Decision-support infrastructure
Yieldothen runs entirely in the browser, so analysts and investors can operate from any location with a stable connection. Recommendations are generated from statistical modelling rather than intuition, and every output is logged for independent review.
Transparency layer
Every model run writes to a log that any account holder can audit. Entries below illustrate the format community members use to check our outputs against actual market behaviour over time.
| Timestamp (AEST) | Model | Signal Latency | Model Confidence | Outcome Validation |
|---|---|---|---|---|
| 09:14:02 | Volatility-Cluster v3 | 0.38s | 88.1% | Confirmed |
| 09:12:47 | Risk-Weighted Screen | 0.41s | 82.6% | Confirmed |
| 09:10:55 | Trend-Persistence v2 | 0.52s | 79.4% | Awaiting close |
| 09:08:31 | Volatility-Cluster v3 | 0.36s | 90.0% | Confirmed |
Core capabilities
The platform is designed around a single remote operator managing positions or recommendations without a trading desk behind them. Each module is scoped to reduce manual review time, not to replace judgment.
Each dataset is tested for statistical significance before a model is promoted to live status. The brief summarises sample size, confidence interval, and the conditions under which the model's output should be discounted.
Positions are screened against detected volatility clusters, flagging periods where historical correlations break down. The engine does not block action; it attaches a risk note to the recommendation instead.
Daily summaries are generated without manual compilation, formatted for either personal review or distribution to a small client base. Reports reference the same log entries visible in the public feed.
Methodology
The pipeline is deliberately linear. Each stage produces an artefact that can be inspected on its own, rather than a single opaque output.
Market, economic, and sentiment feeds are pulled on a fixed schedule and checked for gaps or duplication before anything is processed. Incomplete feeds are flagged rather than silently filled.
Cleaned data is run through the relevant model, which outputs a probability range rather than a single figure. The range is retained in the log alongside the final recommendation.
Before publication, an analyst reviews any output that falls outside the model's expected confidence band. This step exists to catch feed errors and context the model cannot see, not to override the statistics on preference.
Performance metrics
We publish model accuracy against a plain market benchmark rather than relying on descriptive claims. The table below shows the format used in the full dataset available to account holders.
| Model | Period | Model Accuracy | Benchmark | Verified |
|---|---|---|---|---|
| Volatility-Cluster v3 | Rolling 90d | 71.4% | Index baseline | 2024-11-02 |
| Risk-Weighted Screen | Rolling 90d | 68.9% | Index baseline | 2024-11-02 |
| Trend-Persistence v2 | Rolling 30d | 64.2% | Index baseline | 2024-11-01 |
Figures above illustrate the reporting format only. Current performance figures for each model are visible in full on the live log after account initialisation.
Access tiers
Each tier differs in API rate limits and the depth of historical data available, not in the accuracy of the underlying models. All tiers read from the same public log.
Yieldothen runs the same models and the same public log regardless of where you log in from. There is no office-bound feature set and no regional delay in log updates.