v0.3.2

New Features

• HPC sliding-window submission (isce_base.py) — each step now runs a lightweight sbatch manager job that keeps ≤max_concurrent_hpc child jobs active at all times, refilling immediately on completion. Replaces the old batch-sequential approach. Consecutive steps with equal command counts (e.g. run_13–run_16) are merged into a single group-manager. Steps are chained via --dependency=afterok.
• Per-command elapsed time in sbatch logs (isce_base.py) — sbatch scripts print START/DONE/FAIL with elapsed seconds. Group tasks also print total elapsed across all grouped steps.

Bug Fixes

• HyP3 file paths (hyp3_base.py) — hyp3_jobs.json now saves to workdir root (was workdir/hyp3/); downloaded ZIPs go to workdir/hyp3/; retry job files save to workdir root; legacy out_dir=workdir entries auto-migrate to workdir/hyp3/.
• watch command ignores --interval flag (cli/main.py) — _proc_local_watch read args.refresh_interval but argparse stores it as args.interval; interval was always 60 s regardless of user input. Fixed.
• insarhub processor refresh/download re-processes retry job files (cli/main.py) — _find_job_files globbed hyp3*.json, matching hyp3_retry_jobs_<ts>.json files from past runs alongside hyp3_jobs.json. On refresh/download, stale retry files were loaded as separate processors, causing duplicate downloads or incorrect status. Fixed: retry files excluded from glob.
• Orbit files downloaded twice (cli/main.py) — -d -O flags triggered two orbit downloads. Fixed: skips explicit download_orbit() call when downloader already handled it.
• retry() runs locally after HPC submission (isce_base.py, cli/main.py) — hpc_mode excluded from saved config, so retry defaulted to local. Fixed: retry() auto-detects HPC from job metadata (slurm_job_ids/hpc_manager/hpc_array) and writes it back to config.hpc_mode so _step_executor() routes correctly. _load_local_processor also restores max_concurrent_hpc and HPC config fields from insarhub_config.json.
• Manager job killed mid-run leaves step stuck PENDING (isce_base.py) — elif n_cmds > 0 and not job_ids: was dead code (always False inside if job_ids: block); SLURM-killed managers with incomplete commands never resolved to FAILED. Fixed: condition is now elif n_cmds > 0:.
• _parse_time_s mis-parses 2-part SLURM time strings (isce_base.py) — "30:00" was treated as 30 h 0 min (108,000 s) instead of 30 min 0 s (1,800 s), overestimating group-manager walltime 60×. Fixed: 2-part strings now parsed as MM:SS per SLURM spec.
• Group-manager step stuck PENDING when job gone from SLURM (isce_base.py) — if group_task_dir was absent from saved metadata (old jobs), n_cmds resolved to 0, preventing the SUCCEEDED/FAILED transition. Fixed: n_cmds now stored in job metadata at submission; refresh uses it directly with file-count as fallback.
• refresh() shows only one RUNNING command in manager mode (isce_base.py) — with sliding-window, multiple commands run concurrently but only one showed RUNNING. Fixed: per-command status now derived from .done/.fail files; all in-flight commands show RUNNING.
• Hyp3_SBAS MintPy output in workdir root (mintpy_base.py) — base class run() passed self.workdir to TimeSeriesAnalysis; Hyp3_SBAS inherits without override, so all MintPy outputs scattered to workdir root instead of workdir/mintpy/. Fixed: uses self.mintpy_dir; same correction for _geocode_diagnostic_files.
• ISCE cleanup misses merged/interferograms (isce_sbas.py) — cleanup targeted isce/interferograms/ (nonexistent); real stackSentinel output is isce/merged/interferograms/. Large intermediate files were never deleted.
• HyP3 auth failure submits to wrong user (hyp3_base.py) — when re-auth failed for a pool user, credits=0 was overwritten by self.client.check_credits() on the previous user's client; jobs were then submitted under the wrong account. Fixed: credits check guarded by auth result.
• self.batchs updated per-loop-iteration (hyp3_base.py) — on multi-user refresh, if any user failed, their batch was silently dropped from self.batchs. Fixed: assignment moved after loop.
• Missing filename key in HyP3 file metadata crashes download (hyp3_base.py) — direct dict subscript raised KeyError for auxiliary entries lacking filename. Fixed: file_meta.get('filename') with skip on empty.
• delete_job_folder blocks on ~-prefixed workdir (settings.py) — Path(workdir) without expanduser().resolve() made relative_to() always raise ValueError, returning 403 on every delete. Fixed.
• _run_folder_select_pairs uses folder.parent as workdir (folders.py) — downloader config received parent directory instead of job folder; sub-paths written one level up, potentially colliding with sibling jobs. Fixed: workdir=folder.
• Analyzer stop_event leaks on step error (routes/analyzer.py) — early return on step exception bypassed _stop_events.pop(job_id), leaking events indefinitely. Fixed: pop before return.
• Refresh overwrites .insarhub_cache.json filenames with empty list (routes/processor.py) — if no jobs had SUCCEEDED yet, filenames=[] overwrote a valid cache from a prior successful refresh. Fixed: preserves existing filenames when current refresh yields none.
• Retry job files appear as selectable job entries in GUI (routes/processor.py) — hyp3*.json glob matched hyp3_retry_jobs_<ts>.json; selecting one for refresh returned only the retry batch status. Fixed: retry files excluded.

GUI

• Hyp3_S1 max_workers in settings panel (defaultconfig.py) — parallel download threads now configurable via Job settings group (default 4, range 1–16).

Refactor

• Centralized path layout (config/paths.py) — Hyp3Paths, ISCEPaths, MintPyPaths dataclasses replace all hardcoded workdir / "subdir" strings across hyp3_base.py, isce_base.py, isce_s1.py, mintpy_base.py, hyp3_sbas.py.
• Path dataclass coverage extended — remaining hardcoded path literals replaced with dataclass properties: isce_sbas.py now uses ISCEPaths for isce_dir, slc_dir, dem_dir; cli/main.py uses Hyp3Paths/ISCEPaths in _has_zips and _find_job_file; utils/batch.py and utils/tool.py use Hyp3Paths.output_dir for ZIP discovery.
• Hyp3Processor renamed to CloudProcessor (core/base.py) — ABC renamed to reflect generic cloud-backend semantics rather than HyP3 specificity. Updated across core/__init__.py, __init__.py, processor/hyp3_base.py, commands/processor.py, cli/main.py, core/engine.py.

Docs

• Contributing guide — new tab in MkDocs navigation (EN + ZH). Split into Overview, Backend, and Frontend pages.
• Backend contributing guide — architecture overview, path conventions, per-section instructions for adding new processors/downloaders/analyzers. Each section includes a "Adding a New Base X" subsection (with code examples for CloudProcessor/LocalProcessor, BaseDownloader, BaseAnalyzer) and an "Extending an Existing Base X" subsection with switch tabs (Hyp3Base, ISCE_Base, ASF_Base_Downloader, Mintpy_SBAS_Base_Analyzer).
• Frontend contributing guide — conda Node.js install, uvicorn backend startup from InSARHub root, module reference tables grouped by area (Entry & Global, Map, Search & Scene Selection, Jobs & Results, Settings, Utilities), backend communication pattern, settings panel, Vite proxy, build output, code style.

---