What is validated.
What is experimental.

Tidava is a market-research workspace, not a prediction guarantee. This page documents our validation status, data limitations and the live prediction ledger so claims can be checked.

No performance claim is made. Several visual panels are experimental or descriptive. Do not treat a screen, score or confidence percentage as a probability of profit.

Live out-of-sample ledger

No graded predictions yet. The ledger is collecting closed-bar predictions; accuracy is not claimed until outcomes exist.

Validation status

Category	Examples	Status
Backtested physics models	Prism: Hurst, RMT, MST, Rough Vol, Hawkes	Backtested, not statistically significant Stored daily series 2010–2026; permutation tests p≈0.83–1.00. Useful as risk context, not validated alpha.
Deterministic server models	Calendar Edge, Composite Signal, Mean Reversion (OU)	In-sample only Formula is reproducible; historical statistics are in-sample. No validated directional edge.
Experimental client models	Transformer Forecast, Neural Edge, Probability Field, HMM/regime panels	Experimental heuristic Fixed hand-tuned weights or heuristic state machines; not trained ML and confidence percentages are not calibrated.
Descriptive panels	Order Flow, Vol surfaces, Price Replay, Wavelet, Entropy, Phase Space	Descriptive Visualize market structure, realized volatility or flow; no validated directional prediction.

Data sources and limitations

US equities: Alpaca. The default free feed is IEX-only and is not consolidated market volume; historical data may be limited. A paid SIP/consolidated feed requires a separate subscription/licence.
Crypto: Binance public market data. Venue fragmentation applies; it is not a consolidated global price.
FX/metals/indices: broker or Yahoo feeds. They may be delayed and are not necessarily redistributable.
Daily research panels: Yahoo Finance daily adjusted closes, re-downloaded on demand; not point-in-time stored history.

How to read the panels

Descriptive panels answer “what is happening?”. Experimental panels answer “what might happen next?” with no validated edge. Only the prediction ledger reports outcomes, and only after the forecast horizon closes.

Reproducibility rules

Signals are calculated on closed bars only; live forming bars must not repaint predictions.
Predictions are timestamped with model, symbol, timeframe, bar close and horizon.
Accuracy is reported only on graded predictions, with sample counts.
No costs, slippage or funding are included unless explicitly stated.
Backtests overlap forward windows; effective sample size is smaller than raw signal counts.