Reference
Every number on this site comes from an open federal dataset. This page lists the source for each chapter, what we did to it, the readiness-score formula, and the limitations a careful reader should keep in mind.
| Dataset | Used in | Vintage | What we do with it |
|---|---|---|---|
| EIA-860 · annual | Retirement schedule, atlas, time machine | 2022, 2023 | Generator-level inventory. We split the "Retired and Canceled" sheet by EIA status code (RE → retired; CN / IP → canceled / postponed). |
| EIA-860M · monthly | Time machine, retirement schedule headline, atlas | Every month, Apr 2016 → latest | Same shape as EIA-860 but updated monthly. We use exact sheet-name routing (Operating, Planned, Retired, Canceled or Postponed) — the _PR proposed-retirement sheets are intentionally skipped because they duplicate rows that already exist on the base sheets with the retirement date inline. |
| EIA-923 · monthly Page 1 | Capacity factor on the atlas | Latest annual per plant | Annual net generation / (nameplate × 8,760) for each plant. |
| EIA-930 · hourly | Generation clock | Trailing 12 months | Mean MWh by hour-of-day × fuel bucket, per balancing authority. |
| EPA CAMD CEMS | CO₂ intensity on the atlas | 2024, 2025 daily files | Gross load + CO₂ per plant per year. We clip CO₂ intensity to NaN when gross_load_mwh < 1,000, because small denominators produce physically impossible values (NG ≈ 0.4, coal ≈ 1.0; anything above ~2.5 t/MWh is a data artifact). |
| LBNL Queued Up | Queue reality check | Latest annual snapshot | Dedup'd by queue_project_id across vintages to avoid double-counting. active + suspended are included in the headline active-queue figure (matches LBNL's own convention); operational projects have already connected and are excluded. |
| HIFLD · substations, transmission lines | Atlas (readiness score) | Public refresh, 2025 | Equirectangular distance from each plant to every substation and line midpoint, ≥230 kV bucketed as "high voltage." |
| EIA gas pipelines · interstate + intrastate | Atlas (readiness score) | Latest available | Distance + interstate-miles-within-25-mi per plant. |
The atlas's readiness score is a 0–100 composite of four sub-scores, each on the same 0–100 scale, weighted as follows:
Each sub-score is visible on the detail card. A missing input scores a neutral 50 rather than zero, so sparse-data plants are not penalized for missing fields.
The atlas focuses on plants where existing interconnection service
agreements (ISAs), transformers, gas taps, and on-site infrastructure
may still be usable. That reuse window decays after retirement: ISA
termination rules across PJM, MISO and ERCOT generally take effect
within a few years of permanent retirement, and prime movers are
typically scrapped on a similar timeline. Including plants that
retired in 2015–2019 on the map would show sites where this
infrastructure is less likely to remain. Those plants are shown
separately as historical context, not as current prospects. The
"Ready now" preset further narrows to
capacity_factor ≤ 20%, retirement inside the near-term window
(last unit retiring 2024–2028), and readiness ≥ 70.
This started as a personal exercise: I wanted to understand the U.S. energy landscape better, and the most effective way I know to learn a domain is to build something with its data. The goal was to force myself to find the public datasets that exist, figure out what they actually measure, and see how they fit together.
I'm not an energy professional. I work in software, and some of what's here is certainly naive or oversimplified — the scoring weights are reasonable guesses, not calibrated models, and I'm sure an experienced grid planner would see gaps I don't. That's fine. This is a learning tool first.
The amount of structured, freely available federal energy data is remarkable. EIA alone publishes plant-level inventories (860/860M), monthly generation (923), hourly grid-level fuel mix (930), and retail pricing (861) — all downloadable as bulk files. LBNL publishes interconnection queue data across every ISO. HIFLD publishes substation and transmission-line coordinates. None of it requires an account or approval; most of it has years of history.
The more interesting part was joining it. The time machine, for example, replays 120 individual monthly EIA-860M Excel files — one for every month since April 2016 — and reconstructs a county-level retirement schedule from each one. That kind of longitudinal view isn't published anywhere; you have to build it by stitching the archives together. The readiness score joins four separate spatial datasets (plant locations, substation coordinates, transmission lines, pipeline routes) into a single per-plant composite. Most of the pipeline effort was in cleaning and aligning these sources, not in the visualization.
The entire pipeline and site were built with substantial help from Claude, Anthropic's AI assistant. I described what I wanted to see and worked through the implementation interactively: the data ingestion, normalization, scoring logic, and D3 charts all came together in conversation, iterating on data joins, scoring formulas, and chart designs as I went. Working this way let me move faster than I would have alone, and cover more of the data than I'd have had the patience to wrangle by hand.