Systematic Crypto · Hyperliquid Perpetuals

The research comes first. The fund follows.

Most funds raise capital, then find their edge. We do the opposite. Axon Ridge runs live systematic models on Hyperliquid perpetuals — real capital, verifiable on-chain — and doesn't open to outside investors until the evidence clears a standard most funds never impose on themselves.

See the methodology Live track record →
Live data ingestion · 51 perpetual markets streaming
GRID-50 K-of-N books + specialist roster in paper validation · track them live →
28+
Model Families Audited
Every model family in the modern literature gets the same audit: identical OOS protocol, identical cost model. No favourites. Most fail — that is the point.
15/15
Seeds Positive at the Final Gate
Promotion runs through a three-gate funnel — CPCV signal screen, cost sweep on locked predictions, then a 15-seed causal walk-forward with pre-registered thresholds. The current production family passed every seed.
~3 yr
L2 Order-Book History
Continuous L2 ladders, trades, funding and OI from April 2023, streamed live into a dedup-safe store. Real microstructure, not daily candles.
4 bps
Per-Leg Cost Floor
Every Sharpe quoted on this site is net of a realistic crossing cost. Gross-Sharpe charts are a salesman's tool, not a quant's.
The Process

An adversarial stance toward our own ideas.

Every paper claims an edge somewhere. Most evaporate under honest cost accounting, a single regime change, or a different random seed. Our working assumption is that every candidate signal is noise until proven otherwise — and we spend more effort trying to break an idea than to defend it. We know this works because we ran the audit on ourselves: our own early promotions turned out to be selection on noise, and the current process was rebuilt on the other side of that finding.

Promotion now runs through a pre-registered three-gate funnel — Grid A (signal structure under combinatorial purged cross-validation), Grid B (execution-cost sweep on locked predictions), Grid C (a fifteen-seed causal walk-forward that could actually have been traded). Every threshold is written down before the first candidate runs; the deploy object is the seed median, never the best seed; and when the backtest can't separate two variants, paper trading adjudicates — not a third sweep. Only a human moves anything from paper to live capital. The full protocol is public: the Grid A/B/C paper.

Operating Principles

Four rules we don't break.

These are hard constraints, not guidelines. They exist because promising results reliably dissolve the moment any one of them is relaxed — and we'd rather kill an idea early than lose capital later.

01
Evidence over hype
Every claim links to a measurable test, a primary source, or a file diff. Minimum search budgets are enforced before any finding is written down. Fluent summaries without underlying work are rejected.
02
Costs are not optional
All Sharpe figures are annualised dollar Sharpe at 4 bps per leg. Backtests without realistic fees and slippage are not findings — they are debunk paths. Information coefficient is never a stand-in for dollar return.
03
Negative evidence is required
Every promising finding must be paired with an active search for its failure modes — replications, falsifications, regime breaks. Debunked claims are preserved, not deleted, so the same mistake isn't repurchased later.
04
Pre-registered gates, seed medians
Promotion requires all three funnel gates — CPCV signal screen, cost survival on locked predictions, and ≥ 70% of fifteen pre-registered seeds positive on a causal walk-forward. Deployment uses the seed median, never the best seed; backtest iteration is a budget, not a free resource.
The Pipeline

From data to capital, in five steps.

Each stage has its own gate. Most candidates die at stages two and three, and that is the point. Survivors enter paper validation before any live capital is ever risked.

01 · Ingest

Market Data

L2 book, trades, funding, OI and mark feeds streamed from Hyperliquid into a dedup-safe ClickHouse store. Live since April 2026; archive from 2023.

02 · Screen

Grid A — Signal

Candidates train under combinatorial purged cross-validation on a 50-coin panel. Fold consistency, ridge floor, regression control, coin-half stability — the signal must be a property of the cross-section, not a lucky ticker.

03 · Validate

Grids B & C — Cost, then Causality

Grid B sweeps execution on locked predictions at 0/4/8 bps per leg. Grid C retrains the survivor as a monthly expanding walk-forward under 15 pre-registered seeds — the test you could actually have traded.

04 · Paper

Paper Validation

Survivors run on live data in simulated execution — same fills, same costs, same scheduling — until pre-registered gates clear over a sustained window.

05 · Deploy

Live Execution

Risk-gated IOC limit orders. Per-coin notional caps, leverage limits, daily loss kill, per-strategy drawdown kill, manual override. Every fill audited.

Current Book

Market-neutral by construction — live, paper, and next in line.

Our first generation of directional single-coin strategies taught us an expensive lesson: their edge was a beta-shaped hot streak, and we promoted it at the peak (the post-mortem is public). Everything that runs live now is a long/short book that cancels directional exposure by construction. New candidates earn their way through paper on identical instrumentation before any promotion decision — which is made by a human, not a pipeline. Specific signal mechanics are not disclosed.

BOOK · 01 LIVE

Specialist K-of-N book — market-neutral

The live-capital book: validated per-coin signals ranked daily against each other, long the strongest, short the weakest, dollar-neutral, with staleness guards that force-flatten any leg whose signal stops refreshing. Small real capital, every fill audited against the paper mirror.

Shape
long/short · neutral
Cost convention
4 bps / leg
Status
live · real capital
Rebalance
daily
BOOK · 02 PAPER VALIDATION

GRID-50 pooled ranker — two constructions

The first family through the full Grid A/B/C funnel: one pooled cross-sectional ranker on a 50-coin universe, 15/15 seeds positive on the causal walk-forward. Two book constructions trade side by side in paper — same predictions, different execution shape — because the backtest genuinely couldn't separate them (write-up).

Universe
50 coins
Gate C
15/15 seeds +ve
Status
paper · 2 books
Rebalance
daily · K-of-N
ROSTER PAPER VALIDATION

Specialist roster + conviction-sized mirrors

The paper lab: single-coin specialists from the Grid-era funnels, each on a notional $10k book scored hourly on live prices, plus conviction-sized mirrors that scale notional by signal strength — an A/B test of position sizing running on identical signals. This roster is a microscope and a candidate pool, not live capital: it tells us which names carry genuinely bidirectional signal and feeds the next book generation. Every strategy is tracked publicly.

Scoring
hourly · live px
Status
paper · tracking
Sizing A/B
flat vs conviction
Centring
causal rolling
Transparency

What we can say — and what we won't.

The most common failure mode in this industry is overclaim. The second is overshare. We try to be precise about both — explicit on what we have built and measured, deliberately quiet on the recipes themselves.

What we can defend

Built, measured, in flight.

  • A continuous Hyperliquid data stack — live L2, trades, funding, OI — backed by three years of archive.
  • A large, audited architecture screen with consistent OOS protocols, costed at 4 bps per leg.
  • A pre-registered three-gate promotion funnel (Grid A/B/C) with published methodology and published failures — including our own.
  • A live market-neutral book on real capital, plus a paper roster on identical instrumentation.
  • A live execution stack with documented risk gates, kill switches, and post-fill reconciliation.
  • A preserved record of negative findings — debunked claims, replication failures, methodology corrections.
What we won't disclose

The recipe stays in the kitchen.

  • Specific signal mechanics, feature sets, or model architectures used inside the live sleeves.
  • Asset selection, position cadence schedules, or rebalance triggers at a level that could be replicated.
  • That simulated Sharpe figures will reproduce at scale — markets, fills and costs move. We track the drift live.
  • A multi-year live track record. The live book is recent; the track record is being earned, not asserted.
  • Any regulatory status as an investment adviser, fund, or pooled vehicle. We're not, and we don't pretend to be.
Early Access

When the live book has a story to tell.

We don't publish often. When the live sleeves reach a defensible track record, when methodology meaningfully shifts, or when we open to external capital, this list hears about it first. No marketing copy, no daily commentary.

// thanks — you'll hear from us when it matters.