Local Pipeline First¶

This is the recommended first pipeline-development step for new or modified computations.

Before adding routes, Celery tasks, auth decorators, or dataset registration, make the actual analytical workflow work in a local, inspectable way.

Why Start Here¶

This sequence reduces confusion:

• if the science is wrong, you catch it before integration work
• if the data assumptions are wrong, you see them before HTTP and queue layers hide the problem
• if the output shape is unstable, you avoid publishing bad interfaces

The Local-First Sequence¶

flowchart LR
    A[Choose one ROI and one output] --> B[Run the core function locally]
    B --> C[Validate files, geometry, and fields]
    C --> D[Make it repeatable from shell or script]
    D --> E[Only then add Django Computing API]

What Counts As Local¶

Any of these are valid starting points:

• local sample files on your machine
• your own GEE assets
• public CoRE Stack layer metadata discovered from Public API References and STAC Specs
• existing CoRE Stack public layer inventories exposing gee_asset_path

Minimum Shape Of A Good First Prototype¶

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

def run_pipeline(state, district, block, output_dir, source=None):  # (1)!
    roi = load_roi(state, district, block)
    raw = load_source_data(source=source, roi=roi)
    processed = compute_result(raw, roi)
    validated = validate_result(processed)                           # (2)!
    path = write_outputs(validated, output_dir)
    return {
        "roi": roi["name"],
        "output_path": path,                                         # (3)!
        "summary": summarize(validated),
    }

1. Keep the local entry point simple and explicit.
2. Validate the result before you start thinking about APIs or publication.
3. Return a path and a summary so the run is easy to inspect and compare.

What To Validate Before Moving On¶

• does the region of interest match what you intended?
• are the output fields stable and understandable?
• if the output is spatial, does it open cleanly in QGIS or Python?
• if the output is temporal, are the time windows and units clear?
• can you run it twice without changing behavior unexpectedly?

Practical Exit Points¶

Only After This Page¶

Once the local output is stable, the next developer job is to decide:

• does it need an HTTP route?
• does it need async execution?
• does it need publication or dataset registration?

Those decisions belong to later pages.