Spaces:
Sleeping
Sleeping
Commit History
-
docs: shorten hf summary
5706fdf -
docs: refresh winner readme
75907ae -
fix(frontend): ship prebuilt app bundle
98b73e0 -
fix(deploy): build frontend in node stage
cd52852 -
feat(frontend): add SQL command center UI
f86ef5b -
fix: set app_port=8000 in HF Space frontmatter (HF proxy defaults to 7860)
0376765 -
Merge HF Space initial commit into winning-env
bc10764 -
chore: merge HF Space scaffolding (frontmatter + .gitattributes)
b9659fa -
initial commit
Verified signature32d0d02 -
fix(phase2): strip all floats from [STEP] lines + sterile stdout wrapper
Canary findings #11/#12 show the validator's parser scans ALL stdout for floating-point numbers and rejects any outside the open interval (0, 1). Defense in depth: 1. _SterileStdout wrapper — installed before any import that could print, drops any line not starting with [START]/[STEP]/[END]. Stray library prints can no longer leak a float onto stdout. 2. [STEP] lines now text-only: 'task=X action=submit_query result=Y'. No reward, cumulative, rows, step count — nothing parseable as a float. 3. clamp_score hardened: after rounding to .4f, re-check and clamp again. Prevents 0.00004 -> 0.0000 and 0.99995 -> 1.0000 edge cases. 4. sql_env.grader.strict_clamp got the same post-round guard. Added two regression tests for the round-to-endpoint edge cases. stdout float scan (bogus key path): only 0.0500 and 0.3000 appear, both strictly in (0, 1). Exit 0.0bc9d26 -
feat: SQL Repair OpenEnv submission — Phase 1 validator passes
- 3 SQL repair tasks (easy/medium/hard) with SQLite-backed env - FastAPI server with all required endpoints (/health /tasks /reset /step /grader /baseline) - /reset accepts empty body (Phase 1 requirement) - inference.py: HTTP client + OpenAI-compatible LLM caller - Strict (0,1) score clamping with NaN/inf -> 0.5 fallback - Every task emits exactly one [START]/[END] even on crash (Phase 2 lesson) - Sterile stdout: only bracket lines on stdout, diagnostics on stderr - pyproject.toml + uv.lock + server/app.py:main + openenv-core>=0.2.0 - openenv validate .: PASS - 8/8 unit tests pass269f632