1. The Data Engine

DeFiStar.io does not rely on third-party APIs. We perform direct Remote Procedure Calls (RPC) to the blockchain nodes for over 9 networks. This ensures the yield and TVL you see are not cached or delayed.

Direct Contract Querying

Using the eth_call method, we query smart contracts directly to retrieve raw hexadecimal data, which our engine decodes into usable financial metrics. For example, to help calculate liquidity in Compound V3:

function getTokenSupply($rpcUrl, $tokenAddr) {
    $supplyHex = rpcCall($rpcUrl, 'eth_call', [
        ['to' => $tokenAddr, 'data' => '0x18160ddd'], 'latest'
    ]);
    return $supplyHex ? hexToDec($supplyHex) : '0';
}

2. Analytical Calculations

Once we have the raw "State" of the blockchain, our specialised modules perform the following calculations:

// Persistence Score Calculation (yield-persistence.php)
$volatilityPenalty = min(100, ($stdDev / $avgAPY) * 100);
$consistencyBonus = ($minAPY / $avgAPY) * 100;
$persistenceScore = max(0, min(100, $consistencyBonus - ($volatilityPenalty * 0.5)));

// Decay Rate (comparing first vs last third of period)
$decayRate = (($firstThirdAvg - $lastThirdAvg) / $firstThirdAvg) * 100;

3. The Risk Radar & Safety Score

Our safety.php and risk-radar.php modules work in tandem to audit protocol safety. The Safety Score (0-100) is a weighted calculation:

Penalties & Bonuses

Penalties are deducted from the base score based on the following tiers:

• Low TVL: −20 pts (under $500K) / −15 pts ($500K–$1M) / −8 pts ($1M–$10M).
• Low Utilisation (<5%): −15 pts. Utilisation 5–20% incurs −5 pts.
• High Utilisation (>80%): −10 pts. Above 90%: −20 pts. Morpho and Morpho Blue are exempt from this penalty.
• Extreme APY (>15%): −15 pts. APY 10–15% incurs a scaled penalty up to −10 pts, plus an additional −5 pts if TVL is under $5M.
• Zombie Pool (APY <1% and TVL <$1M): −20 pts.
• APY / Utilisation Mismatch (APY >8% with Utilisation <30%): −6 to −10 pts — yield is not backed by genuine borrower demand.
• Emerging Chain Risk: Applied when TVL is under $1M on BNB Chain (−15 pts), Linea (−10 pts), or Scroll (−5 pts).

Penalty Amplifiers: When a pool's base score is below 70, all penalties are multiplied by 1.5×. If the asset is also USDT, FRAX, or USDD an additional 1.1× multiplier applies (combined cap: 1.65×). Total penalties are then hard-capped at the lesser of 50 pts or 60% of the base score.

Bonuses are awarded for healthy pool behaviour: +10 pts for TVL exceeding $100M, +12 pts for utilisation in the 40–70% sweet spot, and +10 pts for APY in the stable 3–7% range.

Two-Tier Scoring System

DeFiStar uses two related but distinct score types:

• Aggregate Entity Scores (0–68) — shown on the Stablecoins, Protocols, and Blockchains ranking pages. These measure each entity's overall ecosystem strength: TVL dominance, adoption breadth, yield stability, track record, and chain maturity. Scores are capped at 68 because they serve as inputs to the pool-level calculator. The 32 points above 68 are reserved for pool-level bonuses.
• Per-Pool Safety Scores (0–100) — shown on individual pool, safety audit, and dashboard pages. These combine the three entity scores using the 40/40/20 weighting above, then apply pool-specific penalties and bonuses to produce a final composite score.

Entity aggregate scores (stablecoins, blockchains, protocols) are displayed on their native 0–68 scale — the true maximum for the foundational scoring model. Pool-specific bonuses add up to 32 additional points, bringing the maximum combined per-pool safety score to 100. This preserves the mathematical relationship: entity score (max 68) + pool bonuses (max 32) = pool score (max 100).

4. Historical Data Depth

The accuracy of our predictive models and trend analysis depends on historical data depth. Below is our current data archive per protocol:

5. Stablecoin & Oracle Verification

In stablecoin-analytics.php, we don't just show prices. We verify stablecoin health using Price Oracles directly on-chain. This ensures that de-pegging events are detected at the oracle level—the same level used by protocols for liquidations.

// Peg price fetched via Chainlink Oracle (stablecoin-analytics.php)
$pegPrices = $cronData['stablecoins'] ?? [];
$pegStatus = (abs(1 - $peg) > 0.01) ? 'WARNING' : 'HEALTHY';

// Market dominance calculation
$dominancePct = ($asset['total_tvl'] / $globalTvl) * 100;

6. Prediction Engine Methodology

DeFiStar operates four independent prediction engines and a signal confluence system. Each prediction is stored with a target verification date and automatically checked against actual outcomes. Our philosophy: fewer predictions, higher accuracy. Low-accuracy signal types are retired; high-accuracy signals receive stricter quality gates.

Accuracy Calculation

All accuracy figures across the platform use a consistent methodology:

• Only definitive outcomes count: Predictions are marked correct or incorrect. Legacy neutral outcomes are excluded from the accuracy denominator.
• Rolling windows: Every engine page displays 7-day, 30-day, and all-time accuracy to track performance trends.
• Per-type breakdown: Accuracy is reported per signal type (e.g., "Cooling Down: 49%", "Extreme Surge: 84%") for maximum transparency.

Verification Process

Verification runs automatically via verify-all.php (v4.0) on a 6-hourly cron schedule. Each engine uses verification logic that exactly matches its prediction engine:

• APY Momentum: Requires a meaningful magnitude move — 5% relative change OR 0.15 percentage points absolute, whichever is larger. Trivial fluctuations do not count.
• TVL Momentum: Stabilisation verified within a 5% base band (MAE-adjusted per pool). Honest MAE measurement — actual TVL movement is recorded even for correct predictions.
• Surge Predictor: Requires 3%+ APY increase from starting level. All outcomes are correct or incorrect — no ambiguous verdicts.
• Reliability: Checks whether current APY remains above the 5% benchmark (for "maintained" predictions) or has dropped below it (for "declining").

// APY Momentum: Magnitude-aware verification (V3.0 aligned)
$minMove = max($startApy * 0.05, 0.15);

if ($pred['prediction_type'] === 'cooling_down') {
    $isCorrect = ($startApy - $currentApy) >= $minMove;
}

// CRITICAL: Only correct or incorrect — NEVER neutral
$outcome = $isCorrect ? 'correct' : 'incorrect';