Spaces:
Running
Running
| title: "Repro - Learning Unanimously Acceptable Lotteries" | |
| emoji: 🎯 | |
| colorFrom: yellow | |
| colorTo: red | |
| sdk: static | |
| pinned: false | |
| tags: | |
| - trackio | |
| - trackio-logbook | |
| - open-experiment | |
| - icml2026-repro | |
| - paper-daiccpXZfU | |
| - theory | |
| - membership-queries | |
| # Repro - Learning Unanimously Acceptable Lotteries | |
| Current claim-by-claim verification starts at [Current verification](#/current), with the strongest result shown first in the [illustrated report](#/current-report). The [final release report](#/release-report) gives the claim-by-claim forecast, compute accounting, and remaining judge risk. All five exact contracts are scientifically **VERIFIED** in the cumulative candidate; publication and any score change remain blocked until release checks and live judging. | |
| The fixed command is `uv sync --frozen && uv run --frozen python repro/src/verify.py && uv run --frozen python repro/src/publication_gate.py`. All research compute ran on Hugging Face `cpu-upgrade` with no GPU. A [self-contained marimo tutorial](https://huggingface.co/spaces/DineshAI/daiccpXZfU/blob/main/notebooks/reproduction.py) embeds the conclusions without requiring expensive reruns. | |
| The previous judged artifact is preserved under [Historical rejected baseline](#/historical-rejected-baseline). Its pages remain unchanged and are not the current verifier. | |