AI data operations · Public energy data
A working review application for monthly EIA solar and battery data.
Onboard accepts two monthly workbooks, maps them to one typed inventory, matches project records, and shows exactly what changed. Additions, removals, monitored field changes, and questionable source values stay in a persistent review queue until an operator records a decision.
The operational question
EIA publishes a refreshed generator inventory every month. An operations team importing the Planned worksheet needs to know which solar and battery projects appeared, disappeared, changed schedule or status, or contain values that deserve verification before they reach an internal tracker.
Onboard converts both workbooks into the same typed schema, matches records by Plant ID and Generator ID, records the precise differences in each field, and creates a review queue. It preserves the original rows and source fingerprints so every result can be traced and reproduced.
Working application
These captures come from the running FastAPI application after executing the bundled May to June comparison. The interface filters 529 review items, shows the observed source value and recommended action, persists each operator decision, and updates the remaining count.
CSV and XLSX inputs are validated, scoped, typed, and matched with Plant ID plus Generator ID.
Each decision is written into the run's exported review queue and remains available after a restart.
The suite covers public metrics, additions, removals, field changes, source quality, duplicate keys, review persistence, cleaning, provenance, and API validation.
The real datasets
Scope: Technology = Solar Photovoltaic or Batteries. The workbook's Provenance sheet records both source URLs, complete SHA-256 file fingerprints, the 17 monitored fields, and every execution rule.
System flow
Read the Planned worksheet from the May and June 2026 EIA-860M generator inventories.
Select Solar Photovoltaic and Batteries records; normalize identifiers, capacities, coordinates, statuses, and planned dates.
Match Plant ID + Generator ID and detect additions, removals, and changes to 17 monitored fields.
Send project changes and questionable source values to a review queue instead of silently syncing them.
Export raw extracts, clean data, monthly changes, review records, source hashes, rules, and measured metrics.
Actual records
66342::GAIASPlanned operation dateThe system does not infer a construction date from nearby projects or external context.
67770::30200Capacity validationBoth capacity values are preserved and flagged because a plausible explanation is not the same as a verified correction.
58639::BESS8Planned operation dateThe monthly change is recorded as a schedule shift and held for approval before a downstream project tracker is updated.
Inspect the evidence
Reproducible outputs
The formatted workbook is the fastest review. The individual files make every stage available for inspection or rerunning.
Where AI fits
Claude can propose schema mappings and transformations only from a fixed list of approved actions. It cannot send executable Python to the runtime.
The published EIA result uses the deterministic planner, so a reviewer can regenerate every artifact without an API key or a changing model response.
Source anomalies and consequential monthly changes remain pending until a person approves, rejects, or explicitly keeps the source value.
Claim boundary: these are measured results from one portfolio run on public EIA data. They are not client outcomes, accuracy claims, estimates of time savings, or production benchmarks.