git_commit string | phase string | robustness_items_completed list | optional_items_skipped list | headline dict | shared_system_prompt_body_caveat string | cost_summary_usd dict |
|---|---|---|---|---|---|---|
1916c1fbbdabcd9bfb03067193a902c57e99d957 | robustness_consolidation | [
"1_cleaner_negatives_recalibration",
"2_wrong_author_scaffold",
"3_30b_hallucination_audit",
"4_length_disqualifier_clean_rubric"
] | [
"5_sampler_noise_floor",
"6_extra_anchors"
] | {
"cleaner_negatives_strong_flag_rate": "0/18 = 0%",
"calibration_caveat": "30B-A3B raw-base prompt-induced topical paraphrase rate 8/77 (10%) is comparable to 30B-Ryan-SFT 9/77; absolute strong-flag interpretation on seg-17 cells is upper-bounded.",
"wrong_author_buck_paired_delta_substance_vs_RP": {
"mean":... | The wrong-author scaffold's system-prompt body is byte-identical to the rigorous Ryan scaffold body except for author attribution and the 2 exemplar excerpts. The body was selected by best-of-N optimization against Ryan style in segment 11 phase 0. Outcome B partly reflects shared Ryan-tuned register description, not p... | {
"cleaner_negatives_drafting": 0.4,
"paraphrastic_recall_v2_haiku_rerank": 1.2,
"wrong_author_sampling_tinker": 0.4,
"wrong_author_buck_judging": 5.5,
"wrong_author_evhub_judging": 5.5,
"total_phase_estimated": 13
} |
README.md exists but content is empty.
- Downloads last month
- 71