Fleet Power Draw + Approximate Headroom — Stacked

Draw and headroom stacked so the top edge of each area is approximate total running-generator capacity. Both series are 7-day rolling averages (daily actuals are noisy at this stacked-area scale; see the previous report for unsmoothed daily points).

Source: InfluxDB MARAGON_Generators (2024-09-17 floor) + gen_telemetry.db kw_rating roster. Same headroom approximation as the prior report — fleet-level floor (not per-group), current-roster kW ratings applied retroactively, Mesa run_status as the running gate (not the live "newest-wins" override). Full caveats at the bottom.

MW Stacked — Full History

2024-09-18 to 2026-09-23 · 7-day rolling average, draw (green) + headroom (orange) stacked

Shaded region, 2025-12-22 to 2026-01-24: operator-confirmed gas supply disruption — feeder lines froze over, cutting supply to several sites (GN, Will, Stout, WW). This is a near-force-majeure case: the orange band widens here not because spare capacity opened up, but because supply itself couldn't fill the gens that were still nominally "Running." See the footer for why this reconstruction can't tell that apart from real headroom on its own.

MW Stacked — Last 180 Days

Zoomed view of the same stacked series

Share of Capacity (draw % + headroom % = 100%) — Full History

2024-09-18 to 2026-09-23 · 7-day rolling average, normalized to approximate capacity each day

Same 2025-12-22 to 2026-01-24 gas-line-freeze window shaded — this is where the orange (headroom) share visibly balloons in the normalized view, for the same supply-side reason, not a demand drop.

Share of Capacity — Last 180 Days

Zoomed view of the same normalized series

What "capacity" means here: the top edge of the stacked MW chart, and the 100% line in the percentage chart, is not measured directly — it's load + headroom, where headroom is the approximation described in the previous report (kw_rating × fraction of the day each gen's Mesa run_status read "Running", summed fleet-wide, floored at 0 against load). Treat the capacity line as an upper-bound estimate of what the currently-running fleet could carry, not a hard nameplate ceiling and not the live dashboard's per-group-floored figure. In particular this does not know about historical disabled_gens, preemptive_hold, or Scheduled Load derates (none are logged with a changelog, only current state) — so any day those were active, the true spare capacity was smaller than shown here.

Why the shaded gas-freeze window inflates the orange band: when generators keep reporting "Running" (Mesa's status string, which is what gates capacity here) while load collapses because supply — not demand — is constrained, capacity stays counted but load craters underneath it, and the arithmetic reads that as headroom. It isn't: there was no more gas to burn, so the gens could not have carried more load no matter how many miners were awake. This reconstruction has no visibility into gas supply, so a supply-side event and a genuine low-utilization day look identical in the chart — only outside context (like this operator note) tells them apart. Compare this to a scenario where the generators themselves shut down on low pressure instead: then both load AND capacity drop together for those gens, and headroom does not spike at all — it's specifically the "gens still counted as running, nothing to feed them" combination that produces the inflated band seen here.

Reading the percentage chart: outside the shaded window, the fleet has spent most of its history in the 75–95% utilized band (draw as a share of approximate running capacity), with other low-utilization stretches mostly corresponding to the same known dips flagged in the prior report (e.g. around 2026-04-04, 2026-07-06/07) — not investigated further here. A day where the green band hits ~100% means running gens were fully loaded with no spare estimated at all that day, not that every generator in the fleet was on.