How We Test
This is the protocol any number measured by this site has to follow. It is deliberately ordinary — calibrate, state the conditions, cook, log, break it on purpose, repeat, and photograph everything including the failures.
What this page is for
Every measured figure on The Gentle Fork has to link back here. If you want to decide whether a number is worth trusting, compare its test log against this protocol. The method hub explains why the protocol is useful even when a page contains only a reader-run check, and how cited figures are handled.
The protocol
1. Calibrate before the session. Every probe is checked at the ice point and the boiling point before a testing session, and the offset is written into the log. An instrument that has drifted is retired from the session rather than compensated for in our heads.
2. State the conditions before cooking. Equipment, quantities, ambient temperature and probe placement are recorded first. Deciding what to record after seeing the result is how measurement turns into storytelling.
3. Cook it and log continuously. Medium temperature and internal temperature are read at intervals rather than once at the end, because the shape of the curve is usually more informative than the final number.
4. Take a batch past the window on purpose. The failure temperature cannot be inferred. Somebody has to keep going until the dish breaks, note where that happened, and describe what specifically failed.
5. Repeat. Once tells you what happened. More than once starts telling you whether it will happen again. We publish the number of runs so you can weigh the claim yourself.
6. Photograph the failures too. The failed batch gets the same lighting and the same framing as the successful one. Side-by-side photographs are the single most persuasive thing this site can publish.
7. Publish with provenance, or do not publish. Anything measured carries the log reference. Anything from elsewhere carries the source and retrieval date. Anything untested carries no number at all.
The test log
Every run fills in the same fields. A page cannot publish a measured number unless the corresponding log entry exists.
| Field | What goes in it |
|---|---|
| Tester | The name of the person who cooked it. Not a house byline, not an alias. |
| Date | The day the run happened, not the day it was published. |
| Equipment | Hob type, pan material and diameter, oven, vessel, probe model class. |
| Ambient | Kitchen temperature and anything unusual — a window open, an oven still warm from a previous run. |
| Quantities | Masses rather than volumes wherever the result depends on them. |
| Probe placement | Where the sensor sat, described precisely. A badly placed probe invalidates everything downstream. |
| Readings | The medium temperature and the internal temperature, with times. |
| Result | What the food was actually like, in plain language, plus the photograph. |
| Failure point | Where it stopped working, found by deliberately taking a batch past the window. |
| What went wrong | Every run that failed, and the diagnosis. These are published, not discarded. |
| Run number | Which attempt this was, so the published replicate count is honest. |
To keep your own log, copy these field names into a notebook or spreadsheet and fill in one row per run. They are the same fields the protocol uses.
Honesty about sample size
A result from a single run must say "one run" rather than leaving the sample size implicit. A stated small sample is useful. An unstated one is a quiet claim to rigour the evidence has not earned. The same applies to equipment: one hob is one hob, and a single-appliance record must say so.
What we do not do
- We do not interpolate. If we measured at two points, we publish two points, not a smooth line through the gap.
- We do not round to look tidy. A measurement that landed awkwardly is published awkwardly.
- We do not average across conditions. Results from different pans, masses or ovens stay separate.
- We do not generate anything. No estimated temperatures, no generated photographs, no invented sources.
- We do not measure food safety. Those figures are cited from named authorities on the page that uses them, never derived here.
Questions
What instruments do you use? The protocol calls for probe thermometers cross-checked against each other and against the ice and boiling points, plus a surface thermometer for pan work and, where possible, a logging probe for anything running longer than an hour. A page describes the class of instrument and its calibration state rather than turning the protocol into a product placement.
How is probe placement decided? By finding the slowest-heating point of the food and putting the sensor there, then describing that placement so you can reproduce it. For thin items and fillets this is the largest single source of error in home measurement, so pages that depend on it show the placement rather than describing it vaguely.
Why does ambient temperature matter? Because it changes how fast things cool, how much carryover you get, and how a low-temperature hold behaves. A result from a warm kitchen in August is not automatically the same as one from a cold one in February, and pretending otherwise is how unreproducible numbers get published.
How many replicates is enough? It depends on how much the result moves. A stable, obvious transition may need very few runs; a marginal one needs more. Rather than pretending to a standard we cannot always meet, we publish the count and let you judge. A page saying "one run" is more useful than a page implying ten.
Do you control for ingredient variation? Only partly, and we say so. Egg size, fat percentage, fish thickness and cut geometry all move results. Where variation is likely to matter more than our measurement precision, the page states that rather than presenting one number as definitive.
What happens to a result that contradicts a published source? It gets published alongside the source, with both figures visible and our conditions stated. We do not quietly adopt whichever number is more convenient, and we do not assume our own kitchen is right.
Can I run this protocol at home? Yes — the log fields in the table above are all you need. Calibrate, write down your conditions before you cook, log as you go, and record the run that failed. If your numbers differ from a measured figure on this site, check your calibration offset and probe placement first, then compare conditions field by field; a difference you can explain is more useful than one you average away.
Try it yourself
Start with calibration — it is two minutes and it changes everything downstream. Then copy the log fields and run one dish properly. Most cooks are surprised by how different the reading is from the assumption.