Tier-1/2 confirmatory analysis — dev tier1_taskvector_eb5d091bf5ed, confirmatory tier1_taskvector_confirmatory_532eb68bd5e2
| pair | task | eligible (own vector dev) | selected condition (dev acc) | conf: zero / few / own / selected / best control | selected − best control [95% CI] |
|---|---|---|---|---|---|
| qwen7b->qwen3b | antonym | yes (0.88) | replace_donor_fewshot | ridge | L25 (0.84) |
| qwen7b->qwen3b | capital | yes (0.82) | ridge | L14 | a1.0 (0.14) |
| qwen7b->qwen3b | morph | no (0.00) | procrustes | L2 | a1.0 (0.00) |
| qwen7b->qwen3b | addthree | no (0.00) | procrustes | L2 | a1.0 (0.00) |
| qwen7b->qwen3b | H6 partial (1 task(s)) | ||||
| qwen3b->qwen7b | antonym | yes (0.91) | replace_donor_fewshot | procrustes | L19 (0.84) |
| qwen3b->qwen7b | capital | yes (1.00) | ridge | L11 | a1.0 (0.91) |
| qwen3b->qwen7b | morph | no (0.03) | procrustes | L2 | a1.0 (0.00) |
| qwen3b->qwen7b | addthree | no (0.00) | procrustes | L2 | a1.0 (0.00) |
| qwen3b->qwen7b | H6 not supported (0 task(s)) | ||||
| qwen7b->llama3b | antonym | yes (0.84) | procrustes | L14 | a1.0 (0.38) |
| qwen7b->llama3b | capital | yes (1.00) | procrustes | L11 | a1.0 (0.05) |
| qwen7b->llama3b | morph | no (0.00) | procrustes | L2 | a1.0 (0.00) |
| qwen7b->llama3b | addthree | yes (0.77) | procrustes | L2 | a1.0 (0.00) |
| qwen7b->llama3b | H6 not supported (0 task(s)) |
Note: CIs are exact paired item bootstraps (2 000 draws) on per-item correctness stored in the confirmatory results.