A sample graph is a figure. It is not, by default, an observation. The scientific failure mode of hobby weather websites was to publish a pretty PNG with no caption, then let visitors treat it as “the station.” A demo image that omits station identity, period, units, and the word illustrative is a mislabeled product. It teaches the wrong lesson even when the underlying logger was honest.
The historical URL /samp-graph.php sat on the Cumulus host as a sample or demonstration graph, not as the operational temperature chart. A related test path, /test/bramptonweather/samp-graph.php, used the same idea in a named sandbox. Neither address is a live feed. This article is about the caption contract those URLs implied.
Historical context
Cumulus and related PHP tooling needed screenshots: a graph that showed what a template would look like, a library example, a layout test. Operators copied those images onto live pages. Incoming links then treated /samp-graph.php as if it were current weather. The original weather-station site also kept a real temperature time-series chart and a JPGraph project page. Those three URLs are easy to conflate because all three can emit a line plot. They answer different questions.
- Operational chart — this station, this interval, these units, this sensor.
- Sample graph — an illustration of how a chart might look.
- Library project — how software draws axes, not what the air did.
The restoration keeps the sample URL as a methods page so the backlink still names the right problem: unlabeled demos.
The caption is part of the measurement
Meteorology already solved this for professional figures. A plotted series is usable when a reader who never saw the website can recover:
- What was measured — temperature, not “weather.”
- Where — station name, identifier, and location at a resolution that distinguishes this yard from the next town.
- When — start, end, timezone, and sample interval.
- In what unit — °C or °F, hPa or inHg, mm or inches, stated on the figure.
- What kind of series it is — observed samples, a model forecast, a derived index, or an illustration.
- Whether it is this station — or a vendor demo, a neighbor’s file, or a synthetic sine wave used to test a library.
Omit any of those and the PNG is decoration. WMO instrument practice treats identity, time, and exposure as part of the observation, not as optional HTML. A METAR is the extreme form of the same contract: location, time, and quantities in one labeled line. A hobby graph that drops the labels is worse than a sparse METAR, because it still looks complete.
Call the sixth item out loud. “Sample,” “demo,” “example,” or “not this station” belongs in the title or the first sentence of the caption. CSS that makes the plot look like the live dashboard is not a substitute.
How unlabeled demos mislead
They mislead in ordinary, repeatable ways.
Wrong station. A template ships with a sample from the developer’s logger. The adopting site never replaces the file. Visitors cite “Brampton” or “Mesa” or no place at all. Later archives inherit the error.
Wrong time. A static PNG from 2008 remains on a page that says “current.” The image timestamp in the filesystem is not the observation time. If the plot includes a clock in the corner, that clock is only as good as the server that stamped it.
Wrong unit. A sample drawn in Celsius is dropped into a Fahrenheit site, or the axis is cropped so the unit label disappears. A 25 on an unlabeled axis is not a temperature.
Wrong evidence family. A demo that overlays a forecast ribbon on observed points, with one legend, trains readers to treat model output as a thermometer. Forecast and observation can share a canvas only if the legend says which line is which.
False completeness. Smooth curves hide missing hours. Sample data are often gap-free because they were invented or cherry-picked. Real loggers drop samples. A demo that never shows a gap teaches operators to “fix” gaps by interpolation and then forget they did.
False precision. Library examples love two decimal places. Sensors do not.
None of these failures requires malice. They are what happens when a figure is treated as chrome.
What an honest sample graph looks like
Put the caption in text next to the image, not only inside pixels. Pixel text dies when the image is reused. A sufficient caption, as prose:
Illustrative sample only. Not an observation from this website’s station. Variable: air temperature. Unit: °C. Period: 00:00–24:00 local, timezone stated. Interval: 5 minutes. Source: synthetic or named third-party file. Generated: date of the image file.
If the sample is from a real station, name that station and the date, and still say it is a sample for layout, not today’s weather. If it is from a test path such as the Brampton weather sandbox on this host, the caption should carry that name rather than implying the operational station.
Do not use a sample as a source for climate remarks, heat records, or “the station confirmed.” Those claims need an operational series with siting notes, as on the temperature-chart page.
Operational charts versus samples versus libraries
Keep the three products on three URLs and three titles.
The operational temperature chart must survive a forensic test: could a later analyst align it with another record? That test fails if the file is actually a leftover sample.
The sample URL may show a realistic-looking diurnal wave. Realism is not provenance. A sine wave with a daily period can mimic a clear-sky cycle and still be fiction.
A plotting library can render either. The library page should not be cited as a weather source. If you need to explain how axes were drawn, that is software history. If you need to explain what the air did, you need the logger, the unit, and the screen.
Practical checklist for operators
- Filename and URL contain
samp,sample,demo, ortestwhen the figure is not operational. - Visible caption states illustrative status before any numbers.
- Station, period, timezone, variable, and unit are in HTML text.
- The live dashboard does not hotlink the sample file.
- Replacement of the sample is part of commissioning a new template.
- Archives store the caption with the image.
- No climate or warning language is attached to a demo PNG.
If you cannot fill the caption, do not publish the graph.
Modern relevance
Unlabeled figures still leak into blogs, social posts, and “AI” slide decks as if they were data. The correction is editorial, not graphical: name the station, the period, the unit, and the evidence type. TNET’s research library treats atmospheric series as one evidence family only when those fields are present. See the evidence families behind a connection outlook. Related Cumulus publishing notes sit on the Cumulus legacy hub.
This page does not display a sample plot. It states the rule the historical sample URL should have made obvious.
Sources
- World Meteorological Organization, Guide to Instruments and Methods of Observation (WMO-No. 8): community.wmo.int
- National Weather Service, METAR decode key (complete labeled observation line): weather.gov (PDF)
- National Weather Service, forecasts and observations as distinct products: weather.gov
- TNET, The evidence families behind a connection outlook