Training & validation results
A static report of the results obtained during training and walk-forward validation of the EventHorizon Crypto model (BTC/USDT). No live feed, no external data dependency — every number here is embedded directly in this page.
Accuracy at best validation epoch
The amber marker is the 52% reference used in the supplied dashboard; the baseline definition should be checked against the source run configuration. Accuracy above this reference is descriptive. Statistical significance depends on the original test configuration, intervals and model-selection procedure.
| Horizon | Accuracy | vs. baseline (52%) |
|---|---|---|
| 5s | 60.5% | |
| 15s | 52.4% | |
| 30s | 48.0% | |
| 1min | 50.0% | |
| 5min | 50.2% | |
| 15min | 49.8% | |
| 30min | 50.5% | |
| 1h | 50.5% | |
| 5h | 54.0% | |
| 1d | — no data |
Arithmetic check: the eight displayed values from 5s to 1h average 51.5% (rounded to one decimal). The public project README separately reports 51.2% mean honest-validation accuracy for a multi-horizon model and 60.7% for an H0-focused model; the exact run mapping behind this page’s 50.7% / 60.5% figures was not supplied. The summary therefore keeps the displayed horizon scores but flags that run-level reconciliation is needed.
Validation pipeline
The supplied dashboard reports 60.5% at 5s. Public project summaries document several validation methods, but they do not provide enough run-level output to independently reproduce the exact result or assert that this selected checkpoint passed every test.
Walk-forward & embargo
The public README documents walk-forward analysis and an embargo between training and evaluation. Exact fold boundaries and gap length for the displayed run are not included here.
Block bootstrap
Block bootstrap is documented for uncertainty estimation. Block lengths, replicate count, interval endpoints and the run log are not exposed in this static page.
Permutation test
The documented pipeline includes permutation testing with shuffled targets and aligned masks. The null design, repetitions and p-value for the exact 60.5% result are not provided here.
Gap bootstrap
Public project documentation describes a paired gap bootstrap comparing model and baseline. The exact paired inputs and result for this dashboard’s run cannot be confirmed from the public summary.
Temporal stability
Temporal-stability analysis across partitions is described in the research README. The 15-day partition width and per-partition outputs shown in the old copy are not corroborated by that summary.
Method summary checked against the public research records: Crypto H0 Edge research repository and EventHorizon Crypto README. These summaries document methods and limitations but do not expose every run-level interval, p-value, block setting or per-tier PnL result.
Reported validation result, negative economic outcome
Promising validation signal; tradability not established
The supplied dashboard reports 60.5% accuracy at 5 seconds against its 52% reference (+8.5 percentage points). Public research summaries report 60.7% for an H0-focused model and 59.9% for H0 in a multi-horizon model. They also state that the validation set was used for checkpoint selection, not as a final independent test. The exact 60.5% run and its p-value / confidence interval remain unverified from the public summary.
The project README reports that the strategy was economically unviable under the tested maker-execution assumptions, and that maker-fee sensitivity was evaluated separately. The public summary does not establish the old page’s more specific claim that net PnL was negative across every Binance VIP fee tier; that tier-by-tier statement should be checked against the underlying result files before being restored.
These findings are historical and specific to the evaluated data, checkpoint and execution assumptions. Validation does not guarantee future performance, and a positive classification metric is not evidence of profitability.