Index Catalog & Strategies¶

SignalForge currently publishes six rules-based indices, a capital-protection wrapper (HAVEN), and a regime meta-allocator (ARIA). This page describes each at methodology level: what it holds, how it decides, how often it rebalances, and its risk posture.

No performance figures on this page

All SignalForge strategies are validated against historical backtests and simulation batteries before publication. Backtest results do not guarantee future performance, and specific validation figures are not published here. Validation evidence — rolling-tranche analyses, Monte Carlo batteries, stress tables, and tearsheets — is available from ACM on request.

Catalog summary¶

Index	Ticker	Mechanism	Universe	Rebalance cadence	Risk posture
Digital Gold Index	DGI	Macro-signal allocation, core/growth sleeves	USDi (CPI-indexed stable), gold (PAXG), BTC	Daily evaluation, threshold-triggered rebalancing	Long-only; volatility-targeted BTC sleeve; HAVEN-eligible
Just Real Return Token	JRRT	Real-return allocation from macro signals	PAXG, USDi	Daily	Long-only, two-asset, unlevered; not HAVEN-eligible
Basis Yield Index	BYI	Binary rotation: delta-neutral basis trade vs JRRT core	BTC/ETH/HYPE basis legs, JRRT core	Daily	Market-neutral income profile; no leverage, long-spot base; not HAVEN-eligible
apyUSD Levered Carry	APYC	Levered fixed-yield carry loop	Pendle PT-apyUSD vs Morpho apxUSD debt	Daily maintenance within an LTV band; periodic rolls	Levered carry, apxUSD-denominated; peg and capacity disclosures; HAVEN-eligible
Hard-Asset Trend Index	ATLAS	Ensemble momentum, long-only	BTC, ETH (wstETH), HYPE (staked), gold (PAXG), USDi cash sleeve	Daily	Fully unlevered, volatility-targeted, drawdown-throttled; HAVEN-eligible
Levered Hard-Asset Trend	ATLAS-L	Same signal engine as ATLAS, levered execution	ATLAS universe plus perpetual futures	Daily	Gross exposure capped by a published invariant; per-profile attested jump bounds; HAVEN-eligible

All indices publish from a base value of 100, model transaction costs explicitly, and make every decision on prior-day data — there is no intraday discretion anywhere in the family.

DGI — Digital Gold Index¶

A multi-asset macro allocation across USDi (a CPI-indexed stable asset), gold (PAXG), and BTC, organised as two sleeves:

a core sleeve (USDi + gold) that plays defence, and
a growth sleeve (BTC) that takes risk when macro conditions permit.

The split is driven by two macro signals derived from public data: the level of 10-year real yields and the inflation slope (long-dated minus short-dated breakeven inflation), each mapped through a normal CDF into a score. High real yields and a flattening inflation slope shift weight toward the growth sleeve; the same scores split the core sleeve between gold and USDi. Macro inputs come from published FRED series.

Two overlays then apply:

Momentum filter — the BTC allocation is scaled down while BTC trades below its 20-day moving average.
Volatility targeting — the BTC leg is sized against a per-profile volatility target using a covariance-based portfolio volatility estimate, and rebalances trigger only when weights drift past a per-profile threshold (thresholds widen automatically in high-volatility regimes).

Profiles range from conservative to aggressive by volatility target (2% to 20% annualized) with matching leverage caps and rebalance thresholds. DGI carries a signed jump-exposure bound and is HAVEN-eligible. Its halt policy allows a wider signal-staleness window than the rest of the family by design, because its macro inputs publish only through the prior calendar day.

JRRT — Just Real Return Token¶

A two-asset real-return index: gold (PAXG) versus USDi (CPI accrual), with dynamic weights derived from the same macro pair as DGI — 10-year real yields and the inflation slope, each mapped through a normal CDF. The gold weight is the average of the two scores; USDi takes the remainder; the index compounds daily on the weighted returns. Rebalancing is daily, and profiles add optional volatility targeting at increasing targets (5%, 10%, 15%).

JRRT deliberately publishes without a signed jump-exposure bound: both legs are considered gap-benign, but no bound has been signed off, so the risk-engine contract carries an explicit machine-readable refusal and JRRT cannot be wrapped by HAVEN.

BYI — Basis Yield Index¶

A binary daily rotation between two legs:

a delta-neutral basis trade on Hyperliquid — long spot, short perpetual, across BTC, ETH, and HYPE at fixed sleeve weights — and
JRRT core as the defensive leg.

A funding-based Basis Score (a normalized z-score of the weighted perpetual premium) decides the rotation: the index allocates to the basis trade when funding conditions score favourably, and to JRRT otherwise. A volatility guardrail is checked first: if trailing volatility exceeds its threshold, the index forces the defensive leg regardless of the score. Transaction costs are charged at regime switches only.

The basis leg is a composite asset: every published output decomposes it into executable long and short legs per asset, and the JRRT core leg resolves live as a child index. The structure runs at 1× leverage across all profiles on a long-spot base (gross exposure 2× NAV, net 0× when the basis trade is on). BYI publishes without a signed jump-exposure bound and is not HAVEN-eligible.

APYC — apyUSD Levered Carry¶

An apxUSD-denominated NAV of a daily-maintained levered carry loop: buy fixed-yield principal tokens (Pendle PT-apyUSD), borrow apxUSD against them on Morpho at a target loan-to-value, maintain the position within a relative LTV band, and roll to the next maturity ahead of expiry. The apxUSD/USD price is explicitly outside the index — subscribers own that exposure directly, and the methodology says so.

Profiles span a target-LTV ladder from unlevered through a levered base profile, plus a dynamic profile that throttles leverage when either of two leading stress signals fires (a sharp single-day jump in stablecoin borrow rates, or a multi-day slide in the apxUSD price) and restores it only after a confirmation period. The risk-engine contract additionally publishes carry-inversion and yield-spike de-risk rules, a hard de-risk LTV well inside the market's liquidation threshold, and a signed jump-exposure bound.

APYC's disclosures are notably blunt: peg exposure is owned by the subscriber, capacity is explicitly limited and disclosed per methodology version, and venue liquidity constraints are recorded. Its live history includes the June 2026 apxUSD depeg event, and the methodology's stress behaviour through severe peg scenarios is part of the validation evidence available from ACM.

ATLAS — Hard-Asset Trend Index¶

The USD NAV of a mechanical, long-only, fully unlevered ensemble-momentum portfolio over BTC, ETH (held as wstETH), HYPE (staked), and gold (PAXG), with USDi as the cash sleeve. All decisions use prior-day data. The daily rule stack:

Ensemble signal — the average sign of returns over short, medium, and long lookbacks, clipped to long-only.
Trend gate — an asset's signal is zeroed unless price is above its long-term moving average.
Crowding filter — a long is trimmed while perpetual funding for that asset is unusually rich (a sign the trade is crowded).
Sizing — inverse downside-deviation weighting with per-asset volatility targets and per-asset caps.
Portfolio volatility target — the whole book is scaled to a portfolio-level volatility target with gross exposure capped at 1.0.
Cash floor — each profile reserves a fixed cash allocation (the only parameter that differs across profiles).
Smoothing — weights are exponentially smoothed to limit turnover.
Drawdown throttle — risk weights are halved while the index is materially below its high-water mark, releasing at a tighter level.
Costs — modeled per unit of one-sided turnover.

ATLAS carries a signed jump-exposure bound and is HAVEN-eligible. Its published disclosures include concentration effects in the recent sample (a gold-heavy regime), sensitivity to lookback selection, and a transaction-cost ceiling.

ATLAS-L — Levered Hard-Asset Trend¶

ATLAS's levered sibling: an identical signal engine, but the portfolio volatility target may request gross exposure above 1×, up to a published gross-exposure cap of 3.5× — an invariant the consumer can enforce (the sum of absolute non-cash weights never exceeds the cap).

Execution structure is part of the methodology: exposure is carried spot up to 1× NAV, and perpetual futures above that, with gold levered via a gold-backed perpetual. Signals publish first-class spot-weight and perp-weight splits, and vault directives scale that split linearly — the executor never re-derives it. A fixed fraction of perpetual notional is reserved as idle margin, and the engine de-risks to spot-only if a margin buffer is breached.

Three profiles — conservative, core (flagship), aggressive — differ by portfolio volatility target (15%, 20%, 25%) with proportionally scaled drawdown throttles. Each profile carries its own signed jump-exposure bound, so protected products over different profiles are sized independently. Disclosures cover levered bear-market behaviour and the modeling assumption used for gold-perpetual funding.

HAVEN — the capital-protection wrapper¶

HAVEN is a structuring primitive that converts any qualified strategy into a capital-protected (par) or defined-outcome (sub-par floor) fixed-tenor product using a capital-protection mechanism: two assets and a rebalancing rule, no options or derivatives.

Mechanism¶

A HAVEN tranche holds exactly two arms:

Floor arm — a yield-bearing asset that manufactures the guarantee: a fixed-yield principal token redeeming at par, or a USDi / T-bill-style wrapper accruing at an underwritten rate.
Risky arm — shares of the wrapped strategy at NAV.

Every monitoring step (daily in backtest; ticks of up to 30 seconds in production):

Floor(t)   = Principal x F x (1 + r)^(-time to maturity)   # PV of the guarantee
Cushion(t) = NAV(t) - Floor(t)                             # the risk budget
m(t)       = multiplier policy
Target(t)  = clip( m(t) x Cushion(t), 0, NAV )             # risky-arm allocation

where F is the protected fraction of principal (F = 1.00 for par protection; F = 0.90 caps the maximum loss at maturity at 10%).

Three load-bearing properties follow:

Manufactured convexity. Rebalancing multiplier x cushion into the risky arm makes exposure grow in trends (the ratchet) and shrink toward zero as NAV approaches the floor. The cost is paid in path — choppy markets bleed cushion — not in an option premium.
Only jumps can breach. Under continuous monitoring, a continuous price path cannot take NAV through the floor — the engine de-risks along the way. A breach requires a single-block loss larger than 1/multiplier, so 100 / multiplier is the alarm number published with every directive: the one-shot loss that would pierce the floor. Safe sizing is therefore bounded by the strategy's worst plausible single-block loss — which is exactly what the signed jump-exposure attestation certifies. This is why HAVEN multipliers can run materially higher than conventional capital-protection sizing: the multiplier is licensed by a per-strategy signed bound, and the engine refuses to run a jump-budget multiplier over any strategy that lacks one. The refusal is enforced in code, not by convention.
Self-coordination. When the wrapped strategy goes defensive, its live jump exposure falls and the multiplier rises automatically — the wrapper and the strategy de-risk and re-risk in concert.

Rebalancing is deliberately asymmetric: de-risking executes immediately; re-risking is drift-gated, so the wrapper reacts fast to danger and slowly to recovery. Below a configured cushion buffer the risky arm is sold entirely — a circuit breaker, not a tombstone: floor accrual can rebuild the cushion and re-enter. Tenor and floor yield are return levers as powerful as the multiplier, and extending tenor adds participation without lowering the gap-breach threshold.

Guarantee semantics¶

Principal is protected at maturity only. Early exits realize mark-to-market value: with a USDi-style floor, early exits track NAV closely; with a principal-token floor, early exits can price below par earlier in the tranche. An early-exit value table is produced and disclosed for every product. Structural residual risks that no multiplier removes — floor-asset failure, strategy-share redemption failure, and jumps beyond the attested bound — are published as disclosures, not treated as engineering targets.

The validation gauntlet¶

A HAVEN configuration ships only if it passes all of:

Full protection across all rolling historical tranches at the floor.
Full protection across a zero-drift Monte Carlo — the strategy's returns with drift removed, probed with block-bootstrap and fat-tailed resampling, so protection is never credited to the strategy simply having gone up historically.
Stress scenarios — large one-day gaps, fast crashes, grinding bears, and whipsaws — breaching only beyond the attested jump bound, and then with limited severity.
Intra-tranche drawdown consistent with the product's target audience.
A produced and disclosed early-exit value table.

Per-product results of this gauntlet, including behaviour beyond the attested bound, are part of the validation evidence available from ACM on request.

ARIA — the regime meta-allocator¶

ARIA is a simulation-path strategy (not a config-driven index) that allocates across four self-defending carry sleeves and one return sleeve using two deterministic daily rules — no forecasts, no discretion, prior-day data only:

Sleeve	Role	Self-defense
Funding rotator	Carry — best-of-venue delta-neutral perpetual funding	Exits when trailing funding turns negative
Levered stable-yield carry	Carry — looped staked-stablecoin yield	Levered only while the net spread over borrow costs stays positive by a margin
Fixed-rate ladder	Carry — laddered fixed-yield principal tokens	Fixed-rate by construction; issuer caps and pool-quality filters
Credit rotator	Carry — DeFi credit pools with a minimum hold	Trailing-yield rotation with a switching buffer
Momentum + productive cash	Return — BTC/ETH momentum, cash parked in USDi	Momentum gate sends the sleeve to USDi

Rule 1 — the carry-regime dial. A trailing measure of available carry is banded (rich / mid / thin) with hysteresis; the band sets the split between carry and momentum sleeves.
Rule 2 — the carry-engine selector. Among carry sleeves, trailing realized return picks the engine, with a switching buffer and a minimum hold to prevent churn. Capacity-aware allocation modes exist for larger deployments (winner-take-all versus a capped waterfall).

ARIA's live jump exposure — the input to its HAVEN attestation — is derived mechanically from the current carry weight and a per-engine risk budget, so the protection wrapper always sizes against what ARIA is actually holding.

The strategy catalog: from index to vault product¶

Indices become investable through the strategy catalog: the set of live, published products the platform can subscribe to. Each catalog entry carries the product's identity, its methodology reference, its verbatim risk-engine contract, and — for vault products — the underlying index and profile it subscribes to.

Current vault products are HAVEN wrappers over the family:

Product	Underlying	Protection floor (F)
HAVEN(ARIA) — defined-outcome flagship	ARIA	0.95
HAVEN(DGI)	DGI base	0.90
HAVEN(ATLAS) — launch tier	ATLAS base	0.90
HAVEN(ATLAS-L core)	ATLAS-L core	0.90 and 0.85 tiers
HAVEN(ATLAS-L aggressive)	ATLAS-L aggressive	0.90 and 0.85 tiers

All share a common spine: a jump-budget multiplier licensed by the underlying's signed bound, a USDi floor arm, 12-month tenors, and 30-second production monitoring. The index-subscription path is the regulatory-preferred pattern: the vault subscribes to a published index's signed jump-exposure attestation, and each execution directive arrives as a complete instruction — floor value, cushion, multiplier, target weight, per-asset splits — with zero decisions left to the executor. The vault itself never levers: its target strategy weight is always at most 100%.

How directives are consumed and gated on the platform side is covered in Execution & Strategy Layer.

How an index is defined and computed¶

At concept level:

Declarative, versioned definitions. Each index is a single declarative definition — identity, universe, weight methodology, profiles, risk-engine contract, and status — validated against a schema on load. Only definitions marked live are published.
Factory routing. A calculator factory routes each index to its engine: one generic signal-based engine driven directly by the declarative definition, and dedicated, proven calculator engines behind a common interface for the more specialised mechanisms (trend ensembles, carry loops, binary rotations). Formula evaluation for the generic engine runs in a restricted sandbox.
Deterministic backtest loop. Backtests and production share the same per-day loop: compute returns, apply yesterday's weights, drift, compute today's targets from data up to today, apply optional volatility targeting, check the rebalance threshold (with cooldowns and volatility-adjusted widening), record. Given the same inputs, the loop always produces the same outputs.
Composite-asset expansion. Composite holdings — delta-neutral basis pairs, child indices — are decomposed into executable legs in every published output, so a consumer always receives positions it can actually trade.

Research → validation → freeze¶

Every strategy travels a controlled path before publication:

Hypothesis first. Each experiment is logged before any backtest runs, with pre-registered kill criteria, data provenance, and the parameter neighbourhood to be swept. Killed experiments stay in the registry — a documented dead end saves the next researcher from re-testing it.
Real data only. Backtests run on real venue data pinned in frozen, checksummed snapshots that are hash-verified on load. There is exactly one sanctioned synthetic exception — a documented accrual convention for USDi history before the asset launched — and it is recorded in the affected products' disclosures.
Discipline rules. Costs are modeled from day one; parameters are selected on plateaus, not peaks; strategies are stress-tested across market epochs; reporting is required to be honest about weaknesses.
Validation gauntlet. Candidates face rolling-window analysis, walk-forward out-of-sample testing with an explicit overfitting verdict, Monte Carlo batteries, and — for protected products — the full HAVEN gauntlet described above.
Freeze. A shipping methodology is frozen as a hash-committed artifact set. From that point, published risk parameters change only through a controlled methodology version bump (see governance).

Validation evidence

The outputs of this workflow — tearsheets, tranche analyses, stress tables, and walk-forward results — are maintained per product and per methodology version. They are not published on this site; validation evidence is available from ACM on request.