Fleet Power Draw & Approximate Headroom — Full History

Daily actual (thin) with trailing 7-day rolling average (bold). Two series: generator power output (MW, solid ground truth) and an approximate reconstruction of "spare running-generator capacity" (MW, the historical stand-in for the dashboard's "could be using X MW more" tile).

Data pulled 2026-09-24. Source: InfluxDB MARAGON_Generators bucket — true floor of that bucket confirmed by unbounded-range query is 2024-09-17. Today (partial day) excluded; last complete day is 2026-09-23 UTC.

Headroom here is an approximation, not a replica of the live dashboard tile — see method note below before treating it as authoritative.

Power Draw — Full History

2024-09-18 to 2026-09-23 · 736 daily points

Shaded region, 2025-12-22 to 2026-01-24: operator-confirmed gas supply disruption — feeder lines froze over, cutting supply to several sites. Not detected from telemetry; noted here from operator knowledge. See headroom chart below for how this same event reads on the capacity side.

Power Draw — Last 180 Days

Zoomed view of the same series

Approximate Headroom — Full History

2024-09-18 to 2026-09-23 · capacity − load, floored at 0, fleet-wide (not per-group)

Shaded region, 2025-12-22 to 2026-01-24: same gas supply disruption (frozen lines). This is the near-force-majeure case discussed below — generators at GN/Will/Stout/WW kept reporting "Running" while load collapsed to a fraction of normal, because supply, not demand, was the constraint. The headroom spike this produces (peak ~22.8 MW on 2025-12-23) is not spare capacity anyone could have filled — there wasn't enough gas to run the gens fully loaded in the first place. Treat any headroom spike coinciding with a supply-side event as a symptom, not an opportunity; see the explanation of this failure mode further down.

Approximate Headroom — Last 180 Days

Zoomed view of the same series

Power draw method: for each UTC day, mean power_avg (kW, already-scaled Mesa 5–15min telemetry) per generator, summed across every gen_id measurement reporting that day, ÷1000 for MW. A down gen reports ~0 kW so it naturally drops out of the sum without a separate status filter. This matches the dashboard's own canonical method (sum of running gens' output) in spirit, but is a daily mean of periodic snapshots, not the live 1 Hz field-collector figure — short-term (sub-day) swings and the exact instantaneous number will differ from what the live status page shows at any given second.

Headroom method (approximation — read before using): capacity(day) = Σ over gens of (nameplate kw_rating × fraction of that day's readings where Influx's run_status field read "Running"); load(day) = the power-draw figure above; headroom = max(0, capacity − load). kw_rating comes from the latest known value per gen_id in gen_telemetry.db (126 of 166 gens seen in the run_status history matched; the other 40 defaulted to 330 kW, the fleet's common per-gen rating — a minor approximation, not a material one). This deliberately mirrors the live dashboard's logic (capacity only from gens whose run status says Running) but differs from it in three ways worth knowing before you quote a number off this chart:

Fleet-level, not per-group-floored. The live tile floors headroom at 0 per generator group before summing, so one group sitting over target can't cancel another's shortfall. This reconstruction floors once at the fleet level, which can read slightly higher than the true per-group figure during any day when groups pulled in opposite directions.
Disabled/held-out gens are invisible here. disabled_gens and preemptive_hold state exist only as CURRENT fields in master_config.json — there is no changelog, so a gen the operator deliberately took off the table on some past date reads as ordinary spare capacity in this series. Same for Scheduled Load derates (this uses nameplate kw_rating as target, not a possibly-derated group target). Net effect: this is best read as an upper bound on true historical headroom, not an exact figure.
Uses Mesa's status string, not the live "newest wins" override. The live dashboard prefers the field-collector's live reading over Mesa's when it's fresher — which means a gen Mesa flags with a fault (e.g. "Low NG Pressure") can still count as running live if the field collector's own frame says Running. This series uses only Mesa's run_status string, so a gas-curtailed gen that Mesa correctly flags as down will NOT count as running here, even in the edge case where it would still be counted live. That makes this series the more conservative of the two for exactly the "did a gas outage get hidden as headroom" question — but it also means it can disagree with the live tile on any given day for a different reason than #1/#2 above.

Two failure modes worth telling apart when a site goes down: (a) the generators themselves stop (e.g. curtailed on low gas pressure) — draw drops for real, and headroom does not rise, because a non-running gen contributes zero to both capacity and load and simply vanishes from the equation; (b) the generators keep reporting Running while the miners/pods go dark or supply can't fill them — draw drops the same way, but headroom spikes, because capacity is still counted while load collapses underneath it. The shaded region above (2025-12-22 to 2026-01-24, operator-confirmed gas line freeze) is case (b) at fleet scale: with supply itself constrained, the gens could not have been loaded further even though the arithmetic says ~20 MW of "headroom" was sitting there. This reconstruction has no visibility into gas supply or pod state, so it cannot distinguish a genuine spare-capacity day from a supply-constrained one on its own — that distinction has to come from context like this, not from the chart.

Several other real multi-MW swings are visible in the raw series (e.g. 2026-04-04, 2026-06-24, 2026-07-06/07) — these were not individually investigated for this chart; ask if you want a specific dip broken down by site/generator, ideally cross-referenced against gen_down_events.category (which separately tracks gas/overload/manual/electrical/etc. outages back to 2026-03-18).