fix: DecisionDistillPolicy.select() returns indices instead of strings when memory <= budget; add regression test

fix: import sys in harness/run.py and harness/clearml_report.py; add distill to CLI episode choices

docs: finalize H2-G results — gemma K=3 sweep 150/0 errors confirmed

docs: pre-register H2-G — cross-validation on gemma-generated episodes

docs: finalize H2-G results — gemma K=3 sweep 150/0 errors confirmed

docs: pre-register H2-G — cross-validation on gemma-generated episodes

docs: embed result figures into final report

docs: update README to final state (distill implemented, results summary, analysis scripts)

docs: restructure HYPOTHESIS.md as the final report

docs: full experiment results — H2 refuted, failure localized to selection

chore: switch MEMORY_BUDGET to 1 for second sweep pass (K=1 vs K=3 comparison)

fix: duplicated observation text in generator _step (text appended twice)

Deepseek