Unverified50% confidenceFactExact time
Timestral still outperforms the original gradient boosting model in the corrected backtest: MASE 0.66 versus 0.71, but the improvement is far smaller than the initially reported 0.49.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/10/2026
First Seen
Valid until: 2/8/2027
Sources
Related Entities
Related Claims
UnverifiedTimestral's MASE on GIFT-Eval is 0.658, close to the engineer's corrected 0.66, while the original leaked result of 0.49 deviates markedly.78% similarUnverifiedAfter re-running with archived forecast data, the MASE was 0.66, only a slight improvement over the original 0.71.77% similarUnverifiedCompared to the deterministic TimeKAN model, PGP-Clinical-TimeKAN reduced MAE by 0.52%.66% similarUnverifiedIn this system, short-term return features (5-minute −0.031, 15-minute −0.021) exhibit negative IC, reflecting a short-term reversal effect, with IC gradually approaching zero and turning positive as the time horizon extends.65% similarUnverifiedThe recommended starting alpha value is 0.10; a stronger alpha value does not mean better results, the degree of variation differs across seeds, and this method is not a universal improvement.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1156102API
curl https://kongchang.com/api/v1/knowledge/claims/1156102MCP
get_claim(id=1156102)