proxy_taillift__PXR__T0.80__C0.25__INTERNAL

Rule. PXR — proxy ramp-carrier: same two-sided test as AB2, but the per-step cap is applied on the PROXY axes for emotions that are not directly rampable
Source. trajectories_v5lift.parquet  |  Family. the tier owner's manifest tiers -- the exact subsets used for training
Manifest tier. proxy_taillift__PXR__T0.80__C0.25__INTERNAL — population 0 chains (0 h). SHAREABLE variant: 0.
Filter. rule=='PXR' and T==0.2 and C==0.25 and dataset in ['mls', 'eurospeech', 'emolia', 'podcast', 'snippets', 'evasnippets'] and qmax>=0.8 and cmax<=0.25
This tier was resampled. The manifest's earlier published filter applied the one-sided test to every tier regardless of its actual rule. Tier populations were always counted under the correct rule, so the sizes quoted here were never wrong — but the earlier selection drew from a contaminated pool, of which 100.0 % were not members of this tier. Under the corrected filter — which tests qmax >= T and cmax <= C, both computed per row under this tier's own rule — the tier turns out to contain nothing at all, which is why nothing is shown below.

This tier is empty — there is nothing to listen to. This tier is EMPTY: under its own two-sided rule no chain in the mined set reaches T=0.80 while keeping every proxy-axis step within 0.25. The 93 chains the earlier (incorrect) filter matched were selected by the one-sided test and were never members of this tier.

It is kept in the index rather than deleted so the ladder stays honest: a rung that nothing reaches is a real and useful result, and an earlier version of this page showed 20 samples here that were never members of the tier.