fix: OpenMM via micromamba/conda-forge + worker phantom job guards

debug: add OpenMM verification steps + startup check so we can see what fails

fix: install OpenMM directly into system Python, remove conda/micromamba

fix: install OpenMM into conda env used by CMD so it's actually importable

fix: remove invalid has_anisou call from BioPython fallback

chore: remove daily rate limiting from MD, function prediction, and ADMET endpoints

fix: remove invalid CaPPD import from BioPython fallback

fix: OpenMM via micromamba + BioPython structural fallback, real error messages

feat: scientifically accurate MD simulation with OBC2 solvent, Kabsch RMSD, per-residue RMSF

fix: always start in-process worker, remove ENABLE_INPROCESS_WORKER gate

fix: logging crash, md/function_predict ligand_smiles column fallback

Fix MD simulation: OpenMM fallback, PDB ID propagation, better error handling

Phase 3a/4/5: ADMET descriptors (RDKit), MD simulation (OpenMM), function prediction (DeepFRI-inspired) — backend + frontend

Phase 2: structured JSON logging, request ID middleware, enhanced health endpoint, Sentry breadcrumbs

Phase 1: harden rate limits — JWT-aware keys, per-endpoint caps, daily limits on all job types

Phase 0c: offload large artifacts to Supabase Storage (docking, pipeline, sequencing)

Phase 0b: durable worker with FOR UPDATE SKIP LOCKED, per-type concurrency, separate container

Phase 0a: SSRF protection + auth on docking/sequencing, worker design doc

Fix _TRP_RINGS -> _TRP_RING_ATOMS name error

Docking viewer upgrade: Molstar GPU rendering, proper interaction geometry, 3D dashed-line rendering

Fix KeyError: payload['smiles'] -> payload.get('ligand_smiles')

Docking: minimal INSERT (only id/status/smiles), startup table check, migration SQL

Fix docking INSERT: only use columns guaranteed to exist in Supabase

Major docking rewrite: match frontend contract (pdb_id+smiles input, nested poses/interactions output, PDB fetch from RCSB, auto grid center)

Fix docking route mismatches: GET /status/{id}, frontend pdb URL prefix

Fix docking route: POST /api/docking/run to match frontend

Fix docking: add pdb_to_pdbqt_receptor() for protein PDBQT conversion

Fix unbounded _SESSION_EVENT_COUNTS dict leak (cap at 1000)

Use Dockerfile-installed Vina first, fall back to download

Fix Settings attribute names: SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY

fix: undefined query_length variable in _run_blast

fix: Linux Vina binary, Open Babel PDBQT, pipeline_worker utcnow

test: reduce vina timeout to 10s for faster diagnostic

fix: indentation error at line 610

fix: revert to async subprocess (no thread pool to avoid import hang), keep --cpu 1

fix: use ThreadPoolExecutor with subprocess.Popen (no asyncio subprocess), add --cpu 1 and stdin=/dev/null

fix: use asyncio.create_subprocess_exec with file handles + asyncio.wait_for (no threads, no pipes)

fix: use poll() polling loop instead of proc.wait(timeout=N) to avoid Python 3.14 pidfd-wait hang

fix: use Popen + wait(timeout=N) + killpg layered kill approach

fix: use /usr/bin/timeout OS-level command instead of Python subprocess timeout

fix: use subprocess.DEVNULL for Vina output — eliminates ALL I/O blocking

fix: use dedicated ThreadPoolExecutor(max_workers=1) + subprocess.run(file handles, timeout=N)

fix: use asyncio.create_subprocess_exec with file descriptors + wait_for — no threads, no pipe buffer deadlock

fix: add vimontest timeout diagnostic with sleep test

fix: set deployed commit hash

fix: update commit hash in vimontest

chore: add deployed commit hash to vimontest for debugging

fix: use file handles instead of PIPE to avoid Vina pipe-buffer deadlock on Render

test: vimontest runs Vina with 10s timeout to diagnose hangs

fix: use process group kill (killpg) for reliable Vina timeout