Changelog#

Changelog#

All notable changes to SYMFLUENCE are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]#

Added#

  • The SYMFLUENCE agent interface (symfluence agent): hands off to an installed coding-agent CLI (Claude Code, Codex, Gemini, …) primed as the SYMFLUENCE agent through four provider-agnostic layers, each wired through whatever mechanism the CLI declares in the launcher registry: (1) the packaged skills; (2) an identity block with live project context (detected configs, domain directories) and operating house rules, delivered via the CLI’s system-prompt flag or the AGENTS.md preamble; (3) a dependency-free MCP server (symfluence agent mcp); (4) packaged specialist subagents (calibration-debugger, platform-scout).

  • Agent modes (symfluence agent model / symfluence agent code): two first-class session modes, each a frozen ModeProfile (agent/modes.py) that shapes priming end to end. Modelling primes the operational skills (explore-platform, run-workflow-locally, debug-calibration), both subagents, modelling house rules (never edit platform source; drive everything through the workflow/MCP tools; report in hydrological terms), and a headless tool allowlist for the native chat. Coding keeps the full skill set and the host CLI’s own permissions. Bare symfluence agent opens the TUI agent screen; agent doctor diagnoses the setup per mode (with --json); agent mcp --mode serves one mode’s tool profile; skills materialize per-mode and copy whole skill directories.

  • MCP background jobs and platform-inspection tools: the agent MCP server exposes 15 tools. start_workflow_job / get_job_status / cancel_job / list_jobs run workflow executions as detached background jobs (a stdlib wrapper records the exit code durably; records and logs live in the agent cache; cancellation signals the process group), so a chat session can start an hours-long calibration without blocking the MCP connection. read_run_log, list_domains, calibration_status, get_results_summary, get_plot_paths, and compare_experiments answer the modelling questions (log tails, domain inventory, calibration progress, headline KGE/NSE metrics, figures, cross-experiment ranking) by reading the domain_* tree directly with stdlib CSV parsing. update_config is the one writer: a line-preserving edit of the user’s experiment YAML that must pass typed SymfluenceConfig validation and backs the original up first.

  • Agent home screen with a suspend round-trip: a minimal Agent home in the TUI (mode 7) — two mode cards (Model / Code) over two dim context lines (detected config with a cycle key, runtime readiness); diagnostics behind a d details modal. Starting a coding session round-trips: the TUI suspends, the primed coding-agent CLI runs full-screen in the same terminal, and the home screen returns when it exits. Terminals that cannot suspend fall back to the classic AgentHandoff exec after the TUI exits (--direct, a one-shot prompt, no TTY, or a missing TUI extra hand off immediately).

  • Native modelling chat: symfluence agent model in the TUI opens a SYMFLUENCE-native chat screen driving headless Claude Code over its stream-JSON protocol — one bounded subprocess per turn, resumed via Claude Code’s own session store (agent/headless.py: tolerant NDJSON parser → typed events; session ids persisted per project+mode). Assistant prose streams into the conversation; tool invocations render as compact expandable cards (never raw JSON); a run sidebar polls the domain tree and background-job records independently of the agent, so an hours-long calibration keeps ticking on screen. esc interrupts a turn, ctrl+e exports the conversation as Markdown, and each turn’s footer shows duration, cost, and the running session total. Codex/Gemini (or a failed stream) fall back to the suspend round-trip with modelling priming. Stdlib-only driver; no new dependencies.

  • Interactive permission approvals in the modelling chat: tools outside the modelling allowlist route through a hidden approve_action permission-prompt bridge (agent/approvals.py) — the MCP server blocks, the chat pops an allow/deny modal, and no reply is a denial. Outside the chat the same tools stay hard-denied. ADR-0004 rewritten for the current architecture (human-in-the-loop at the layer that executes actions).

  • MESH multi-GRU preprocessing fixes and elevation-band precip correction: GRU-dependent hydrology parameters (ZSNL/ZPLS/ZPLG/…) expand to one value per GRU so multi-GRU domains initialize; new forcing keys MESH_PRECIP_LAPSE_RATE (orographic gradient for elevation-banded forcing) and MESH_PRECIP_MULTIPLIER (uniform gauge-undercatch correction), both defaulting to no-op; SKIP_WARM_START documented as defaulting to true so calibrations are reproducible regardless of machine history.

Changed#

  • Logging-protocol overhaul (ADR-0005 enforcement, from an audit of the paper-run log corpus where WARNING+ERROR lines outnumbered INFO 1.85:1):

    • One shared protocol module (core/logging_utils.py): canonical file format (~46% smaller lines: TIME LEVEL [logger] message, filename symfluence_{domain}_{experiment_id}_{ts}.log), log_once de-duplicated emission, a single third-party suppression table, and one get_worker_logger bootstrap replacing four bespoke worker setups.

    • Milestone-level INFO everywhere: per-item/per-chunk loops now emit one summary line with per-item detail at DEBUG (an ERA5 96-chunk resume drops from ~190 INFO lines to 3); hot-loop conditions that produced tens of thousands of repeated WARNING/ERROR lines emit once and demote to DEBUG; external-model stdout goes to sidecar files referenced by one log line.

    • Comparable calibration progress: every optimizer emits the fixed schema {ALG} {i}/{max} {unit} ({pct}%) | Best | Improved | Crashes | Elapsed with explicit units (evals/gens/epochs/loops) and [P##] worker tags; GLUE and NSGA-II now stream progress.

    • Honest run reporting: ✗ Failed completions on failed steps, failure-aware workflow end block, run_summary.json schema v2 with error/warning totals counted from actual log records, per-step status+duration aligned with the run manifest.

    • Global --quiet/-q flag (console WARNING+; file log unaffected); --debug unchanged and dominant.

    • setup_logging is idempotent per (domain, experiment id) — re-use no longer re-opens a new log file per facade construction.

  • Agent verb consolidation: agent launch is deprecated (alias for agent code, to be removed after one release); the already-deprecated agent start/agent run and the agent list/agent skills verbs are removed (agent doctor covers the last two).

Fixed#

  • Stale plugins no longer vanish silently: a plugin whose register() fails on SYMFLUENCE API drift (e.g. importing a removed alias) now logs a WARNING naming the incompatible package instead of disappearing from the registry at DEBUG level; a worker-contract test sweep (AST-resolved method calls on every in-repo worker class) guards against the API-drift class of calibration-voiding AttributeErrors observed in the paper-run logs.

  • Review follow-ups to the agent interface and CLI option handling (#300, #299): --dry-run binary <tool> now previews instead of executing; global option normalization no longer steals host-CLI flags from agent-session pass-through (structure-aware boundary at REMAINDER subcommands, option names derived from a single spec); the MCP server survives non-object JSON-RPC input and bounds subprocess output via a temp file; the agent screen forwards -- pass-through args and validates the preselected runtime; priming failures degrade with visible warnings and the launch card only claims layers that actually activated (a SYMFLUENCE-generated AGENTS.md is refreshed, a user-authored one is reported as blocking injection); frontmatter parsing consolidated into resources.parse_frontmatter.

[0.9.2] - 2026-07-07#

Archival release accompanying the SYMFLUENCE paper series (“Out-standing in Every Field”, “The Registry as Social Contract”, “From Configuration to Prediction”; Eythorsson et al., 2026, Water Resources Research, submitted). First release archived on Zenodo via the GitHub integration.

Added#

  • Citation metadata: CITATION.cff (GitHub “Cite this repository”) and .zenodo.json (Zenodo deposit metadata) with the full author list and ORCIDs (#269).

  • Guard test validating every shipped paper calibration config against the strict config schema.

Changed#

  • Paper supplementary configs trimmed to the final manuscript (#268): removed the four cut experiments (decision ensemble, sensitivity analysis, large-sample and large-domain Iceland calibration), reduced the forcing ensemble to the four reported products (ERA5, RDRS, AORC, CONUS404), and kept only the 130 fixed-seed (42) calibration combinations the paper reports; all READMEs updated to the manuscript’s section and figure numbering.

Fixed#

  • Shipped calibration configs for the five JAX-native models (HBV, HEC-HMS, SAC-SMA, TOPMODEL, Xinanjiang) failed strict validation due to stale keys from older plugin schemas (initial_params; HBV also smoothing, smoothing_factor) and could not be loaded; the stale keys are removed (85 configs).

  • Paper-config seed guard repointed after the #268 trim removed the config it loaded (#270).

[0.9.1] - 2026-06-14#

Note: Versions 0.8.0–0.9.0 were tagged and published without CHANGELOG entries. This section documents the changes since the v0.9.0 release; the 0.8.x/0.9.0 history is not reconstructed here.

Added#

  • Canonical forcing contract (CFIF): a single canonical forcing reader keyed to CF standard names (CFIF) that model preprocessors consume through one path, with automatic resampling of the source to each model’s cadence. Adopted by SWAT, Wflow, IGNACIO, LISFLOOD, CWatM, PCR-GLOBWB, RHESSys, CLM-ParFlow, and mHM.

  • Model-ready data store for attributes and observations: new attributes reader plus observation/streamflow loaders that source from the model-ready store. SUMMA soil/land class and HYPE GeoData (soil/land/elevation) now read from the store; StreamflowMetrics and the remaining calibration loaders are routed through it.

  • Windows binary builds for many models — WATFLOOD (bundled Waterloo CHARM prebuilt), ParFlow, CLM-ParFlow, PRMS, VIC, PIHM, mHM, WRF-Hydro, and glacier — via MinGW/MSYS2 POSIX shims and CMake/NetCDF wiring.

  • WATFLOOD: full Unix and Windows source builds; CHARM runs on the datastore; parallel-calibration plumbing.

  • GSFLOW: datastore-driven inputs and a distributed-groundwater option.

  • mHM: CARRA forcing support and distributed-morphology (MPR) mode.

  • WATFLOOD and GSFLOW standalone postprocessors registered.

Changed#

  • Agent interface simplified to a launcher. symfluence agent launch now hands off to an installed coding-agent CLI (Claude Code, Codex, Gemini, …), primed with the SYMFLUENCE skills, instead of running a bespoke in-house LLM agent. It detects the CLI on PATH and uses whichever API key is set (ANTHROPIC_API_KEY / OPENAI_API_KEY / GEMINI_API_KEY); override with SYMFLUENCE_AGENT_CLI, skip skill materialization with SYMFLUENCE_NO_SKILLS. The skills now ship inside the package. (net ~ −7,000 LOC)

  • Config schema canonicalization (toward 1.0): promoted data-handler, HYPE/NGEN/FUSE/GR/mizuRoute/path, multi-gauge, optimizer/evaluator, and parameter-bounds flat keys to typed Pydantic fields; RECOGNIZED_FLAT_KEYS is now closed; the state and data_assimilation sections are wired into the flat-key transform; IGNACIO/GNN/LSTM schemas reconciled with the registered ones. Strict config-key validation is enforced and dogfooded on shipped configs.

  • Documentation: root ARCHITECTURE.md at-a-glance, an ADR practice with the initial ADR set, and a threat model (docs/security.md).

  • Evaluation: unified catchment-area resolution between trial and final-eval metrics (resolves the FUSE final-eval KGE divergence).

  • Performance: RTI review item 20 hot-spots — rasterio hoist, Julian-date vectorization, coastal-partition rewrite — plus a CLI performance gate.

  • Core layering restored: no core → upper-layer imports (RTI item 19).

Fixed#

  • Data acquisition: MERIT-Basins region table and dead-host download; Daymet gridded OPeNDAP route (descending y, DAP2); HRRR LCC-grid windowing; CERRA longwave variable name; NEX-GDDP-CMIP6 synthetic-airpres survival and hardening; DEM merge pixel-grid snap; MSWEP remote path and default version; SMHI skip-if-raw-exists guard; EM-Earth S3 handler (403 visibility, credentials, month window); WSC GeoMet paging corruption and NLDAS-2 v2.0 variable names; USGS NWIS gauge-local → UTC timestamps; GRACE atomic .part downloads.

  • HYPE: clone over https:// with retry, replacing the deprecated git:// SourceForge transport that intermittently broke nightly and release builds.

  • CLM build: NetCDF library symlinks for PIO (Debian multiarch), git safe.directory re-assert after HOME change, mpi-serial pin, and Perl XML::LibXML install.

  • Hardcoded *3600 hourly-conversion bugs in the Wflow and IGNACIO forcing paths.

  • Windows builds: WATFLOOD DLL staging, PIHM timegm guard, ParFlow mkdir/path compatibility, PRMS NetCDF path quoting and gnu17, VIC version macros, mHM NetCDF detection, and WRF-Hydro checkout.

  • Glacier: earthaccess auth for NSIDC, RGI region-name transform, and coldstart fixes.

  • PRMS: resilience-handler tracebacks (exc_info=True) and HRU parameters driven from the datastore.

  • SWAT preprocessor registered via core.registries.R to avoid a circular import.

Removed#

  • Deprecated register* decorator registry API — registration is now R-facade only. (breaking)

  • Deprecated *Postprocessor spelling aliases — use *PostProcessor. (breaking; external plugins must update)

  • HYDROGEOSPHERE model stub. (breaking)

  • The CalMIP example stub.

  • DPE (differentiable PE) and other dead/external config keys; shipped configs curated accordingly.

  • The bespoke in-house LLM agent (symfluence.agent internals) and the optional [llm]/openai dependency — replaced by the launcher above. symfluence agent start/run are deprecated aliases for launch. (breaking: the agent now requires an external coding-agent CLI)

CI / Release#

  • release-binaries: build MODFLOW 6 from source on Linux, build ngen on Linux/macOS, skip CLM on Windows (ESMF/CIME on MinGW unsupported), close presence-check holes and surface failed optional builds, and don’t clobber the latest release on a tagless workflow_dispatch.

  • Strict config-key validation dogfooded on shipped configs (kept out of the lint job, which has no symfluence install).

  • Require the PostProcessor-spelling plugin releases.

[0.7.1] - 2026-03-01#

Fixed#

  • HPC file-locking workaround: Added tempfile fallback (attempt 4) to ERA5 NetCDF writer for parallel filesystems (Lustre/GPFS/BeeGFS) where HDF5 fcntl() locking fails even with HDF5_USE_FILE_LOCKING=FALSE. Writes to $TMPDIR first, then moves the file to the target path.

  • mizuRoute Makefile build: Collapse multi-line LIBNETCDF before appending RPATH flags to avoid breaking backslash continuation (caused “missing separator” errors on CI).

  • ngen linker (expat): Detect system UDUNITS2 via pkg-config/ldconfig (not just explicit env var) so -lexpat is added for XML symbol resolution.

  • Wflow calibration: Fix snow parameters, routing, and unit conversion issues. Add Oudin PET, log-transform bounds, and sub-daily resampling support.

  • FUSE preprocessing: Make streamflow observations optional.

  • MPI launcher: Add fallback when preferred launcher fails; prefer mpirun over srun for launcher detection.

Added#

  • SPDX license headers on all source files.

  • Coverage tracking in cross-platform CI workflow.

  • Unit tests for NGEN, JFUSE, HYPE model runners and configs.

  • Unit tests for core modules (file_utils, validation, path_resolver).

  • Unit tests for _safe_to_netcdf fallback chain.

  • WATFLOOD: Expand calibration to 16 parameters with tests.

  • Bow River preset and default DEM_SOURCE to Copernicus.

  • Config-hash stage invalidation for workflow orchestrator.

[0.7.0] - 2026-02-10#

Note: This release jumps from the last tagged release v0.5.2 directly to v0.7.0. Versions 0.5.3–0.6.0 were development milestones on the develop branch and were never tagged or published to main. Their changes are included in this release and documented below in the [0.6.0] and [0.5.x] sections for historical reference.

Breaking Change: This release refactors the CLI to a subcommand architecture. All existing CLI commands will need to be updated.

Added#

  • New CLI Commands

    • symfluence gui launch - Panel-based web GUI for workflow management

    • symfluence data download|list|info - Standalone dataset acquisition without a full project

    • --shapefile option for symfluence data download to derive bounding box from a shapefile

  • Documentation Improvements

    • New CFuse model guide (experimental differentiable PyTorch-based FUSE)

    • New JFuse model guide (experimental JAX-based FUSE with gradient support)

    • New WM-Fire model guide (wildfire spread simulation with RHESSys)

    • Comprehensive CLI reference documentation with all commands and options

    • Expanded configuration parameter reference

    • Fixed Python API examples in calibration documentation

    • New testing guide for contributors with pytest patterns and CI integration

    • Added agent-assisted contribution workflow documentation

  • Agent Improvements

    • Enhanced show_staged_changes with diff statistics and file summary

    • Improved generate_pr_description with auto-detection of modified files

    • Better PR descriptions with context-specific sections and testing checklists

    • Added documentation for fuzzy matching threshold parameter

  • HBV Module Refactoring

    • New losses.py module with differentiable NSE/KGE loss functions and gradient utilities

    • New parameters.py module with parameter bounds, defaults, and scaling utilities

    • Lazy imports in __init__.py for optional JAX dependency

    • Reduced model.py by ~650 lines through modularization

  • HBV Parameter Regionalization

    • Neural transfer functions for spatially-varying parameter estimation

    • Support for catchment attribute-based parameter prediction

  • HBV Calibration Improvements

    • Hydrograph signature metrics for multi-objective calibration

    • Enhanced optimizer with adaptive learning rates

  • New modular command structure in src/symfluence/cli/:

    • argument_parser.py - Main parser with subcommand structure

    • validators.py - Validation utilities

    • commands/ directory with category-specific handlers

Changed#

  • Complete CLI Refactor

    • Replaced flat flag-based interface with modern two-level subcommand architecture

    • New structure: symfluence <category> <action> instead of symfluence --flag

    • 9 command categories: workflow, project, binary, config, job, example, agent, gui, data

    • Eliminated complex mode detection logic

    • Archived old cli_argument_manager.py for reference

  • New CLI Structure

    # Workflow commands
symfluence workflow run [--config CONFIG]
symfluence workflow step STEP_NAME
symfluence workflow list-steps
symfluence workflow status

# Project commands
symfluence project init [PRESET]
symfluence project pour-point LAT/LON --domain-name NAME --definition METHOD
symfluence project list-presets

# Binary/tool commands
symfluence binary install [TOOL...]
symfluence binary validate
symfluence binary doctor

# Configuration commands
symfluence config validate
symfluence config validate-env
symfluence config list-templates

# Job commands
symfluence job submit [WORKFLOW_CMD] [SLURM_OPTIONS]

# Example commands
symfluence example launch EXAMPLE_ID
symfluence example list

# Agent commands
symfluence agent start
symfluence agent run PROMPT

# GUI commands
symfluence gui launch

# Data commands
symfluence data download DATASET --bbox W S E N
symfluence data list
symfluence data info DATASET

  • HBV Sub-Daily Parameter Scaling

    • Implemented exact exponential scaling for recession coefficients (K0, K1, K2)

    • Formula: k_sub = 1 - (1 - k_daily)^(dt/24) replaces linear approximation

    • Eliminates ~5-13% error in recession behavior at sub-daily timesteps

    • Flux rate parameters (CFMAX, PERC) continue to use linear scaling

    • Added FLUX_RATE_PARAMS and RECESSION_PARAMS constants

  • MESH Preprocessing Consolidation

    • Moved configuration defaults to dedicated config_defaults.py

    • Streamlined config_generator.py and meshflow_manager.py

    • Expanded parameter_fixer.py with robust parameter handling

    • Updated NALCMS to CLASS land cover mapping (wetland, snow/ice corrections)

    • Added unit conversion utilities for forcing data

    • Improved logging (replaced debug prints with proper logger calls)

  • FLUXCOM ET Acquisition Rework

    • Replaced stub ICOS downloader with fully functional _download_from_icos() using ICOS Carbon Portal metadata API

    • Smart variable matching (exact name, then substring, then fallback)

    • Automatic unit conversion (W/m2, mm/hr, mm/day)

    • Nearest-neighbor fallback for basins smaller than the grid cell

  • MODIS ET Acquisition Improvements

    • Spatial subsetting via sinusoidal projection for large tiles with small domains

    • Improved special-value masking (all MODIS QC codes, not just fill value)

    • SDS name matching now tries exact match before substring search

  • Reporting Module Cleanup

    • Simplified plotter implementations

    • Improved shapefile handling

  • Example Notebooks — Typed Config Migration

    • Migrated notebooks 02a–03b from legacy yaml.safe_load + flat dict pattern to SymfluenceConfig.from_minimal() typed config API

    • All 9 example notebooks (01a–04b) now use the same typed, validated, frozen config pattern

    • Removed MAF-specific language from all example notebooks; acquisition comments now reference DATA_ACCESS: 'cloud' or 'maf' config setting

    • Fixed pre-existing bugs: 02a NameError on config access, 03b cf.managers wrong variable name, 03b config overwrite bug, 03b outdated CLI syntax

Fixed#

  • MESH lumped mode routing: switched from run_def to noroute mode to correctly preserve lower-zone baseflow (wf_lzs); run_def in MESH 1.5.6 bypasses STGGW storage, causing zero baseflow

  • Traceback handling in MESH postprocessor

  • Added missing pydantic runtime dependency to pyproject.toml

  • Fixed invalid escape sequences in __init__.py warning filters for Python 3.12+ compatibility

  • CI pipeline now correctly fails on unit/integration test regressions (removed || true from pytest invocations)

Documentation#

  • Added comprehensive sub-daily simulation section to HBV model guide

  • Documented parameter scaling approaches (exact vs linear)

  • Added validation methodology for sub-daily implementations

Migration Guide#

Old Command (v0.6.x)

New Command (v0.7.0)

symfluence --calibrate_model

symfluence workflow step calibrate_model

symfluence --setup_project --create_pour_point

symfluence workflow steps setup_project create_pour_point

symfluence --get_executables summa

symfluence binary install summa

symfluence --validate_binaries

symfluence binary validate

symfluence --doctor

symfluence binary doctor

symfluence --init fuse-provo --scaffold

symfluence project init fuse-provo --scaffold

symfluence --list_presets

symfluence project list-presets

symfluence --workflow_status

symfluence workflow status

symfluence --list_steps

symfluence workflow list-steps

symfluence --pour_point 51/-115 --domain_def delineate --domain_name Test

symfluence project pour-point 51/-115 --domain-name Test --definition delineate

symfluence --agent

symfluence agent start

symfluence --example_notebook 1a

symfluence example launch 1a

Benefits of new CLI:

  • Clearer command organization and discoverability

  • Better help messages (symfluence workflow --help shows workflow-specific options)

  • Easier to extend with new commands

  • Industry-standard pattern (like git, docker, kubectl)

[0.6.0] - 2025-12-29#

Added#

  • Calibration Observation Data Utilities

    • download_smhi_discharge.py - Download discharge data from Swedish Meteorological Institute

    • prepare_streamflow_for_calibration.py - Convert discharge CSV to calibration format

    • setup_calibration.py - Automated calibration setup with parameter bounds

    • Calibration demo tests for Elliðaár (Iceland) and Fyris (Sweden) basins

  • CARRA/CERRA Data Processing Improvements

    • Fixed CARRA longitude normalization in spatial subsetting

    • FORCING_TIME_STEP_SIZE configuration support (10800s for CERRA)

    • FORCING_SHAPE_ID_NAME configuration support with default ‘ID’

Changed#

  • CLI orchestrator integration completed with full workflow execution support

  • Version bump to 0.6.0 across all version references

Removed#

  • Deprecated CONFLUENCE backward compatibility

    • Removed CONFLUENCE.py wrapper file

    • Removed ./confluence shell script

    • Removed CONFLUENCE_DATA_DIR and CONFLUENCE_CODE_DIR configuration support

    • All documentation now uses SYMFLUENCE exclusively

Fixed#

  • CARRA spatial subsetting for small basin extents

  • EASYMORE remapping failures for CARRA datasets

[0.5.11] - 2025-12-15#

Changed#

  • Enhanced ngen outlet detection

  • Cleaned up technical debt

Fixed#

  • Mypy type errors in config property inheritance

  • Completed typed config migration for base classes

Improved#

  • Centralized evaluation metric logic

  • Improved linting and added ruff tests to pyproject.toml

[0.5.3] - 2025-11-15#

Added#

  • Support for cloud acquisition of:

    • Copernicus DEM

    • MODIS land cover

    • Global USDA-NRCS soil texture class map

    • Forcing datasets: ERA5, NEX-GDDP, CONUS404, AORC

  • Agnostic pipeline for cloud-based ERA5 matched workflows

  • Full cloud-integrated workflow tested with ERA5

  • Made MPI worker log generation optional

  • Initial t-route support (in progress)

[0.5.2] - 2025-11-12#

Major: Formal Initial Release with End-to-End CI Validation#

This release marks the first fully reproducible SYMFLUENCE workflow with continuous integration.

Added#

  • End-to-End CI Pipeline (Example Notebook 2a Equivalent) Integrated a comprehensive GitHub Actions workflow that builds, validates, and runs SYMFLUENCE automatically on every commit to main.

    • Compiles all hydrologic model dependencies (TauDEM, mizuRoute, FUSE, NGEN).

    • Validates MPI, NetCDF, GDAL, and HDF5 environments.

    • Executes key steps (setup_project, create_pour_point, define_domain, discretize_domain, model_agnostic_preprocessing, run_model, calibrate_model, run_benchmarking).

    • Confirms reproducible outputs under SYMFLUENCE_DATA_DIR/domain_Bow_at_Banff.

    • Runs both wrapper (./symfluence) and direct Python entrypoints equivalently.

Changed#

  • Updated external_tools_config.py to include automatic path resolution for TauDEM binaries (e.g., moveoutletstostrms → moveoutletstostreams).

  • Expanded logging and run summaries for CI visibility.

  • Protected main branch to require successful CI validation before merge.

Notes#

This release formalizes SYMFLUENCE’s reproducibility framework, guaranteeing that all supported workflows can be rebuilt and validated automatically on clean systems.

[0.5.0] - 2025-01-09#

Major: CONFLUENCE → SYMFLUENCE Rebranding#

This is the rebranding release. The project is now SYMFLUENCE (SYnergistic Modelling Framework for Linking and Unifying Earth-system Nexii for Computational Exploration).

Added#

  • Complete rebranding to SYMFLUENCE

  • New domain: symfluence.org

  • PyPI package: pip install symfluence

  • Backward compatibility for all CONFLUENCE names (with deprecation warnings)

Changed#

  • Main script: CONFLUENCE.py → symfluence.py

  • Shell command: ./confluence → ./symfluence

  • Config parameters: CONFLUENCE_* → SYMFLUENCE_*

  • Repository: DarriEy/CONFLUENCE → symfluence-org/SYMFLUENCE

Deprecated#

  • All CONFLUENCE naming (removed in v0.6.0)