σ(portfolio_share × portfolio_return).

Part 5 — The Sentindex NAV

The Sentindex is a tracked NAV computed daily from the two-layer portfolio.

• Inception: at $100.00.
• Daily return: weighted blend of Strategic Core + each included Strata portfolio per the regime-dynamic split.
• Cost model: 1 bp daily drag per portfolio on gross long exposure.
• Rebalance cadence: Strategic Core monthly; Strata on conviction-change events (no scheduled rebalance — the basket structure is stable, only the weights respond to conviction updates).
• Current state: . Strata portfolios contributing. Gross exposure , net .

The Sentindex NAV is the product. The substrate engine + the two-layer architecture is the methodology.

Part 6 — What this is not

• Not a forecast factory. We don't predict CPI prints. We classify regimes from observable substrates and act on them.
• Not curve-fit. No optimization over historical returns. Sizing is a rule, not a backtest result.
• Not a black box. Every input, every rule, every threshold is in the substrate browser or this page.
• Not advice. Informational and educational only. The terms are here.

Cadence

How it's different from most macro writing

• Substrate-driven, not pillar-driven. The previous version of this site was organized around seven thematic "Repricing pillars." That model has been retired. Substrates are smaller, more numerous, more falsifiable, and each one has a single published rule. The framework is the methodology, not a narrative.
• Pre-registered everything. Every substrate's classification rule is committed before being graded. Every sleeve sizing rule is locked. Rejected research isn't quietly deleted; it stays in the kill log.
• No look-ahead, no curve-fit. No weights estimated from in-sample performance. ALFRED first-release vintages where applicable. Walk-forward state computation only.
• Public when it kills itself. If a substrate or a trade family fails, the kill log is public. The product is the discipline, not the back-test.
• Free where data is free. Public data, public rules, public regime read. Paid is for depth and actionable position-level detail — not for things any reader can compute from a public source.

Contact

Reply to any Regime Letter you receive. That's the channel.