Methodology

How the data is collected

Everything on this site comes off public APIs, pulled on a fixed schedule. Prices, inventories, supply and the Mont Belvieu spot come from the US Energy Information Administration's open data API: the weekly heating-season price survey, the Weekly Petroleum Status Report and the daily spot series. Heating degree days come from NOAA's Climate Prediction Center. There is no scraping, no estimation and no text-extraction step; every figure comes straight off a structured agency series.

Each pull covers a trailing window, not just the newest week, and a missed or partial run backfills on the next one. If one week's job fails, that week fills in automatically once the pipeline runs again; nothing is silently skipped or left blank.

The seasonal survey

EIA's residential and wholesale price survey runs October through March. From April through September those series publish nothing, by design, and the site treats that silence as expected rather than flagging it stale. EIA has occasionally run reduced off-season surveys; where a legitimate summer reading exists in the record, we keep it, so a summer chart can show the odd real point inside the pause.

How comparisons are computed

Week over week, versus four weeks ago, and year over year are each the change between the latest value and the value from that many days back on the same series. There is no smoothing or averaging in these figures; they are two readings compared directly.

"Vs 5-year median" compares the current value to the median of that same calendar week across the prior five years. Median, not average, so one freak winter does not skew the baseline. EIA's own publications use a 5-year average, so our figure and theirs can differ slightly; ours is always labeled "vs 5-yr median".

Two figures on this site are derived, both pure arithmetic on published series. Days of supply, also called domestic product-supplied coverage, is primary stocks divided by the trailing-four-week average of domestic product supplied (EIA's own convention). It is a coverage ratio, not a depletion forecast: it does not net out exports or account for demand changes, so it never predicts a run-out date. The spread pages subtract one published price from another for the same week: residential minus wholesale for the same area (a broad gross-spread indicator, not a clean dealer margin), and US residential minus the Mont Belvieu spot from the survey's week. Nothing is modeled.

Units, stated plainly

Prices are dollars per gallon at EIA's three decimals, with weekly moves quoted in cents. Inventories are millions of barrels; EIA publishes them in thousands of barrels and we convert for readability. Supply, production, exports, imports and demand are thousand barrels per day. We write that unit as "kb/d" on charts and tables and as the spelled-out "million barrels per day" in running copy (2,910 kb/d and 2.910 million barrels per day are the same number); we never abbreviate it "Mb/d", because the M would collide with the M in "million barrels" and read a thousand times too large. The weekly inventory series counts propane and propylene together, per EIA's definition, and the site says so wherever it matters.

What "normal" means, and where it breaks down

"Normal" on this site always means the five-year median described above. Five years is short enough to compute cleanly and recompute every year, but short enough that an unusual stretch, like the export boom of recent years, moves the baseline with it. If the prior five years were themselves tight, "at normal" can still mean tight in absolute terms. Treat our figure as a fast, consistent yardstick, not the final word on what is typical.

The off-season price treatment

The EIA residential and wholesale price survey runs October through March, and EIA publishes no monthly residential propane series that continues through the summer, so from April through September the price pages deliberately hold their final surveyed readings rather than showing a gap or an estimate. That is a choice, not a lag: the last winter print is the freshest survey data that exists. Off season those pages label their comparison table "Final weeks of the ... heating season" and lead with the year-scale rows (vs the 5-year median, then year over year), keeping the final week-over-week and month-over-month prints below them as end-of-season readings. The Mont Belvieu hub keeps trading year round; its daily and monthly-average prices live on the spot page.

What "vs 4 weeks ago" means here

On a weekly series the 30-day comparison lands about four weeks back, not a calendar month, so the table labels it "vs 4 weeks ago" rather than "month over month". It compares a reading to the print closest to thirty days earlier on the same series, not to a calendar-month average. In the off-season table that means the final surveyed week against the print about four weeks before it.

How the badges read

Every scored page carries one of three symmetric badges against the 5-year median: "Above normal" (more than 10% above), "Near normal" (within 10% either side), and "Below normal" (more than 10% below). The colors follow the market's risk direction: for inventories, below normal is the alarming side; for prices, above normal is. The badge's hover text carries the date of the data it rests on, which matters most during the off-season price freeze.

Update cadence and freshness

The pipeline runs on the EIA weekly cycle; the brief that reads it goes out once a week. Every data page carries two dates: "Data through" is the newest reading in the data itself, and "page updated" is when the page was last rebuilt. A delayed or failed pull shows up immediately as a stale data date instead of a silently frozen number. During the April-September survey pause the price pages' data date sits at the end of the season, which is the freshest data that exists.

For the plain-English version of how to read a page once the numbers are in front of you, see how to read this.