SightBTC AI: A Probabilistic 24-Hour Bitcoin Direction Intelligence Layer

Whitepaper v0.1

Date: 09 July 2026

Status: Draft for product/research positioning

Important notice

SightBTC AI is described in this document as a research and informational product concept. It is not financial, investment, trading, legal, or tax advice. It does not provide guaranteed outcomes, guaranteed profit, or risk-free trading signals. Cryptocurrency markets are volatile, and any model-generated directional signal can be incorrect. Users remain fully responsible for their own decisions.

Abstract

Bitcoin traders operating on short time horizons face a noisy decision environment shaped by volatility, fragmented market signals, derivatives behavior, sentiment shifts, and sudden liquidity events. Many existing signal services present directional calls with excessive certainty, limited context, and weak risk framing. SightBTC AI is proposed as a research-first intelligence layer for estimating Bitcoin's probable 24-hour directional bias using machine-learning classification, market-regime analysis, and confidence-based signal thresholds.

The system focuses on a narrow use case: classifying the next 24-hour BTC direction as UP, DOWN, or NEUTRAL / LOW CONFIDENCE. Rather than predicting exact prices or promising profitable trades, SightBTC AI translates multi-source market conditions into a structured signal containing direction, confidence score, time horizon, timestamp, market-regime note, and explicit risk disclaimer. The product is designed for directional Up/Down traders and crypto users seeking a disciplined second opinion before time-bound decisions.

This whitepaper defines the product thesis, intended users, signal structure, input categories, modeling approach, validation principles, delivery roadmap, and risk limitations. The central claim is not that short-term Bitcoin direction can be predicted with certainty, but that probabilistic framing, model discipline, and transparent uncertainty can improve how traders interpret noisy market conditions.

1. Executive summary

SightBTC AI is a proposed AI-assisted research product for estimating Bitcoin's likely direction over the next 24 hours. It is designed to support human decision-making, not replace it.

The public output is intentionally simple:

• UP — the model estimates BTC is more likely to close higher over the next 24 hours.
• DOWN — the model estimates BTC is more likely to close lower over the next 24 hours.
• NEUTRAL / LOW CONFIDENCE — current conditions do not justify a directional signal.

Each signal includes a confidence score, market-regime context, update timestamp, interpretation note, and risk warning. This structure is intended to reduce hype, avoid forced predictions, and help users compare their own view against a model-based directional estimate.

SightBTC AI is narrow by design. It does not attempt to cover every crypto asset, every timeframe, or every trading style. Its initial thesis is that a focused 24-hour BTC direction signal can be more understandable, measurable, and responsible than broad, vague market commentary.

The product's value depends on four principles:

1. Probabilities over promises — outputs should communicate uncertainty, not certainty.

2. Context over hype — confidence and market regime matter as much as direction.

3. Validation over storytelling — model performance must be tested against realistic baselines.

4. User responsibility over automation — the signal is a research input, not a trading command.

2. Market problem

Bitcoin is one of the most liquid crypto assets, but short-term BTC trading remains difficult because the market is noisy, reactive, and structurally fragmented. Traders often make decisions while watching multiple incomplete sources of information: price charts, volume, funding rates, open interest, news, macro events, social sentiment, influencer posts, and liquidation data.

The problem is not lack of information. The problem is lack of structured interpretation.

Short-horizon directional traders face several recurring issues:

• Information overload. More indicators do not automatically create a clearer decision. Traders may see bullish momentum, bearish funding, positive sentiment, and high volatility at the same time.
• Emotional decision-making. Fast BTC moves can create fear of missing out, panic exits, revenge trades, and overreaction to recent candles.
• Conflicting narratives. Social media and market commentary often explain price action after the fact, but provide weak forward-looking discipline.
• Binary calls without confidence. A simple "BTC up" or "BTC down" call is incomplete if it does not communicate uncertainty, invalidation risk, and market context.
• Overpromising signal services. Some products sell certainty where none exists. This can create false confidence and poor risk behavior.

A responsible directional research product should not pretend to remove risk. Instead, it should help users interpret uncertainty more consistently.

3. Product concept

SightBTC AI is a research-first signal layer that estimates Bitcoin's probable direction over the next 24 hours.

The product converts market inputs into a structured signal:

• Asset pair: BTC/USD or BTC/USDT
• Direction: UP, DOWN, or NEUTRAL
• Confidence score: probability-like model confidence measure
• Time horizon: next 24 hours
• Generated timestamp: when the signal was produced
• Market-regime note: short explanation of current conditions
• Interpretation note: plain-English reading of the signal
• Risk disclaimer: reminder that the signal can be wrong

The initial public experience can be delivered through a landing page, formatted whitepaper, mock signal examples, manual Founder Pass payment flow, and periodic research updates. A later validated version may include scheduled signal generation, a signal archive, confidence history, regime tracking, dashboard views, and alert preferences.

Non-goals

SightBTC AI is not:

• A trading bot
• An exchange
• A broker
• A portfolio manager
• A guaranteed-profit signal group
• A promise of exact price targets
• A substitute for risk management
• Financial advice

This boundary matters. The system is designed to inform directional thinking, not to execute trades or encourage blind following.

4. Target users and use cases

4.1 Primary users

Up/Down directional traders.

These users make time-bound decisions on whether BTC will move up or down within a defined period. They need fast interpretation, but not necessarily a long research report before every decision.

Crypto traders seeking a second opinion.

These users already monitor charts and market context. SightBTC AI gives them a model-assisted reference point that can confirm, challenge, or complicate their own view.

Quant-curious retail users.

These users are interested in AI and machine-learning market interpretation, but need outputs that are readable without advanced statistical background.

Researchers and product observers.

These users may follow the product to observe how AI-assisted market interpretation is framed, tested, and communicated.

4.2 Example use cases

• Checking model bias before a 24-hour directional trade
• Comparing personal chart analysis against an AI-assisted classification
• Observing when the model chooses NEUTRAL instead of forcing a call
• Tracking how confidence changes across different market regimes
• Reviewing historical signals for learning and research
• Using confidence and regime notes as prompts for deeper analysis

The system is most useful when treated as a structured second opinion. It is least useful when treated as an instruction to trade.

5. Signal design

SightBTC AI signals are intentionally compact. A user should be able to read the core output in seconds while still seeing the context required for responsible interpretation.

5.1 Signal classes

UP means the model estimates BTC has a stronger probability of closing higher over the next 24-hour window.

DOWN means the model estimates BTC has a stronger probability of closing lower over the next 24-hour window.

NEUTRAL / LOW CONFIDENCE means the model does not have enough confidence to issue a directional signal. This may happen when inputs conflict, volatility is abnormal, liquidity conditions are unstable, or estimated probabilities remain too close to baseline.

The NEUTRAL class is central to the product. A system that must always say UP or DOWN may look decisive, but it can become less trustworthy. In uncertain regimes, the most useful answer may be "unclear."

5.2 Example signal

BTC / USD — 24h Direction Signal

• Direction: UP
• Model confidence: 61%
• Time horizon: Next 24 hours
• Market regime: Volatile bullish compression
• Generated: 09 Jul 2026, 02:00 UTC
• Interpretation: The model currently estimates a moderate upward directional bias over the next 24 hours. Confidence is above the neutral threshold, but volatility remains elevated.
• Risk note: This signal is probabilistic and may be wrong. It should not be used as the only basis for a trade.

5.3 Confidence bands

A practical signal framework may map model confidence into simple user-facing bands:

• Low confidence: no directional signal; output NEUTRAL.
• Moderate confidence: directional signal with caution language.
• High confidence: stronger directional signal, still explicitly not guaranteed.

Confidence should not be presented as certainty. A 61% model confidence does not mean the outcome is safe, inevitable, or profitable. It means the model currently assigns more weight to one directional outcome than the other under its learned framework.

6. Data input categories

SightBTC AI's modeling concept relies on structured, timestamped market features. Input categories may evolve, but the initial research direction includes the following groups.

6.1 Price action

Price action features may include returns over multiple lookback windows, candle ranges, high-low structure, moving-average relationships, local trend direction, distance from recent highs/lows, and drawdown or rebound behavior.

6.2 Momentum

Momentum features may measure short-term acceleration, trend strength, reversal pressure, moving-average slope, relative strength, or cross-window momentum divergence.

6.3 Volatility

Volatility is critical for BTC. Features may include realized volatility, intraday range expansion, compression patterns, volatility breakouts, and abnormal movement flags. A directional signal during calm compression may have a different meaning than a signal during a high-volatility shock.

6.4 Volume and liquidity proxies

Volume behavior can help identify participation, exhaustion, and abnormal market activity. Liquidity proxies may include volume imbalance, spread-like measures where available, order-book-derived features where technically feasible, or exchange-level activity indicators.

6.5 Derivatives indicators

Where data is available and reliable, derivatives features may include funding rates, open interest, futures basis, options-derived information, and liquidation cluster proxies. These can help detect crowded positioning or leverage-driven risk.

6.6 Market structure

Market structure features may represent support/resistance proximity, breakout or breakdown attempts, range boundaries, trend continuation patterns, and failed moves.

6.7 Sentiment and news proxies

Sentiment-derived inputs may include social, news, or attention-based signals if they can be collected legally, reproducibly, and with acceptable data quality. These features should be treated carefully because they can be noisy, manipulated, or lagging.

6.8 Macro and context features

Bitcoin can react to broader risk sentiment, liquidity expectations, major economic events, and crypto-specific news. Macro/context features may include event flags or broad risk-on/risk-off proxies where they improve validation results.

6.9 Data principle

Inputs should be measurable, timestamped, reproducible, and auditable where possible. The model should avoid hidden, manually adjusted inputs that make performance difficult to verify.

7. Modeling approach

The initial SightBTC AI research approach should prioritize simple, testable baselines before complex black-box systems.

7.1 Prediction target

At each observation time, the model evaluates the next 24-hour BTC outcome. A basic label may classify whether BTC closes higher or lower after 24 hours. More advanced labeling may introduce a minimum move threshold so that very small changes are treated as neutral/noise rather than meaningful direction.

Example labeling approach:

• UP label: BTC return after 24 hours is above a positive threshold.
• DOWN label: BTC return after 24 hours is below a negative threshold.
• NEUTRAL label: BTC return remains within a defined low-movement band, or model confidence does not exceed the publication threshold.

The exact threshold should be selected through validation, not marketing preference.

7.2 Candidate algorithms

The model stack should remain practical and testable. The current research direction uses simple baselines first, then compares them against stronger machine-learning and sequential models.

• Logistic regression for an interpretable directional baseline. This is useful because it exposes whether simple OHLCV momentum, volatility, and moving-average relationships contain any measurable signal before heavier models are introduced.
• Gradient boosting / XGBoost-style classifiers for tabular market features. These models are strong candidates for OHLCV, volatility, momentum, liquidity-proxy, funding, and regime features because they can capture non-linear interactions without requiring a very large deep-learning dataset.
• Random forest / tree ensembles as robustness checks against overfitting and feature instability. These are useful for comparing whether signal quality comes from broad feature structure or only from a narrow boosted-model fit.
• LSTM / recurrent sequence models for temporal patterns across recent candles and feature windows. LSTM-style models may be tested after the baseline is stable, especially for short sequences where the order of returns, volatility compression, and breakouts matters.
• Temporal convolution or lightweight transformer-style sequence models may be evaluated later if the dataset expands enough to justify them. These should not be used as the first public model unless they outperform simpler baselines under walk-forward validation.
• Ensemble / meta-model layer to combine several model families into a single probability estimate. A meta-model can reduce reliance on one algorithm by weighting outputs from logistic regression, XGBoost-style models, sequence models, and transparent rule-based filters.

A simple model that performs consistently and can be explained may be preferable to a complex model that performs well only in a narrow backtest.

7.3 Current lightweight script

The current prototype includes a lightweight OHLCV-only script that produces three user-facing fields:

• Vector: UP, DOWN, or NEUTRAL
• Confidence: directional confidence percentage
• Reasoning: a short plain-English explanation based on momentum, volatility, and moving-average context

This prototype is intentionally modest. It is useful for testing the product format and user experience, not for making public performance claims.

7.4 Calibration and confidence

Raw model output should not automatically be treated as reliable probability. Where possible, confidence scores should be calibrated using standard techniques and evaluated across time. The product should track whether higher confidence bands actually correspond to better realized directional performance.

If calibration is weak, confidence language should be conservative.

7.5 Market-regime layer

Short-term model behavior can differ significantly depending on market regime. A model trained across all conditions may perform differently during trend continuation, chop, volatility compression, sudden liquidation cascades, or low-liquidity weekends.

Candidate regimes include:

• Trending up
• Trending down
• Range-bound
• Volatility compression
• Breakout attempt
• Breakdown risk
• High-volatility shock
• Low-liquidity / abnormal conditions

The regime layer has two purposes. First, it can support model performance analysis by showing where signals work or fail. Second, it improves user communication by explaining why a signal is strong, weak, or neutral.

8. Validation and backtesting principles

A directional research product must be validated carefully. Backtests can be misleading if they use future information, overfit parameters, ignore transaction realities, or report only favorable periods.

8.1 Validation goals

SightBTC AI validation should answer several practical questions:

• Does the model classify 24-hour BTC direction better than simple baselines?
• Does performance improve at higher confidence bands?
• How often does the model return NEUTRAL?
• Which regimes produce stronger or weaker results?
• Are errors concentrated during specific volatility or liquidity conditions?
• Does performance remain stable over time?

8.2 Candidate metrics

Useful metrics may include:

• Directional accuracy by class
• Precision and recall for UP and DOWN signals
• Performance by confidence band
• Performance by market regime
• False-positive and false-negative analysis
• Coverage rate: how often the model produces directional vs NEUTRAL output
• Calibration quality
• Stability across market cycles

For a research signal, coverage rate is especially important. A model that is accurate only because it issues very few signals may still be useful, but users should understand that behavior. A model that issues frequent signals with weak accuracy may be dangerous if marketed aggressively.

8.3 Anti-overfitting rules

Responsible validation should follow several rules:

• Use chronological train/test separation.
• Prefer walk-forward validation where possible.
• Avoid leakage from future candles, future news, revised data, or post-event indicators.
• Compare against simple baselines, such as naive trend-following or previous-period direction.
• Report unfavorable periods, not only best-case examples.
• Re-test after any major feature or threshold change.
• Treat backtest performance as historical evidence, not a future guarantee.

8.4 Baseline comparison

SightBTC AI should not be evaluated only against random chance. It should also be compared with practical baseline strategies and simple indicators. If a complex model cannot outperform a transparent baseline under realistic validation, the baseline may be the better product foundation.

8.5 Reporting limitations

Any public performance reporting should include a clear explanation of methodology, time period, sample size, confidence thresholds, and limitations. Backtest performance does not guarantee future results.

9. Product delivery model

SightBTC AI is currently framed as an active-development product with intermediate research results. The delivery model should move from controlled validation to public proof, then to beta testing and commercial rollout.

9.1 Current stage — Development / pre-beta

The current phase focuses on making the signal credible before aggressive public claims:

• Product brief
• Landing page copy
• Whitepaper
• Public update content package
• Mock signal examples
• Risk and disclaimer copy
• Historical BTC dataset
• Baseline classifiers
• Walk-forward validation framework
• Source cross-checks and confidence-calibration work

Founder Pass price: $12.99. This is a pre-launch reservation for future lifetime product access after launch; it does not provide immediate access while the product is still in development. It may also include a 30-day test-alert window: if testing produces one signal above 85% confidence, Founder Pass holders receive that single high-confidence alert as a practical opportunity to evaluate the signal and potentially offset the pass cost. Crypto payment may be supported. It should not be described as an investment, token, equity, profit-share, or financial product, and no cost recovery or profitable outcome is guaranteed.

9.2 Early Q4 2026 — Public results event

The next public milestone is a five-day public results event. For five consecutive days, SightBTC AI should publish BTC direction results publicly with direction, confidence, timestamp, interpretation, and risk notes.

The event goal is to demonstrate signal readability, transparency, operational cadence, and practical model behavior under real public observation. It should not be framed as a guarantee of future performance.

Post-event lifetime-access reservation price: $49.99.

9.3 Post-event — Full beta testing phase

After a successful public event launch, the product can move into broader beta testing:

• Regular signal production
• Signal archive
• Confidence and regime display
• Web delivery and optional alert channels
• Model-performance notes
• User feedback loops
• Continued validation against independent data sources
• Clear user-facing disclaimers

The beta target is to maintain >85% predictability in validated target conditions. This should be treated as a target and validation criterion, not as a guaranteed outcome.

9.4 Q1 2027 — Commercial rollout

After beta completion and final optimization, SightBTC AI can transition to subscription-based access:

• $24.99/month
• $199.99/year

Future additions may include web dashboard views, alert preferences, confidence history, regime tracking, model comparison, API access, and additional horizons or assets if validation supports expansion.

Expansion should remain evidence-driven. Adding more assets or timeframes before validating the core BTC 24-hour signal would increase complexity without necessarily improving product value.

10. Risk, limitations, and compliance posture

SightBTC AI should treat risk language as part of the product, not as a hidden footer.

10.1 Market risks

Bitcoin can move sharply due to news, macro events, liquidity shocks, exchange incidents, regulatory comments, large liquidations, and market manipulation. These events can invalidate any short-term signal.

10.2 Model risks

The model may be wrong because of overfitting, poor feature selection, regime change, noisy labels, data leakage, weak calibration, or insufficient historical representation of current conditions.

10.3 Data risks

Data sources can be delayed, incomplete, revised, inconsistent across exchanges, or temporarily unavailable. A model is only as reliable as the data pipeline supporting it.

10.4 User-behavior risks

Users may overtrust a signal, ignore position sizing, chase losses, use excessive leverage, or treat confidence as certainty. Product copy should actively avoid encouraging this behavior.

10.5 Compliance posture

The product should consistently state that it provides research and informational content only. It should not claim to provide individualized investment advice, guaranteed returns, automated execution, or risk-free opportunities.

Required disclaimer themes:

• Research and informational content only
• Not financial advice
• No guaranteed performance
• Signals can be wrong
• Users are responsible for their own decisions
• Cryptocurrency trading can result in significant losses

11. Roadmap

Now — Development / pre-beta

Continue model refinement, independent-source validation, confidence calibration, and responsible signal-publishing preparation. Founder Pass price: $12.99. This is a pre-launch reservation for future lifetime product access after launch; it does not provide immediate access while the product is still in development. It may also include a 30-day test-alert window with one high-confidence alert if testing produces a signal above 85% confidence. Crypto payment may be supported. It should not be described as an investment, token, equity, profit-share, or financial product, and no cost recovery or profitable outcome is guaranteed.

Early Q4 2026 — Five-day public results event

Publish BTC direction results publicly for five consecutive days, including timestamps, confidence, market regime, interpretation notes, and disclaimers. Post-event lifetime-access reservation price: $49.99.

Post-event — Full beta testing

Move into broader beta testing after successful event launch. Target: maintain >85% predictability in validated target conditions while continuing user feedback, source validation, and product hardening.

Q1 2027 — Commercial rollout

Transition to subscription-based access after beta completion and final optimization: $24.99/month or $199.99/year.

12. Conclusion

SightBTC AI proposes a narrow, disciplined approach to AI-assisted crypto market research: estimate the probable 24-hour direction of Bitcoin, communicate confidence clearly, and admit uncertainty when conditions are unclear.

The product's value does not depend on pretending that Bitcoin can be predicted with certainty. It depends on creating a consistent structure for interpreting noisy directional markets. For Up/Down traders and crypto users seeking a model-assisted second opinion, this structure can support more disciplined thinking while keeping responsibility and risk where they belong: with the user.

SightBTC AI should be built and communicated with restraint: validate before claiming, explain before selling, and use probability language instead of promises. That posture is not only safer; it is also the product's main differentiation from hype-driven signal groups.