All monetary figures denominated in United States Dollars (USD) unless otherwise stated.
Noether Capital is a private, dark quantitative trading operation. It has no external investors, no fund structure, no public visibility. It operates HIRO (Hilbert Intelligent Research & Trading Organism), an AI-native algorithmic trading system designed to industrialise the discovery and harvesting of low-capacity, high-yield, structurally forced market inefficiencies across crypto perpetuals, equity micro futures, commodity micro futures, and Foreign Exchange (FX) micro futures.
HIRO is not a mini hedge fund running scaled-down versions of institutional strategies. It is a micro-alpha extraction organism — a machine that continuously discovers, validates, deploys, and retires many small, fast-cycling, mechanism-bound operators that exploit structural dislocations too small, too messy, or too transient for large firms to pursue. The operator retains 100% of all profits. There are no management fees, no performance fee sharing, no investor reporting, and no compliance overhead from managing other people's money.
The competitive advantage is not prediction accuracy. It is alpha metabolism — the speed at which the system discovers new edges, deploys capital into them, extracts value, detects decay, kills dead operators, and replaces them. At small capital, that metabolism translates directly into extraordinary return on equity because the opportunity set is vast relative to the capital deployed.
Large quantitative firms ($1B+ Assets Under Management (AUM)) are constrained by minimum capacity thresholds, internal approval processes, model governance overhead, and the inability to deploy into opportunities that only support small notional. A $10B fund cannot justify pursuing an edge that absorbs $50K before self-impact degrades it. A $1M firm can extract the full value of that edge and move on.
The global market generates thousands of these micro-opportunities every day — funding rate dislocations, liquidation cascade reversals, session boundary distortions, cross-venue lag, queue depletion snapbacks, post-event normalisation, exchange rule-change effects. Each one is small. Each one is brief. Each one is structurally forced. Collectively, they represent an enormous extractable surface for a system built to harvest them at subscale.
The target is $1M initial capital generating $3M–$4M in net extraction over 12 months, representing 300–400% annual return. This is the engineering target that shapes every architectural decision. It is not a guarantee — it is the design point around which capital velocity, operator count, edge density, and kill speed are calibrated.
300% annual return on $1M = $3M net extraction across approximately 250 trading days = $12,000 average daily net extraction. But daily averages are misleading. Extraction is concentrated in high-density windows — not uniformly distributed. The realistic model is $3K–$5K on quiet days and $20K–$40K on high-density days, with the majority of annual return coming from the highest-density 40–60 trading days.
The $1M does not sit in positions all day. It cycles through multiple opportunities. If average holding period is 2 hours and 30 operators are active, the same $1M is effectively deployed 4–8 times per day. Effective daily notional = $4M–$8M.
| Variable | Conservative | Target | Aggressive |
|---|---|---|---|
| Active operators | 25 | 40 | 60 |
| Avg trades per operator per day | 2 | 3 | 5 |
| Total trades per day | 50 | 120 | 300 |
| Avg holding period | 4 hours | 2 hours | 45 min |
| Daily capital turnover | 3× | 6× | 12× |
| Effective daily notional | $3M | $6M | $12M |
| Required avg net edge per trade | 40 bps | 20 bps | 8 bps |
| Implied daily net extraction | $12,000 | $12,000 | $12,000 |
| Implied annual return | 300% | 300% | 300% |
The Target column is the design point: 40 active operators, 120 trades/day, 2-hour average hold, 6× daily capital turnover, 20 bps average net edge per trade. This is achievable on crypto perpetuals for mechanism-bound, structurally forced opportunities where the edge comes from market plumbing, not prediction.
Noether Capital operates through a Singapore-registered entity. As a proprietary trading firm (not a Digital Token Service Provider), no MAS licence is required, but the system must comply with tax obligations (17% corporate rate) and maintain auditable records.
HIRO implements automated compliance through S36A:
All trading activity is logged to the standard required by Singapore MAS even though the licence threshold is not triggered. This protects the business if regulatory requirements change or if the operation scales to a point where licensing becomes necessary.
The plan lives or dies on two numbers: average net edge per trade (target: 20 bps) and daily capital turnover (target: 6×). If net edge drops to 10 bps, annual return drops to ~150%. If capital turnover drops to 3×, annual return drops to ~150%. Both degrading simultaneously produces ~75%. These two metrics are the primary Key Performance Indicators (KPIs) from Month 3 onward.
HIRO does not seek consistent daily extraction. It seeks maximum extraction during high-density opportunity windows and accepts near-zero activity during low-density periods.
| Window Type | Frequency | Duration | Expected Density | Mechanism Families |
|---|---|---|---|---|
| Funding rate reset | 3× daily (8h cycle) | 15–45 min around reset | High | MP-02 |
| Liquidation cascade | 2–5× weekly | 5–30 min | Very high | FF-02 |
| Session boundary handoff | Daily (Asia→Europe→US) | 30–60 min per handoff | Medium-high | MP-05 |
| Post-news microstructure normalisation | Variable | 5–60 min after event | High | IT-01 |
| Cross-venue dislocation | Multiple daily | Seconds to minutes | Very high (brief) | IL-03 |
| Queue depletion / refill | Continuous during volatility | Seconds to minutes | Medium | LS-01, LS-02 |
| Exchange rule change | Monthly (irregular) | Hours to days | Very high (rare) | MP-07, MP-08 |
| Contract roll / expiry | Monthly/quarterly | Hours | Medium-high | MP-04 |
The aggregate of these windows across 20+ symbols provides sufficient opportunity surface. Not every window produces edge. But enough do, frequently enough, to sustain the target extraction rate when the system is properly calibrated.
HIRO is specified in the Genesis Framework — 77 sections covering every layer from data ingestion to execution. The system discovers alpha through 20+ parallel discovery engines, validates it through multi-stage testing (statistical, counterfactual, ecological), deploys it as operators with dynamic Capital Fractional Unit (CFU) allocation, executes through alpha-aware routing, and retires operators when edge decays. A full architectural summary is available in the Genesis Framework document.
In Year 1, AI inference for Genesis hypothesis generation and adversarial review is provided via frontier model APIs (Claude, GPT-4o). These are used on the research path, not the execution path — API latency is irrelevant because outputs are operator specifications that undergo hours of validation before deployment. By Year 2, if monthly API spend exceeds the cost of a dedicated GPU server (~€200–400/month), the system transitions to self-hosted open-weights models fine-tuned on 12+ months of HIRO operator history, mechanism taxonomy, validation outcomes, and failure modes. This eliminates external data exposure and produces a model specifically calibrated to HIRO's domain. The decision trigger is purely economic: when self-hosted cost < API cost at comparable quality, switch.
HIRO is not hand-coded. It is constructed by the Hilbert AI Software Factory — an AI-driven build system that converts the Genesis Framework specification into verified, tested, deployed code.
The Factory operates on Helsinki (Server B) and uses a dual-AI verification pipeline: Claude (defence) proves code matches specification, GPT-4o (prosecution) actively searches for errors. Code must pass BOTH to be deployed. For Rust execution components (L6), a mandatory 7-stage quality pipeline runs before AI verification: cargo check → clippy → test → Miri → audit → bench → fmt.
This approach provides three advantages over traditional development:
The Factory specification is 59 sections. The Code Creator dashboard tracks 97 build packets across 10 construction steps.
| Server | Location | Specification | Role | Cost |
|---|---|---|---|---|
| A1 (Binance Execution) | AWS Osaka (ap-northeast-3) | Bare-metal or dedicated — 8+ cores, 64GB+ RAM, NVMe, 1Gbps+ | Primary Centralised Exchange (CEX) execution. All Binance orders, market data ingestion, Guardian, Risk Envelope, Position Reconciliation, MSE. <5ms to Binance matching engine. | ~$300/mo |
| A2 (Bybit + DeFi Execution) | AWS Singapore (ap-southeast-1) | Bare-metal or dedicated — 8+ cores, 64GB+ RAM, NVMe, 1Gbps+ | Secondary CEX execution (Bybit, OKX). All DeFi execution — flash loans, oracle liquidations, CEX-DEX arb, Jito Asian bundles. <5ms to Bybit. <100ms to Solana validators. | ~$300/mo |
| B (Research) | Hetzner Helsinki (AX102) | AMD Ryzen 9 7950X, 16 cores, 128GB DDR5, 2TB NVMe | All discovery engines (continuous), Language Engine, Causality Engine, Ecological Simulation, Backtesting, Genesis, Portfolio Optimiser. CPU-heavy, latency-irrelevant. | ~€160/mo |
| C (European DeFi Maximal Extractable Value (MEV)) | Hetzner Frankfurt | Dedicated — 8+ cores, 64GB RAM, NVMe | European DeFi MEV execution. <5ms to Jito Frankfurt relay. Flashbots Protect. When Solana slot leader is a European validator, C fires the bundle instead of A2. | ~€160/mo |
Each server exists because it is the fastest path to a specific extraction venue. A1 in Osaka is <5ms to Binance — the deepest liquidity, most liquidation data, and highest funding volume in crypto. A2 in Singapore is <5ms to Bybit and <100ms to Solana validators — covering the second-largest exchange and all DeFi execution. B in Helsinki provides maximum compute per dollar for CPU-heavy discovery. C in Frankfurt is <5ms to Jito's European relays — when the Solana slot leader is a European validator, C gets the bundle included faster than A2 can from Singapore.
Five servers would produce diminishing returns. Four covers every latency-sensitive destination with <5ms or near-optimal RTT. The marginal gain from a fifth server is single-digit milliseconds on edge cases — not worth the management complexity.
The four-server architecture creates HIRO's primary competitive advantage: cross-domain speed. When A1 (Osaka) detects a liquidation cascade on Binance, it simultaneously executes the CEX reversal trade (<5ms) AND signals A2 (Singapore) via direct tunnel (~70ms Osaka→Singapore) that a DeFi flash loan opportunity may be opening. A2 cross-references against on-chain oracle state (Helius Remote Procedure Call (RPC)) and fires the flash loan via Jito if the oracle hasn't updated yet. Total time from Binance price move to flash loan submission: ~100ms. Pure DeFi bots that only watch the blockchain are still waiting for the oracle to update.
When the Solana slot leader is a European validator, A2 signals C (Frankfurt) instead. C submits the Jito bundle at <5ms to the Frankfurt relay. A2 and C coordinate via direct tunnel (~180ms) to avoid duplicate submissions — the one with the lower-latency path to the current leader fires.
| Link | Round-Trip Time (RTT) | Protocol | Traffic Type |
|---|---|---|---|
| A1 (Osaka) ↔ A2 (Singapore) | ~70ms | WireGuard Virtual Private Network (VPN) | Cross-domain signals: cascade detected → fire flash loan. Time-sensitive. Encrypted. |
| A2 (Singapore) ↔ C (Frankfurt) | ~180ms | WireGuard VPN | MEV coordination: which server submits the bundle based on current slot leader location. Encrypted. |
| B (Helsinki) ↔ A1 (Osaka) | ~250ms | WireGuard VPN | Research→Execution: operator deployments, parameter updates, CFU changes. Not time-critical. |
| B (Helsinki) ↔ A2 (Singapore) | ~250ms | WireGuard VPN | Research→DeFi: DeFi operator deployments, flash loan strategy parameters. Not time-critical. |
| B (Helsinki) ↔ C (Frankfurt) | ~30ms | WireGuard VPN (or Hetzner vSwitch if same DC) | Research→MEV: European DeFi operator updates. Low latency — both in Europe. |
| Destination | From Helsinki (old) | From Optimal Server | Improvement |
|---|---|---|---|
| Binance matching engine (Osaka) | ~250ms | <5ms (A1 Osaka) | 50× |
| Bybit matching engine (Singapore) | ~280ms | <5ms (A2 Singapore) | 56× |
| OKX (Hong Kong / Singapore) | ~260ms | <30ms (A2 Singapore) | 9× |
| Solana validators (Asia) | ~280ms | <100ms (A2 Singapore) | 3× |
| Jito relay (Frankfurt) | ~30ms | <5ms (C Frankfurt) | 6× |
| Jito relay (Amsterdam) | ~40ms | ~10ms (C Frankfurt) | 4× |
| Flash loan: Binance cascade → on-chain liquidation | ~500ms+ | ~100ms (A1→A2 pipeline) | 5× |
| Phase | Servers Active | RPC / Data | Monthly Total |
|---|---|---|---|
| Phase 1 (Month 1–4) | A1 Osaka ($300) + B Helsinki (€160) | Alpaca $99 | ~$575 |
| Phase 2 (Month 5–6) | + A2 Singapore ($300) | + Helius $200, Alchemy $100, data feeds $350 | ~$1,525 |
| Phase 3 (Month 7+) | + C Frankfurt (€160) | Same | ~$1,700 |
Full four-server infrastructure at Phase 3: ~$1,700/month or ~$20K/year. On target extraction of $3.5M+/year, that is 0.6% of revenue. Each server pays for itself if it captures just one additional opportunity per month that the previous architecture would have missed.
The 4-server architecture is monitored by the Visual Infrastructure Mesh (S39A), a real-time topological map showing:
If the Osaka-Singapore corridor degrades, the system automatically stalls only the operators dependent on that specific corridor — cross-venue arb pauses but DeFi execution on Singapore continues independently.
Corridor baselines are maintained in configuration and updated when infrastructure changes are detected by the Horizon Scanning module (S40).
| Tier | Phase | Symbols | Rationale |
|---|---|---|---|
| Tier A: Core Harvest | Month 1+ | BTC, ETH, SOL perpetuals | Deepest mechanism density. 24/7 trading. Highest edge density per hour. Concentrate the machine here first. |
| Tier A: Expanded Core | Month 3+ | BNB, XRP, AVAX, DOGE perpetuals | Secondary crypto with sufficient liquidity and distinct microstructure characteristics. |
| Tier B: Opportunistic | Month 5+ | Micro E-mini S&P 500 (MES), Micro E-mini Nasdaq-100 (MNQ), Micro Gold (MGC), Maximum Capital at Live risk (MCL), M6E micro futures | Monitored continuously. Activated during strong mechanism windows (session opens, macro events, roll periods). |
| Tier C: Research Reservoir | Month 6+ | M2K, Micro E-mini Dow (MYM), Micro Silver (SIL), Micro Copper (MHG), M6B, Micro Japanese Yen (MJY), M6A + expansion candidates | Watched for newborn alpha and structural shifts. Not actively traded unless density justifies activation. |
HIRO does not trade patterns. It trades structural necessity — situations where market architecture forces price displacement regardless of fundamental value. Every operator in the system is bound to at least one mechanism from the Causality Engine's taxonomy (S24Q). The mechanism tells the operator when the edge should exist, when it should not, and what would cause it to disappear.
The identification of structural mechanisms is not speculative — it is grounded in the UHS Forensics & Analogue Engine (S38). This subsystem reverse-engineers historical episodes of exceptional trading economics, isolates the timeless structural components, and searches current markets for analogous conditions.
The engine maintains three libraries:
This provides HIRO with a disciplined, evidence-based approach to alpha discovery rather than random feature scanning.
| Family | ID | Mechanism | Edge Type | Expected Contribution |
|---|---|---|---|---|
| Forced Flow | FF-02 | Margin liquidation cascade | Post-cascade reversal or continuation after forced flow exhaustion | High — most crypto P&L concentration comes from cascade events |
| Market Plumbing | MP-02 | Funding rate reset distortion | Pre/post funding rate positioning and convergence | High — occurs 3× daily with reliable mechanism |
| Market Plumbing | MP-05 | Session boundary distortion | Session handoff inventory reset, opening range exploitation | Medium — daily occurrence, moderate edge per event |
| Liquidity Structure | LS-01/02 | Liquidity vacuum / passive absorption | Post-sweep snapback, refill pattern exploitation | Medium — continuous during volatility |
| Instrument Linkage | IL-03 | Cross-venue fragmentation | Short-horizon dislocation convergence between venues | Medium — frequent but brief |
| Information Transmission | IT-01 | Post-news digestion asymmetry | Overreaction/underreaction normalisation | Variable — depends on news calendar |
| Market Plumbing | MP-07/08 | Fee / liquidation engine changes | Newborn alpha from venue structural changes | Very high per event. Frequency increasing as exchanges compete on features. Primary candidates for S24P transient fast-track deployment. Venue Microstructure Change Monitor (S3D.7a) auto-triggers discovery. |
| Hedging Pressure | HP-01/04 | Delta hedging / inventory pressure | Predictable directional flow from hedging obligations | Medium — requires options data (Phase 2+) |
The top three mechanism families (Forced Flow, Funding Reset, Liquidity Structure) are expected to generate 60–70% of total extraction in Year 1. The remaining families provide diversification and additional extraction during their specific activation windows.
These mechanism families map directly to the UHS Canonical Neighborhood Types defined in S38.9:
| Mechanism Family | S38 Neighborhood | Primary Discovery Engine |
|---|---|---|
| State transmission asymmetry | Lag-Sync (S38.9.1) | S24Q Causality Engine |
| Forced non-economic flow | Forced Exit (S38.9.2) | S24P Transient Fast-Track |
| Venue/fee rule changes | Plumbing Transition (S38.9.3) | S40 Horizon Scanning → S24P |
| Delayed price incorporation | Information Bottleneck (S38.9.4) | S24A Market Behaviour |
| Small-capacity structural edge | Capital-Fragmented (S38.9.5) | S24A, S24P |
| Market maker inventory imbalance | Inventory Stress (S38.9.6) | S24Q Causality Engine |
| Regime transition repricing | Regime Transition (S38.9.7) | S04 → S24A |
The 300% target is not achieved by one or five excellent strategies. It is achieved by running 30–60 operators simultaneously, each capturing a different micro-edge, each with small capital binding, each cycling fast. If each operator averages 2–4 trades per day at $60–$300 net per trade, the aggregate builds.
| Class | Population Target | Avg Lifespan | CFU Binding | Role |
|---|---|---|---|---|
| Durable Core | 10–15 operators | Weeks to months | 3–10 Capital Fractional Units (CFUs) each | Baseline extraction. Stable mechanisms. Consistent but modest per-trade edge. |
| Transient Swarm | 15–30 operators | Hours to days | 1–5 CFUs each | Compound Annual Growth Rate (CAGR) acceleration. Short-lived structural windows. High per-trade edge, brief existence. |
| Symbol-Specific | 5–15 operators | Days to weeks | 1–3 CFUs each | Exploit instrument-specific mechanics. BTC funding timing, SOL liquidation dynamics, etc. |
The transient swarm is the primary CAGR driver. Durable core provides stability and baseline. Symbol-specific operators capture the highest-conviction, most structurally grounded opportunities on individual instruments.
No operator may hold capital for more than 48 hours without demonstrating positive rolling RAR. If Risk-Adjusted Return (RAR) is negative for 48 consecutive hours, the operator is automatically suspended and its CFUs returned to the reserve pool. The Guardian may grant exceptions for specific operators where the bound mechanism has a documented longer activation cycle, but the default is aggressive kill. Capital trapped in a dead operator is capital not available for a live opportunity. At 300% target, capital velocity is the return driver — every idle dollar degrades CAGR.
HIRO does not only search for alpha within the current market structure. The Structural Intelligence & Horizon Scanning module (S40) searches for changes TO the market structure itself — new frictions, new forced flows, new asymmetries.
S40 scans 7 vectors on a dual cadence:
Findings are synthesised by Claude and GPT-4o (adversarial) and converted into Search Directives that feed directly into the discovery engines. This means HIRO can identify and exploit structural changes (e.g., a venue fee change or new DeFi primitive) BEFORE they are visible in price action.
Budget: approximately $30/month for search APIs and AI synthesis.
Target: Zero P&L. System stability. Zero ghost positions. Reconciliation accuracy <1 bps deviation. Attribution engine producing decomposed returns on paper trades. Flash loan contracts tested on testnet. Cross-domain pipeline latency verified <100ms.
| Month | Milestone | Go Gate | No-Go Action |
|---|---|---|---|
| 2 | Paper trading stable | Zero ghost positions. Reconciliation <1 bps. 10+ operators validated. Attribution operational. | Fix infrastructure. No live capital until all gates pass. |
| 3 | First month live ($100K) | Avg net edge >15 bps/trade. Max drawdown <5% deployed. Attribution shows signal alpha > execution drag. | Reduce to $50K. Diagnose edge vs execution vs data. |
| 4 | Second month live | Cumulative P&L positive. Discovery producing 2+ validated operators/week. Capital velocity >3×. | Pause expansion. Focus on pipeline quality. |
| 5 | Scale to $500K | Monthly P&L >$30K on $500K (>70% annualised). Operator count >20. 3+ mechanism families active. | Hold at $500K. Investigate concentration risk. |
| 6 | Scale to $1M | Monthly P&L >$50K on $1M. Capital velocity >5×. Attribution clean (residual <15%). | Hold at current level. Diagnose velocity or edge issues. |
| 9 | Full extraction mode | Rolling 3-month annualised return >150%. Max drawdown <10% NAV. 30+ operators active. | If <100%: review mechanism families. If <150%: strong business, below 300% target. |
| 12 | Year-end assessment | Cumulative return >200%. | If 150–200%: exceptional outcome. If 100–150%: strong outcome. If <100%: review strategic thesis. |
The go/no-go gates are designed so that underperformance is diagnosed and addressed early, not tolerated. But they also acknowledge that 150% annual return is an exceptional outcome and 100% is still excellent. The 300% target shapes the architecture. The gates ensure the system is producing real value even if the target is not fully reached.
| Limit | Threshold | Action |
|---|---|---|
| Per-operator drawdown | 1.5% of operator's allocated CFUs | Operator killed. CFUs returned to reserve. |
| Per-symbol drawdown | 3% of total Net Asset Value (NAV) | All operators on that symbol paused. Symbol enters HOLD. |
| Aggregate daily drawdown | 5% of NAV | L3 Global Kill. All positions flattened. Trading suspended until next session with operator sign-off. |
| Aggregate weekly drawdown | 8% of NAV | System enters review mode. All transient operators retired. Only durable core permitted. Full diagnostic review required. |
| Maximum single position | 3% of 15-minute average volume (MPS) | Pre-order validation rejects oversized orders. |
| Transient operator CFU ceiling | 15% of total deployed CFUs | No new transient operators activated until utilisation drops. |
| Volume-Synchronised Probability of Informed Trading (VPIN) toxicity | >0.80 on target symbol | No new orders. Passive orders cancelled. Transients paused. |
| Aggregate symbol exposure (Anti-Coalescence) | Total exposure across ALL active operators on any single symbol must not exceed 0.5% of that symbol's 1-minute median volume | New orders on that symbol blocked until aggregate exposure drops below threshold. Prevents the swarm from collectively becoming detectable even though each operator individually stays under MPS. This is the quantitative "ghostability" constraint. |
| Temporal Jitter (Anti-Pattern Detection) | When multiple operators fire on the same symbol within the same mechanism window, order submission is staggered across a randomised 200–500ms jitter window | Prevents exchange pattern-matching from linking coordinated orders as swarm behaviour. Each operator's submission time is randomised within the window. No two orders on the same symbol may be submitted within 50ms of each other. Mandatory for all mechanism windows where 3+ operators fire simultaneously. |
| Risk | Mitigation |
|---|---|
| Ghost positions | 60-second position reconciliation (S36.4). Any mismatch = immediate alert + order suspension. |
| Server failure | Two-server architecture. Server A failure: all trading stops safely. Server B failure: trading continues with cached parameters. |
| Exchange outage | Multi-venue routing (S30B). Venue Health Score (S31) auto-pauses on degraded venues. |
| Data feed failure | Provider failover (S3A.9). Stale data detection at 30-second threshold (S36.9). Affected symbols enter HOLD. |
| Alpha decay faster than replacement | Continuous discovery on Server B. Operator birth rate tracked by S24T. Alert if birth rate < death rate for 7 consecutive days. |
| Key Performance Indicator (KPI) | Target | Alert Threshold | Source |
|---|---|---|---|
| Average net edge per trade | ≥20 bps | <15 bps for 5 consecutive days | S35A Attribution |
| Daily capital turnover | ≥5× | <3× for 3 consecutive days | S28 CFU utilisation |
| Active operator count | 30–60 | <20 for 3 consecutive days | S24T Orchestration |
| Operator birth rate | ≥2 per week | <1 per week for 2 consecutive weeks | S24T Discovery pipeline |
| Transient operator rotation velocity | ≥3 births + deaths per day | <1 per day for 5 consecutive days | S24P metrics |
| Signal alpha / total return ratio | ≥60% | <40% (implies returns are from tailwind, not signal) | S35A Attribution |
| Execution drag / signal alpha ratio | ≤30% | >50% (execution consuming too much edge) | S35A Attribution |
| Max drawdown (rolling 30-day) | <8% NAV | >5% NAV triggers review | S32 Risk Envelope |
These KPIs are monitored through the Genesis Numerical Diagnostics & Integrity Dashboard (S39, GNDID). The GNDID provides real-time visibility across four layers:
Design target: the Research Director can answer “Why aren’t we trading?” within 15 seconds of opening the GNDID, without looking at raw server logs.
Manual parameter tuning is available through the GNDID Fine-Tuning Interface: α (Aggression), β (Friction Buffer), γ (Decay Sensitivity), δ (SNR Threshold), ε (Correlation Ceiling). Guardian limits (S33) cannot be overridden.
A real-time panel compositing signals from S24Q (mechanism activations), S24N (language state contradictions), S31.10a (VPIN), S04 (MSE regime), and Venue Health Score into a single ranked view showing per-symbol opportunity density at any moment. The top 5–10 symbols are where capital should be concentrated. This is the primary operational interface for managing the extraction engine.
| Scenario | Avg Net Edge | Capital Velocity | Year 1 Net Return | Year-End NAV |
|---|---|---|---|---|
| Bear Case | 10 bps | 3× | ~75% | $1.75M |
| Base Case | 15 bps | 5× | ~190% | $2.9M |
| Target Case | 20 bps | 6× | ~300% | $4.0M |
| Bull Case | 25 bps | 8× | ~500% | $6.0M |
HIRO's Tier 1 extraction engine operates on a fixed $1M capital base. As profits accumulate, they are swept to a Tier 2 capital reserve on a defined schedule. This serves three purposes: it maintains the ghost footprint (keeping Tier 1 capital small), it protects realised profits from Tier 1 drawdown events, and it builds a capital reserve that can fund Tier 2 strategies, re-inject into Tier 1 during extreme opportunity windows, or be withdrawn as income.
| Rule | Specification |
|---|---|
| Sweep frequency | Weekly (every Friday UTC close). Profits above the $1M Tier 1 base are swept to Tier 2. |
| Minimum sweep | $5,000. If accumulated weekly profit is below $5K, it remains in Tier 1 until the next sweep. |
| Tier 1 replenishment | If Tier 1 NAV drops below $900K due to drawdown, Tier 2 injects capital back to restore $1M. This is automatic up to $100K per injection. Larger replenishments require Guardian approval. |
| Tier 1 maximum | $1.2M. If intra-week profits push Tier 1 above $1.2M, an immediate mid-week sweep is triggered. This prevents Tier 1 from growing beyond the ghost footprint threshold. |
| Tier 2 deployment | Tier 2 capital is held in stable assets (USDT/USDC on exchange, or stablecoin yield if available above risk-free). Tier 2 is not actively traded in Year 1 — it is a reserve and profit accumulation vehicle. |
| Month | Bear (75% Annual Return (AR)) | Base (190% AR) | Target (300% AR) | Bull (500% AR) |
|---|---|---|---|---|
| 1–2 | Phase 0: Paper trading. No P&L. Infrastructure build. | |||
| 3 | $1,010K | $1,025K | $1,040K | $1,060K |
| 4 | $1,020K | $1,055K | $1,085K | $1,130K |
| 5 | $1,040K | $1,100K | $1,170K | $1,280K |
| 6 | $1,070K | $1,170K | $1,310K | $1,520K |
| 7 | $1,110K | $1,260K | $1,500K | $1,830K |
| 8 | $1,150K | $1,360K | $1,720K | $2,200K |
| 9 | $1,200K | $1,480K | $1,980K | $2,650K |
| 10 | $1,260K | $1,620K | $2,280K | $3,200K |
| 11 | $1,330K | $1,780K | $2,630K | $3,870K |
| 12 | $1,400K | $1,960K | $3,030K | $4,700K |
Note: total capital = Tier 1 ($1M fixed) + Tier 2 (accumulated sweeps). Growth shown above is total capital across both tiers. The Tier 1 extraction engine always operates on ~$1M. Profits accumulate in Tier 2.
Month 3–4 shows slower growth because only $100K is deployed (Phase 1 Controlled Live). Full $1M deployment begins Month 5–6. The acceleration from Month 7 onward reflects full capital velocity on all mechanisms.
Noether Capital is a private, dark quant operation. There are no external investors, no fund management obligations, no regulatory overhead from managing other people's money. All profits are retained. The entire organisational drag that slows institutional firms — investor reporting, compliance committees, marketing, lock-up structures — does not exist. This is a structural advantage that compounds over time.
As Tier 2 capital accumulates from weekly profit sweeps, it is deployed through a self-compounding architecture:
| Tier 2 Balance | Deployment |
|---|---|
| $0–$250K | Pure reserve. Held in stablecoins (USDT/USDC) on exchange or in yield-bearing stablecoin protocols if yield exceeds 5% Annual Percentage Rate (APR) with acceptable counterparty risk. Primary function: Tier 1 drawdown replenishment buffer. No active trading. |
| $250K–$750K | Reserve ($250K maintained) + second Tier 1 extraction engine ($500K–$1M). The second engine runs on the same infrastructure (Server B handles discovery for both engines) but operates on a different symbol subset — for example, Engine 1 on BTC/ETH/SOL, Engine 2 on secondary crypto + equity micro futures. Each engine maintains its own ghost footprint. Combined extraction doubles without increasing per-symbol visibility. |
| $750K–$2M | Two Tier 1 engines ($1M each) + Tier 2 active strategies on remaining capital. Tier 2 strategies are longer-horizon, higher-capacity: basis arbitrage (perp vs spot), cross-venue carry, funding rate harvesting at scale, tail-risk hedging. Target: 30–80% AR on Tier 2 capital. Lower velocity, higher capacity, different mechanism families. |
| $2M–$5M | Three Tier 1 engines ($1M each, different instrument universes) + Tier 2 active ($2M+). Add third server for third extraction engine. Total extraction surface spans crypto, equity, commodity, and FX across three independent ghost-footprint engines. Personal income withdrawal enabled from Tier 2 without reducing any Tier 1 engine capital. |
| $5M+ | Maximum self-funded scale. Four to five Tier 1 engines across diversified instrument classes and geographies. Tier 2 at $2M–$3M in active strategies. Remaining capital in long-term wealth preservation (stablecoin yield, treasury instruments, physical assets). Total annual extraction potential: $5M–$15M depending on market conditions. No employees. No investors. No overhead beyond infrastructure and data. |
The dark fund model compounds faster than an institutional model because there is no management fee drag, no performance fee sharing, no investor redemption risk, no compliance overhead, and no organisational complexity. Every dollar of profit is available for redeployment. The trade-off is that growth is limited to retained earnings — but at 300% AR on Tier 1, retained earnings grow fast.
| Cost Category | Detail | Monthly | Annual |
|---|---|---|---|
| Infrastructure | |||
| Server A1 (Binance Execution) | AWS Osaka bare-metal — co-located with Binance matching engine | ~$300 | $3,600 |
| Server A2 (Bybit + DeFi) | AWS Singapore bare-metal — co-located with Bybit, DeFi execution | ~$300 | $3,600 |
| Server B (Research) | Hetzner Helsinki AX102 — Ryzen 9 7950X, 128GB DDR5, 2TB NVMe | €160 (~$175) | $2,100 |
| Server C (EU DeFi MEV) | Hetzner Frankfurt — dedicated, near Jito relay | €160 (~$175) | $2,100 |
| Inter-server VPN | WireGuard mesh: A1↔A2↔B↔C | Included | $0 |
| Domain + Secure Sockets Layer (SSL) | Dashboard hosting, monitoring endpoint | ~$10 | $120 |
| Infrastructure subtotal | $960 | $11,520 | |
| Data Feeds | |||
| Alpaca Markets | Unlimited plan — US equity real-time, Securities Information Processor (SIP) feed | $99 | $1,188 |
| LunarCrush | Crypto social metrics — Phase 1 | ~$100 | $1,200 |
| Santiment | Crypto social + on-chain — Phase 1 | ~$50 | $600 |
| Benzinga Pro | News feed API — Phase 1 | ~$80 | $960 |
| Trading Economics | Macro calendar API — Phase 1 | ~$30 | $360 |
| CryptoPanic | Crypto news aggregator — Phase 1 | Free tier | $0 |
| Wikipedia Pageviews | Wikimedia REST API — Phase 1 | Free | $0 |
| FRED | Federal Reserve macro data — Phase 1 | Free | $0 |
| Binance / Bybit | Market data + liquidation streams — WebSocket | Free | $0 |
| Polygon.io (Phase 2) | Equities L2 — when equity micros go active | ~$80 | $960 |
| Helius Solana RPC | Trader Node — premium Solana RPC with Jito bundle support + Yellowstone gRPC streaming | ~$200 | $2,400 |
| Ethereum RPC (Phase 2) | Alchemy or Infura — premium Ethereum Virtual Machine (EVM) RPC with Flashbots Protect | ~$100 | $1,200 |
| Data feeds subtotal | $739 | $8,868 | |
| AI & Software | |||
| Claude API | Genesis, adversarial review, hypothesis generation | ~$150 | $1,800 |
| GPT-4o API | Adversarial auditor (S11B dual review) | ~$50 | $600 |
| Self-hosted AI (Year 2+) | Replaces API costs when monthly API spend exceeds GPU server cost. Hetzner GPU server ~€200–400/month running fine-tuned open-weights model. | $0 (Year 1) | $0 (Year 1) |
| Software licenses | Misc tools, monitoring, backup | ~$30 | $360 |
| AI & software subtotal | $230 | $2,760 | |
| Factory & Diagnostics (Build Phase) | |||
| Software Factory — AI verification | Claude (defence) + GPT-4o (prosecution) | 0–100/day | 00–,000 |
| GNDID — AI diagnostics | Stall analysis + health reports (S39) | .40/day | 76 |
| Horizon Scanning — SIHS | Search APIs + AI synthesis (S40) | ~0 | 60 |
| Rust Quality Pipeline | cargo audit, Miri, benchmarks — runs locally | /usr/bin/bash | /usr/bin/bash |
| Factory & diagnostics subtotal | ~20 | ~,440 | |
| Trading Costs (Variable) | |||
| Exchange fees (standard tier) | Maker 0.02% / Taker 0.04% on Binance. Blended ~0.03% at 60/40 maker/taker mix. | Variable | See below |
| Exchange fees (VIP tier) | After qualifying: Maker 0.016% / Taker 0.036%. Blended ~0.024%. | Variable | See below |
| Slippage | Estimated 1–3 bps per trade on crypto perpetuals at HIRO's position sizes | Variable | See below |
| Funding rate (net) | Net funding received minus paid. Can be positive (income) or negative (cost). Historically net positive for short-bias strategies. | Variable | See below |
Assumes: Month 1–2 no trading (Phase 0). Month 3–4 at $100K deployed. Month 5–6 scaling to $1M. Month 7–12 full extraction. Exchange fees at VIP tier from Month 4 onward (volume qualification). All figures in USD.
| Line Item | Bear (75%) | Base (190%) | Target (300%) | Bull (500%) |
|---|---|---|---|---|
| Revenue | ||||
| Gross trading P&L | $820,000 | $2,080,000 | $3,300,000 | $5,500,000 |
| Maker rebates earned | $8,000 | $18,000 | $28,000 | $45,000 |
| Net funding income | $5,000 | $12,000 | $20,000 | $30,000 |
| Total gross revenue | $833,000 | $2,110,000 | $3,348,000 | $5,575,000 |
| Trading Costs (Variable) | ||||
| Exchange fees (taker) | ($28,000) | ($65,000) | ($100,000) | ($165,000) |
| Exchange fees (maker, net of rebate) | ($4,000) | ($9,000) | ($14,000) | ($22,000) |
| Slippage cost | ($35,000) | ($80,000) | ($125,000) | ($200,000) |
| Funding cost (periods of net payment) | ($12,000) | ($25,000) | ($38,000) | ($55,000) |
| Total trading costs | ($79,000) | ($179,000) | ($277,000) | ($442,000) |
| Fixed Costs | ||||
| Infrastructure | ($4,320) | ($4,320) | ($4,320) | ($4,320) |
| Data feeds | ($5,268) | ($5,268) | ($5,268) | ($5,268) |
| AI & software | ($2,760) | ($2,760) | ($2,760) | ($2,760) |
| Total fixed costs | ($23,148) | ($23,148) | ($23,148) | ($23,148) |
| Summary | ||||
| Net profit (Year 1) | $730,852 | $1,907,852 | $3,047,852 | $5,109,852 |
| Net return on $1M | ~73% | ~191% | ~305% | ~511% |
| Trading cost as % of gross | 9.5% | 8.5% | 8.3% | 7.9% |
| Fixed cost as % of gross | 2.8% | 1.1% | 0.7% | 0.4% |
| Tier 1 year-end NAV | $1,000,000 | $1,000,000 | $1,000,000 | $1,000,000 |
| Tier 2 year-end balance | $730,852 | $1,907,852 | $3,047,852 | $5,109,852 |
| Total capital year-end | $1,730,852 | $2,907,852 | $4,047,852 | $6,109,852 |
Trading costs (exchange fees + slippage + funding) represent 8–10% of gross revenue across all scenarios. This is the dominant cost. Fixed costs (infrastructure + data + AI) represent only 0.2–1.5% of gross. The business has extreme operating leverage — nearly all incremental gross revenue drops to the bottom line. The most impactful cost optimisation is exchange fee tier negotiation and maker/taker mix improvement, not infrastructure savings.
Noether Capital operates as a private dark fund. No external capital. No investors. 100% profit retention. Growth is funded entirely from retained extraction. The projection below assumes Tier 1 engines operate at the Base Case (190% AR) after Year 1 — a conservative assumption that accounts for mechanism crowding, market structure evolution, and natural edge decay over time.
| End of Year | Tier 1 Engines | Tier 1 Capital | Tier 2 Capital | Total Capital | Annual Extraction | Cumulative Extraction |
|---|---|---|---|---|---|---|
| Year 1 | 1 | $1.0M | $2.0M | $3.0M | $2.0M | $2.0M |
| Year 2 | 2 | $2.0M | $4.8M | $6.8M | $3.8M | $5.8M |
| Year 3 | 3 | $3.0M | $9.5M | $12.5M | $5.7M | $11.5M |
| Year 4 | 4 | $4.0M | $15.1M | $19.1M | $7.6M + Tier 2 income | $19.1M+ |
| Year 5 | 4–5 | $4–5M | $20M+ | $25M+ | $8–10M+ combined | $27M+ |
Assumptions: Year 1 at Base Case (190% AR on $1M = ~$2M extraction, conservative below 300% target). Years 2+ at 190% AR per Tier 1 engine. New Tier 1 engine launched when Tier 2 accumulates sufficient capital ($750K+ surplus above reserve). Tier 2 active strategies begin Year 2 at 50% AR on deployed Tier 2 capital. Personal income withdrawals of $200K/year from Year 2 onward (deducted from Tier 2, not Tier 1). By Year 2, AI inference transitions from API to self-hosted open-weights models fine-tuned on HIRO operator history — eliminating external data exposure and reducing marginal AI cost to near zero.
Infrastructure scales with engine count: each additional Tier 1 engine requires one additional AX102 server (~€160/month) connected via vSwitch. By Year 2, if API spend exceeds ~$1,000/month, a dedicated GPU server (Hetzner GPU line, ~€200–400/month) replaces API calls with self-hosted fine-tuned inference, eliminating external data exposure and reducing marginal AI cost to near zero. Total infrastructure cost at full 4–5 engine scale: ~€1,200–1,500/month ($15K–$18K/year) — still negligible relative to extraction.
Each Tier 1 extraction engine requires its own ghost footprint — operating on a distinct symbol subset to avoid cross-engine coalescence. Engine 1 might run BTC/ETH/SOL. Engine 2 might run secondary crypto + equity micros. Engine 3 might run commodity + FX micros. By Year 3–4, the symbol universe is fully covered across multiple engines, and further Tier 1 scaling requires expanding into new instrument classes or new venues. Beyond 4–5 Tier 1 engines, additional capital is better deployed in higher-capacity Tier 2 strategies.
| Year | Estimated Personal Income | Source | Notes |
|---|---|---|---|
| Year 1 | $0 | — | All profits retained for compounding. Living expenses funded from existing savings. |
| Year 2 | $200K–$400K | Tier 2 withdrawal | First income year. Withdrawn from Tier 2 surplus above engine funding requirements. |
| Year 3 | $400K–$800K | Tier 2 withdrawal + Tier 2 yield | Tier 2 balance supports both engine funding and substantial personal income. |
| Year 4+ | $500K–$1M+ | Tier 2 income + capital gains | System self-sustaining. Personal income no longer constrains capital growth. |
By Year 3, the system is generating sufficient extraction that personal income withdrawals do not constrain Tier 1 or Tier 2 capital growth. The dark fund model means no management fee drag, no performance fee sharing, no fund administration costs — the operator keeps everything.
Noether Capital deliberately operates without external capital, external investors, or public visibility. This is not a temporary constraint — it is the permanent strategic posture. The reasons are structural:
The 300% target is achievable — not guaranteed, but structurally plausible — because of five properties that compound together:
First, the opportunity surface is real. Crypto perpetual markets generate thousands of structurally forced micro-dislocations daily. Funding resets, liquidation cascades, cross-venue lag, queue depletion events — these are not hypothetical. They are observable, measurable, and mechanically explained. The question is not whether the opportunities exist but whether HIRO can capture them efficiently.
Second, the capital is small relative to the opportunity. At $1M, HIRO's per-trade size is far below the detection threshold of institutional participants and far below the capacity ceiling of most micro-alpha opportunities. The system can enter and exit without moving the market, which preserves the full theoretical edge.
Third, the architecture is purpose-built for this extraction model. HIRO is not a general-purpose trading system adapted for micro-alpha. It is designed from the ground up for high operator count, fast cycling, mechanism binding, continuous discovery, aggressive kill, and dynamic capital routing. Every section of the Genesis Framework supports this operating model.
Fourth, the feedback loops are tight. P&L attribution (S35A) decomposes returns daily. Meta-productivity predictions (S24R) anticipate regime changes. Crowding detection (S24P, S24R, S28.5a) routes capital away from degrading edges. Negative primitive feedback (S33A) prevents repeated failures. The system learns from its own operation faster than most manual research processes.
Fifth, the cost structure is minimal. $23K annual fixed costs on a $1M capital base is 2.3%. Even at the bear case scenario (73% return), the firm is generating $731K on $23K of fixed costs. The four-server architecture — Osaka, Singapore, Helsinki, Frankfurt — puts HIRO within <5ms of every major extraction venue on the planet. Each server pays for itself if it captures just one additional opportunity per month that the previous architecture would have missed.
The 300% target is not a prediction. It is the design point that shapes every architectural decision — operator count, capital velocity, kill speed, discovery cadence, mechanism focus. If the system reaches 200%, it has built an exceptional business. If it reaches 150%, it has built a strong business. If it reaches 100%, it has built a good business. The target ensures the machine is built for maximum extraction. The gates ensure it adapts to whatever reality delivers.
Phase 1 operates on CEX perpetuals and micro futures PLUS flash loan extraction (oracle lag liquidations via the A1→A2 cross-domain pipeline). Flash loans are Phase 1 because they require zero capital at risk and the infrastructure — CEX price feeds, Solana RPC, flash loan contracts — is built during Phase 0. The remaining DeFi opportunities (CEX-DEX arb, DeFi perp funding, MEV bundles) and additional CEX opportunities (new listings, basis, options vol, stablecoin depeg) activate in Phase 2 (Month 5+) as the DeFi execution adapter matures.
All DeFi execution is handled by the Alpenglow DeFi Execution Module (S31C), deployed on Server A2 (Singapore). Alpenglow supports: flash loan routing across multiple providers, AMM swap execution with MEV-aware routing (Jito on Solana), oracle-dependent execution with real-time oracle lag monitoring, gas budget management (flash loan operators consume gas budget, not CFUs), and cross-chain execution across Solana and EVM chains. Oracle lag thresholds: Tier-1 oracles (Chainlink, Pyth) at 300ms, Tier-2 at 500ms — monitored by GNDID Layer 2.
| Opportunity | Category | Conservative Annual | Moderate Annual | Infrastructure Needed |
|---|---|---|---|---|
| Oracle lag exploitation | DeFi | $200K | $500K | Solana/EVM RPC, liquidation contracts |
| CEX-DEX arbitrage | DeFi | $100K | $400K | DEX aggregator integration |
| DeFi perp funding arb | DeFi | $100K | $300K | DeFi perp protocol APIs |
| MEV bundle extraction | DeFi | $100K | $300K | Jito Block Engine API |
| New listing exploitation | CEX | $50K | $200K | Exchange announcement monitor |
| Cross-exchange basis at scale | CEX | $100K | $300K | Multi-venue capital deployment |
| Options-implied vol exploitation | CEX | $50K | $200K | Deribit data feed |
| Stablecoin depeg exploitation | CEX/DeFi | $50K | $150K | Stablecoin price monitoring |
| Flash loan amplified extraction | DeFi (PHASE 1) | $500K | $2.0M | Aave/Balancer flash loan contracts, tx simulation — deployed Month 1–2, live Month 3 |
| Phase 2 total (additional DeFi + CEX, excl. flash loans) | $750K | $2.35M | ||
| Phase 1 (CEX + flash loans) extraction surface | $2.7M | $7.0M | ||
| Phase 1 + Phase 2 combined surface | $3.5M | $9.4M | ||
Flash loan extraction is Phase 1 — not deferred. It requires zero capital at risk, uses infrastructure already built for CEX trading (price feeds, server architecture), and adds $500K–$2M in annual extraction potential from Month 3. Combined with CEX extraction ($2.2M–$5.0M), Phase 1 alone covers a $2.7M–$7.0M opportunity surface. The $3M target is achievable from Phase 1 without any Phase 2 expansion. Phase 2 DeFi and additional CEX opportunities add $750K–$2.35M more, pushing the total combined surface to $3.5M–$9.4M.