A student-facing model for crypto analysis
Caja Inversora AI applies stochastic modeling to price and volume data, then publishes the resulting signals before their outcomes are known. Students can read the reasoning, not just the result.
Every signal is timestamped and locked into the public record before markets move, so the log cannot be edited with hindsight.
Methodology
The system is built on a small number of ideas from quantitative finance, translated into plain language for students without a statistics background.
Crypto prices move in ways that are partly random and partly patterned. Stochastic modeling treats that randomness explicitly, describing not one predicted price but a range of plausible outcomes and how likely each one is.
Predictive variance is a measure of how wide that range of outcomes is. A narrow range means the model is relatively confident; a wide one means conditions are unstable and any position should be sized accordingly.
An asymmetric setup is one where the potential loss on a position is deliberately kept smaller than the potential gain. The model does not try to predict direction perfectly; it tries to structure entries so that being wrong costs less than being right earns.
Order-book, volume, and price history are pulled from multiple exchanges and normalized to a common time frame.
The raw series is converted into indicators the model can compare across assets, such as volatility clustering and volume shifts.
The model generates a distribution of likely outcomes rather than a single price target, and estimates the variance around it.
A position size and entry range are calculated so that the downside stays limited relative to the modeled upside.
Community-verified results
Signals are recorded before their outcome is known and reviewed afterwards by other users on the platform. Losing signals stay in the log alongside winning ones.
| Date logged | Model signal | Realized outcome | Variance band | Verification |
|---|---|---|---|---|
| Illustrative row | Reduced exposure, asset A | Within modeled range | Narrow | Confirmed |
| Illustrative row | Increased exposure, asset B | Outside modeled range | Wide | Under review |
| Illustrative row | No position, asset C | Not applicable | Wide | Confirmed |
Illustrative layout of the historical performance chart. Once a public batch of signals has matured, this view is populated with the corresponding realized variance for each entry.
A signal is marked "Confirmed" once at least two independent community reviewers have checked the logged entry, timestamp, and outcome against the raw exchange data, and found no discrepancy.
Each entry carries a timestamp, the input data snapshot used at the time, and the resulting signal. Later edits are not possible; corrections are appended as new, separately dated entries.
Risk management
Capital constraints change what "reasonable risk" means. The model accounts for that directly rather than assuming a large, diversified portfolio.
Instead of maximizing expected return, the default configuration minimizes how much a single misjudged signal can cost. Position sizing is tied to the modeled variance: when the range of likely outcomes widens, suggested exposure shrinks automatically rather than staying fixed.
This does not remove risk from crypto markets, which remain volatile by nature. It changes where that risk sits, keeping any single entry small enough that a wrong call does not require an outsized win to recover from.
Allocation profile
Conservative mode caps suggested position size and widens the variance threshold required before a signal is logged. It is the default for new student accounts.
Collective review
The platform's central claim, that its logs are accurate, is checked by the people using it rather than taken on faith.
A signal is generated by the model and written to the public log with its timestamp and variance band before the market outcome is known.
Once the outcome is available, any registered user can compare the logged entry against raw exchange data and flag inconsistencies.
Entries with no flags after independent review are marked confirmed; disputed entries stay visible with their review status attached.
Student access
Beta access is aimed at university students who want to read a model's reasoning before deciding whether to follow it.
No payment is requested at this stage. Access is granted on a rolling basis to enrolled students.