Chart conventions
Eight conventions run through every view. Each one exists because breaking it would mislead somebody in a specific way, so the reason is given alongside the rule — the reason is the part worth carrying to the next chart.
Colour encodes identity, never rank
Each facility keeps its colour across every chart, and a filter that changes the facility count does not repaint the survivors. If colour tracked rank, an asset would change colour when another one improved, and a reader comparing two screens would conclude the wrong asset had moved.
Tier chips are one hue at three weights, not a traffic light
Tier 1 is a legitimate screening estimate applied to a source that does not warrant a measurement campaign — it is not a failure. Red-amber-green would read as a quality judgement on the analyst rather than a description of the evidence.
An interval is drawn as a band behind its line, never as two lines
Two lines invite the reader to trace the upper one, which is the pessimistic boundary of a distribution rather than an estimate. A band reads as one estimate with a width, which is what it is.
Bubble area is proportional to value, not radius
A radius-scaled circle overstates a large value by the square. On a fleet where one asset carries half the methane, that is the difference between a fair picture and a dramatic one.
A modelled comparator is dashed; a measured one is solid
Sitting below a modelled global mean and sitting below an airborne survey of a neighbouring basin are not the same achievement — and the weaker comparator is usually the more flattering, so the difference has to be visible without a legend.
Lower-is-better axes say so on the axis
Carbon intensity is one of the few figures here where down is good. Leaving the reader to infer that from the unit is how a chart gets read backwards in a meeting.
Every chart carries its data table and a CSV export
A screen reader gets nothing from an SVG of bars, and a regulator disputing a figure needs the numbers rather than a picture of them. The table is always in the page, hidden from sight until asked for but never from assistive technology.
A projection band and a historical interval are different objects
The historical interval is a Monte-Carlo ensemble over parameter uncertainty. The projection band is driver error from hold-out validation, widened with horizon, plus flare-efficiency uncertainty. Drawing them identically would suggest a 2040 figure has the standing of a 2024 one.
A rising flaring share on falling production looks like a problem, and a falling flared volume over the same months looks like an improvement. Both charts are correct. The screening test is written against the share, so that is the one a 2030 verdict turns on — and the Activity & trends view shows both rather than choosing for you.