Methodology

Data Sources

Models Used

Each asset uses one or more quantitative models. Scores range from 0-100 where higher values indicate historically more favorable conditions.

Ciklora Confluence Label

Ciklora Confluence combines three independent data views into one research label: Aligned, Divergent, Neutral, or Low Confidence. The label is context for reading the data together, not a recommendation.

Low Confidence checks run first. When required data is fresh and passes quality gates, matching strong Cycle and Crowd directions show Aligned, opposing strong directions show Divergent, and weaker or mixed readings show Neutral.

• BTC: native BTC has no on-chain staking yield; wrapped-BTC lending variants are outside v1.
• ETH: Yield Context must pass before a visible label can render.
• SOL: visible labels require fresh Cycle, Crowd, and Yield inputs.

Macro Confluence adjusts model confidence using economic indicators. A positive alignment means macro conditions support the signal direction; a negative alignment means they oppose it. The adjustment is small (up to ±20% of base confidence) and is applied only to backtest-validated assets.

Enabled Assets & Indicators

How It Works

1. Compute z-score (value vs 2-year rolling mean) for each indicator.
2. Determine 3-month trend direction (rising / falling / flat) via linear fit.
3. Blend z-score and trend into a per-factor signal using asset-class rules.
4. Weighted sum of factor signals produces an alignment score (−1 to +1).
5. Confidence is adjusted: base × (1 + alignment × 0.20)

Factor weights were determined by walk-forward backtest with ablation testing on 70/30 train-test split. Only factors that improved score-outcome correlation were retained. Results are based on limited macro data history and should be treated as experimental.

Breakout Detection uses technical indicators validated via walk-forward backtest (70% train / 30% test, GO threshold: correlation > 0.05) on BTC data from 2014–2026.

Indicators & Timeframes

Production Strategy

• Daily signal: Volume Profile only (price vs. Value Area High/Low with volume confirmation)
• Weekly signal: S/R (55%) + RSI Divergence (45%)
• Combined:50% daily + 50% weekly, mapped to 0–100 score

Indicators are timeframe-specific based on empirical results. Forced combination across timeframes degrades performance. Currently enabled for BTC only.

On-chain metrics measure Bitcoin network fundamentals directly from the blockchain. Data is sourced from the CoinMetrics Community API (free, no rate limit). Metrics are normalized via 2-year rolling z-scores and combined into a 0–100 score using empirically validated weights from walk-forward backtest.

Metrics & Weights

Pi Cycle Top(Philip Swift, LookIntoBitcoin) is a price-derived BTC cycle-top indicator computed from the 111-day SMA vs the 350-day SMA × 2. Historical BTC cycle tops (2013, 2017, 2021) coincided with the 111×2 SMA crossing above the 350 SMA within ~3 days. The compute module is implemented in src/models/onchain/pi_cycle.pywith 15 unit tests; dashboard surfacing is live in the BTC on-chain panel. A 2026-05 validation pass kept Pi Cycle as overlay-only (weight = 0): the scored implementation did not provide enough usable cross-event samples for statistical promotion, so it remains a stand-alone cycle-state indicator shown for context.

How It Works

1. Fetch daily MVRV, active addresses, exchange flows, and hashrate from CoinMetrics.
2. Normalize each metric using a 730-day rolling z-score (clipped to ±3, mapped to −1 to +1).
3. Compute weighted composite: MVRV × 0.35 + Active Addresses × 0.65.
4. Map composite (−1 to +1) to score (0–100). Above 60 = bullish, below 40 = bearish.

Backtest Results

• Test period: Oct 2022 – Feb 2026 (1,222 samples)
• Composite correlation: r=0.089 (90d forward) — GO
• Monotonicity (90d): 0.75 (GOOD) — higher scores correlate with higher returns
• Score bins: 0–20 → −8%, 40–60 → +13%, 80–100 → +41% (90d forward)

Currently enabled for BTC only. Exchange netflow and reserves are shown for context but excluded from scoring due to negative backtest results. Ablation study confirmed removing exchange_reserves improves composite correlation from 0.024 to 0.071.

Sentiment analysis measures market crowd psychology using the Fear & Greed Index as a contrarian indicator. Extreme fear is interpreted as bullish (historically favorable buying conditions), while extreme greed is bearish. Binance Futures data (funding rate, open interest, long/short ratio) is shown for context but not scored.

Metrics & Weights

Contrarian Logic

1. Fear & Greed raw value (0–100) is mapped via contrarian formula: (50 − value) / 50
2. Extreme Fear (0) → +1.0 (bullish). Extreme Greed (100) → −1.0 (bearish).
3. Composite signal (−1 to +1) is mapped to a 0–100 score. Above 60 = bullish, below 40 = bearish.

Backtest Results

• Test period: Sep 2023 – Feb 2026 (883 samples, 70/30 walk-forward split)
• BTC Fear & Greed: r=0.116 (90d forward, p=0.001) — GO
• ETH Fear & Greed: r=0.063 (90d forward, p=0.077) — GO
• Funding Rate ablation: removing it improves BTC composite from 0.040 to 0.116

Enabled for production crypto assets (BTC, ETH, XRP, SOL, SUI). Backtested on BTC (r=0.116) and ETH (r=0.063). Altcoins use the same Fear & Greed signal since it measures overall crypto market sentiment. Binance Futures display-only metrics have limited history. Data sources: Alternative.me (Fear & Greed), Binance Futures API (funding rate, OI, ratios).

Microstructure analysis examines derivatives market positioning through liquidation data, open interest changes, and orderbook depth. Daily snapshots from Coinalyze provide raw data for informational display.

Metrics (All Display-Only)

BTC only. Data sources: Coinalyze free API (daily liquidation/OI history), Binance Futures API (orderbook depth snapshot). Data is collected daily for informational purposes only.

Macro Factors in AI Analysis

The AI report writer receives these indicators as context when generating daily reports and answering chat questions.

Individual mega-cap stocks (AAPL, MSFT, GOOGL, META, TSLA) use a four-factor composite combining macro and short-term price signals. Loukas cycle analysis on its own underperformed for these names in walk-forward backtesting, so each asset is scored against macro drivers that historically correlate with mega-cap equity returns.

Factors & Weights

Production Status

• AAPL— production ready, 5-year walk-forward backtest cleared
• MSFT, GOOGL, META— backtest cleared, full production migration in progress
• TSLA— elevated monitoring (correlation regime shift detected)
• AMZN, AMD— NO-GO (composite walk-forward negative)

Weights are tuned per stock via walk-forward backtest. Composite re-validates weekly; sustained correlation degradation triggers an elevated-monitoring alert. Source data: FRED (Pal/Alden factors), Yahoo Finance (stock + SP500 close).

The top-100 coin tracker pulls headlines hourly from a curated set of public RSS feeds (no paid API) and scores each headline for sentiment. The per-coin 7-day rolling average drives the bluechip_sentiment_dropalert: when sentiment flips positive→negative andthe 24-hour price change is below −5%, the alert fires.

Pipeline

1. Hourly registry refresh: pull top-100 coins by market cap (CoinGecko), filter stablecoins.
2. Hourly headline pull: fetch RSS feeds, de-duplicate by article URL.
3. Score each headline with an AI sentiment classifier (compound score in [−1, +1]). VADER lexicon (seeded with crypto/finance jargon) is the fallback when the AI classifier is unavailable.
4. Persist scored articles, compute rolling per-coin sentiment averages.
5. Alert engine compares current vs prior 7-day average to detect sentiment drops.

Empirical note: raw VADER misses domain jargon ("bullish", "breakout") as neutral — the fallback path seeds the lexicon with ~22 crypto/finance terms before scoring. Sentiment is provided for context only; it does not feed the 23-asset composite score directly.

What Is Not Included

• Some advanced on-chain metrics— SOPR, NUPL, Puell Multiple, and Reserve Risk have provider integrations (free tier of an external on-chain data provider, rate-limited to ~8 requests/hour) but are currently disabled pending validation. Pi Cycle Top is implemented and shown on BTC. MVRV and Active Addresses are scored into the composite; exchange flows are integrated as display-only.
• Social media sentiment— Paid social-volume providers are not tracked. RSS news headlines for the top-100 coin tracker are scored daily via an AI sentiment classifier with VADER as a fallback. The Fear & Greed Index is the sole scored metric in the crypto sentiment composite; futures funding rates are collected for BTC and ETH as display-only (ablation backtest showed they hurt composite accuracy).
• Whale tracking— Whale transactions and large wallet movements are out of scope. An empirical backtest (2026-04) of ETH large-transfer flows against forward returns found no statistically significant predictive signal. Orderbook depth and liquidation data are collected for BTC as display-only with no validated predictive signal.
• Geopolitical events— The system reacts to price effects, not to the events themselves.
• Price predictions— Scores reflect historical favorability, not future price targets.
• Real-time data— Most data is end-of-day. Major crypto (BTC, ETH, XRP, SOL, SUI) refreshes every 30 minutes via an intraday cron during the 01:00–23:30 UTC window. Indices, individual stocks, metals, FX, and on-chain metrics remain daily.