// EVE TRADING INTELLIGENCE

SIX LAYERS OF
MARKET INTELLIGENCE.

eveOracle ingests live EVE market data every hour, runs ensemble price forecasts, detects market regimes, generates a ranked strategy, and produces a daily command brief — all without touching your wallet. The adaptive layer selects V1/V2/V3 strategy engines based on detected market conditions and tracks decision quality through ISK-based regret calculation.

Launch App Learn More
45
Database Tables
39
Service Modules
8
Sub-Systems
30+
Tracked Items
// HOW IT WORKS

Six layers. One daily brief.

Raw ESI data enters at the top. A prioritised action list exits at the bottom — every hour.

01
Market Ingest
The ingest layer pulls best buy/sell snapshots every hour from ESI across 30+ item types in all four major trade hub regions. Snapshots stored with timestamp; Z-score anomaly detection flags items that deviate >2.5 standard deviations from recent history.
02
Forecast
An ensemble forecaster combines Moving Average, Exponential Smoothing, and Linear Trend models to generate 1-day, 3-day, and 7-day price forecasts with per-model confidence bounds. Trend direction (STABLE / NOISY / VOLATILE) is scored per item.
03
Regime Detection
The system classifies the current market regime per item — STABLE_TREND, VOLATILE, or NOISY — based on price variance and trend consistency. The detected regime drives strategy engine selection in the next step.
04
Strategy Engine
Command selects V1 (aggressive, STABLE_TREND), V2 (conservative, VOLATILE), V3 (quality-gated, NOISY), or HYBRID mode. Capital is split across arbitrage, manufacturing, reserve, and opportunity buckets. Trading preferences apply hard filters before final ranking.
05
Daily Brief
The Execution layer generates a dated command summary with a priority-ordered action list, a task queue with Complete/Skip workflow, and time-sensitive alerts. Each trade is scored on 7 factors: profit potential, confidence, signal strength, urgency, route efficiency, affordability, and complexity.
06
Outcome Evaluation
Every decision is linked back to actual price movement. The Analytics layer calculates forecast accuracy per model (precision, recall, MAE), realized ROI per strategy version, and ISK-based regret — the difference between recommended and optimal actions.
// ORACLE

Market Data & Forecasts

The Oracle layer ingests best buy/sell snapshots every hour across 30+ item types in all four major trade hub regions. An ensemble forecaster (Moving Average + Exponential Smoothing + Linear Trend) generates 1/3/7-day price forecasts with confidence bounds. Z-score anomaly detection flags statistical outliers. Trend analysis scores direction, strength, and momentum.

meltuc.tech/eo/oracle
Item Region Best Buy Best Sell Trend 7d Forecast Confidence
Tritanium JITA 5.21 5.24 ↑ STABLE 5.31 82%
PLEX JITA 52.1M 52.4M → NOISY 51.8M 64%
Mexallon AMARR 82.4 83.1 ↑ STABLE 84.7 77%
Isogen DODIXIE 118.0 119.5 ↓ VOLATILE 114.2 51%
Morphite RENS 9,420 9,510 → NOISY 9,380 68%
▲ Anomalies detected: 2 Z-score > 2.5 threshold · Last ingestion: 4 min ago
// COMMAND

Strategy & Capital Allocation

The Command layer selects the optimal trading strategy engine based on the current market regime: V1 (aggressive, STABLE_TREND), V2 (conservative, VOLATILE), V3 (quality-gated, NOISY), or HYBRID. Capital is allocated across arbitrage, manufacturing, reserve, and opportunity buckets. Trading preferences apply hard filters and soft ranking to personalize recommendations.

meltuc.tech/eo/command
Active Strategy Engine
V3 — NOISY MARKET
Capital Allocation
Arbitrage
25%
25%
Manufacturing
40%
40%
Reserve
20%
20%
Opportunity
15%
15%
Top Recommendations
Action Item Est. Profit Score
BUY Mexallon +18.2M 82%
ARBITRAGE Tritanium +9.4M 71%
HOLD Morphite +4.1M 58%
// EXECUTION

Daily Brief & Task Queue

The daily brief consolidates the top priority actions, key market opportunities, capital recommendations, and risk signals into a single command summary. Tasks are extracted into a priority queue with complete/skip workflow. Alerts surface time-sensitive events. The action prioritizer scores each potential trade on 7 factors: profit potential, confidence, signal strength, urgency, route efficiency, affordability, and complexity.

meltuc.tech/eo/execution
DAILY BRIEF — Apr 15, 2026
CAUTIOUS
Priority Actions
1 Buy Mexallon in AMARR — ensemble forecast shows +2.7% over 3 days with 82% confidence +18.2M ISK
2 Arbitrage Tritanium JITA→AMARR — spread 0.6 ISK/unit at current volume = 9.4M net +9.4M ISK
3 Monitor PLEX JITA — NOISY regime active, hold position until trend resolves above 52.5M HOLD
Task Queue
Execute Mexallon buy order — 500k units @ 83.0 max Complete Skip
Load Tritanium (1.8M units) for JITA→AMARR route Complete Skip
Check Isogen DODIXIE volatility — anomaly flag active Complete Skip
Review reserve bucket — target 20%, currently 18.4% Complete Skip
// ANALYTICS

Accuracy & Regret Tracking

Every strategy decision is linked to actual price movements. The analytics layer calculates forecast accuracy per model, realized ROI per strategy version, and ISK-based regret — the difference between what was recommended and what would have been optimal. Calibration records track how well confidence scores match observed outcomes.

meltuc.tech/eo/analytics
78.4%
Forecast Accuracy
14.2%
Avg ROI
2.1M
7d Regret (ISK)
142
Trades Evaluated
Model Accuracy Comparison
Model Precision Recall MAE
V1 — Aggressive 71.2% 68.4% 0.41
V2 — Conservative 74.8% 72.1% 0.36
V3 — Quality-Gated ACTIVE 78.4% 76.9% 0.29
Ensemble (MA+ES+LT) 80.1% 77.5% 0.27
// OPERATOR

Rollout Control & Shadow Comparison

The Operator layer provides full control over the adaptive intelligence rollout. Four phases: SHADOW_ONLY (observe without risk), LIMITED_ADAPTIVE (adaptive with V3 fallback), FULL_ADAPTIVE (full autonomy), SAFE_FALLBACK (emergency revert). Eight threshold checks auto-trigger rollback on regret spikes, data freshness failures, or mode divergence. Shadow comparisons run all strategy modes in parallel to measure quality without affecting live decisions.

meltuc.tech/eo/operator
Current Phase:
SHADOW_ONLY
SHADOW_ONLY → LIMITED_ADAPTIVE → FULL_ADAPTIVE → SAFE_FALLBACK
8 Threshold Checks
✅ Regret Spike 2.1M < 5M
✅ Data Freshness 4 min < 90 min
✅ Forecast Accuracy 78.4% > 60%
✅ Mode Divergence Low
✅ Confidence Floor 68% > 55%
⚡ ROI Streak 3 trades (warn)
✅ Anomaly Rate 2 items < 5
✅ Shadow Parity V3 leads
Shadow Comparison: V1 vs V3 vs V2 (this week)
V1
+62M
V2
+74M
V3
+91M ▲
V3 outperforming by +17M ISK vs nearest shadow mode. All threshold checks passing. Safe to advance phase.
// CAPABILITIES

Six Layers of Market Intelligence.

eveOracle is not a single tool — it is a layered intelligence system. Each sub-system feeds the next, from raw market ingest through to a daily command brief you can act on immediately.

📥
Market Ingest
Hourly ESI ingestion of best buy/sell snapshots across 30+ item types in all four major trade hub regions. Z-score anomaly detection flags statistical price outliers (threshold: >2.5 sigma) automatically — anomaly count surfaced on the Oracle dashboard.
🔭
Ensemble Forecaster
Three models combined — Moving Average, Exponential Smoothing, and Linear Trend — generate 1/3/7-day price forecasts with individual and ensemble confidence bounds per item per hub. Forecast accuracy tracked via MAE per model.
⚙
Adaptive Strategy Engine
V1 (aggressive, STABLE_TREND), V2 (conservative, VOLATILE), V3 (quality-gated, NOISY), and HYBRID mode. Capital split across arbitrage, manufacturing, reserve, and opportunity buckets. Trading preferences apply hard filters and soft ranking before final recommendation output.
📅
Daily Command Brief
Dated brief with prioritised action list, task queue with Complete/Skip workflow, and time-sensitive alerts. Each trade scored on 7 factors: profit potential, confidence, signal strength, urgency, route efficiency, affordability, and complexity. Generates once per daily cycle.
📊
Outcome & Regret Tracking
Every decision linked to actual price movement. Analytics tracks precision, recall, and MAE per model version (V1/V2/V3/Ensemble). ISK-based regret measures the gap between recommended and optimal action. Calibration records track how confidence scores match observed outcomes over time.
🚦
Rollout Control (Operator)
Four phases: SHADOW_ONLY, LIMITED_ADAPTIVE, FULL_ADAPTIVE, SAFE_FALLBACK. Eight threshold checks auto-trigger rollback on regret spikes, data freshness failures (>90min), or mode divergence. Shadow comparisons run V1/V2/V3 in parallel to measure quality without affecting live decisions.
🚚
Inter-Hub Hauling Scan
On-demand cross-region scan finds buy-low/sell-high routes between trade hubs and ranks them by profit-per-m³, returning full route detail. Persisted inter-hub opportunities are browsable and CSV-exportable alongside same-region arbitrage.
🔌
Agent & MCP Access
Three Oracle capabilities are exposed as MCP tools so AI assistants can pull intelligence programmatically: market forecasts by type, ranked manufacturing/arbitrage opportunities, and inter-hub hauling scans — the same data the in-app screens use.
// GET STARTED

Trade Smarter. Not Harder.

Six layers. One oracle. Live market intelligence, adaptive strategy selection, and ISK-based decision tracking — all in your browser.

Launch eveOracle

Requires a MelTuc account. Create one free.