explore-platform

name	explore-platform
description	Discover what SYMFLUENCE offers by introspecting the live registry and config schema — available models, forcing/observation datasets, optimizers, calibration targets, metrics, presets, templates, and config keys. Start here before choosing a model/dataset to run, or before extending the platform.
when_to_use	["A user asks \"what models / forcings / optimizers / metrics are available?\"","Choosing a HYDROLOGICAL_MODEL, FORCING_DATASET, optimizer, or metric for a config","Checking whether a config key exists / finding the right key name","Before extending the platform (see what's already registered first)"]

Exploring the SYMFLUENCE platform

SYMFLUENCE is registry-driven: models, datasets, optimizers, metrics, calibration targets, presets, and the whole config schema are looked up from a live registry, not a fixed list. Always discover capabilities by querying the platform — never rely on a hardcoded or remembered list, which goes stale. This skill is the discovery entry point for both running the platform and extending it.

1. The one command: symfluence list

symfluence list                # all catalogs + counts
symfluence list models         # registered hydrological/routing models (HYDROLOGICAL_MODEL)
symfluence list forcings       # forcing & attribute datasets (FORCING_DATASET, attribute sources)
symfluence list observations   # streamflow / in-situ observation networks
symfluence list optimizers     # calibration algorithms (OPTIMIZATION_ALGORITHM)
symfluence list targets        # calibration targets (streamflow, snow, ET, ...)
symfluence list metrics        # objective/evaluation metrics (OPTIMIZATION_METRIC)
symfluence list presets        # project init presets
symfluence list templates      # config templates you can copy
symfluence list steps          # the workflow steps (= symfluence workflow list-steps)
symfluence list config-keys    # every recognized config key (the public contract)

Each catalog is read straight from the registry / config schema at runtime, so what it prints is exactly what this install supports — including any plugins.

2. The Python equivalents (when scripting or in a notebook)

from symfluence.core.registries import R
sorted(R.runners.keys())                 # models
sorted(R.acquisition_handlers.keys())    # forcing/attribute datasets
sorted(R.observation_handlers.keys())    # observation networks
sorted(R.calibration_targets.keys())     # calibration targets (e.g. SUMMA_STREAMFLOW)
sorted(R.metrics.keys())                 # metrics

from symfluence.optimization.optimizers.algorithms import list_algorithms
list_algorithms()                        # optimizers

from symfluence.core.config.canonical_mappings import FLAT_TO_NESTED_MAP
sorted(FLAT_TO_NESTED_MAP)               # every recognized config key

R exposes ~30 registries (core/registries.py) — the same .keys() works on any of them (preprocessors, postprocessors, workers, parameter_managers, evaluators, delineation_strategies, bmi_adapters, …).

3. From "what's available" to using it

The catalog names ARE the config values. To run something, set the matching key:

• A model from list models → HYDROLOGICAL_MODEL: <name>
• A dataset from list forcings → FORCING_DATASET: <name>
• An optimizer from list optimizers → OPTIMIZATION_ALGORITHM: <name>
• A metric from list metrics → OPTIMIZATION_METRIC: <name>
• A key from list config-keys → confirm spelling / discover an option

Then author and run the config — see the run-workflow-locally skill (CLI, required keys, stage markers, config validate).

4. From "it's missing" to extending it

If symfluence list doesn't show what you need, add it — the registry is the extension surface. Pick the matching skill:

• A new dataset → add-data-handler
• A new model → add-model-handler
• A new optimizer/search algorithm → add-optimizer
• A calibration that misbehaves → debug-calibration

After registering, re-run symfluence list <kind> to confirm the new entry appears (registration runs at import; if it's absent, the import or entry point isn't wired — see the relevant skill's troubleshooting).

5. Verify before asserting

When a user asks "can SYMFLUENCE do X?", run the matching symfluence list (or R.*.keys()) and answer from the result — do not answer from memory. The live registry is the source of truth.