TERMINAL / GUIDE
Labs model workspace
Read the current outputs of company-built models in one workspace. Compare observed effects, horizons, source freshness and statistical support.
12 / 14Labs model workspace
Interactive interface · ArcDelta dataOpening the workspace…
Start here
- 01Select an asset, venue and forecast horizon. Add models from the company catalog.
- 02Read current conditions and compare conditional frequency with baseline, uncertainty and sample support.
- 03Open horizons, events or data to inspect the selected output. More models automatically reduce chart density.
Inputs
Real ArcDelta history and current observations, plus a separately maintained evidence-review registry.
Running Labs models requires Labs access. This guide is public.
General limitations
Historical samples and simulated outcomes do not establish a price-direction forecast. Small samples, overlapping events and changing regimes limit inference.
- Calculation
- The conditional rate divides observed successful outcomes after X by evaluable X events. The baseline uses eligible outcomes in the comparison period. The effect is conditional rate minus baseline in percentage points.
Rate(Y | X) = successful outcomes / evaluable X events; effect = Rate(Y | X) − baseline - Reading & controls
- Read hits/events, baseline, difference and the 95% interval together. Horizons 5m, 15m, 1h and 4h are separate comparisons.
- Limitations
- Observed frequency is not a calibrated live probability. A wide interval or few events limits precision; viewed history is not untouched holdout data.
- Calculation
- Three distinct models use relative bar range, Wilder ATR relative to close, or root sum of squared log returns over a trailing hour. They compare elevated current readings with later volatility outcomes.
- Reading & controls
- Compare the three rows and horizons. For ATR, also read the matched range/volume control when available.
- Limitations
- Historical samples and simulated outcomes do not establish a price-direction forecast. Small samples, overlapping events and changing regimes limit inference.
- Calculation
- Convergence studies the absolute log price-ratio deviation from its trailing median. Decorrelation studies the trailing correlation of aligned five-minute returns and its later state.
- Reading & controls
- Select two different venues and inspect common timestamp coverage. Interpret the outcome definition of each model separately.
- Limitations
- Convergence is an observed price relationship. It does not demonstrate executable arbitrage after costs.
- Calculation
- The imbalance model studies signed aggressive flow relative to total executed volume. Absorption response adds price-body and opposing-close conditions; it is distinct from Footprint's per-row absorption marker.
- Reading & controls
- Inspect later movement frequency and the matched range/volume control. Open Events to see which closed observations qualify.
- Limitations
- Extreme flow is not an established directional signal. Model and Footprint marks use different units and event definitions.
- Calculation
- Input freshness is checked separately from statistical evidence. Reviews inspect later data, coverage, sample support, uncertainty and comparison controls; the public report contains aggregated results.
- Reading & controls
- Read reviewed/next-review dates. Research, supported, inconclusive, weakened, unsupported, insufficient_data and review_due describe evidence states. A new data epoch requires sample support to be reassessed.
- Limitations
- Fresh inputs do not make a model statistically supported. A due date does not automatically extend evidence validity.