Write the decision before collecting data
A useful test begins with a decision, not a sensor. State which component changes, which measured quantity should respond and what evidence would justify keeping the change. Separate exploratory work from a confirmatory comparison. Exploration can reveal unexpected behaviour, but a hypothesis selected after seeing the data needs a fresh test before it becomes strong evidence.
Define the configuration completely enough to rebuild it. Record component identifiers, geometry versions, assembly settings, crew representation and software versions where relevant. If several properties change together, the result concerns that package. It cannot identify the contribution of one property without additional comparisons or a defensible experimental design.
Qualify the measurement chain
Calibration connects an instrument’s indication to a reference under stated conditions. It does not make every later measurement correct. Check range, resolution, drift, mounting sensitivity and environmental effects for the intended task. A high-resolution display can encourage false confidence when the fixture or test setup introduces much larger variation.
Traceability requires a documented chain of calibrations with stated uncertainties, not simply a familiar laboratory logo. ISO/IEC 17025 addresses the competence of testing and calibration laboratories. Accreditation should be checked against the laboratory’s relevant scope; it is not a blanket endorsement of every service or every engineering conclusion.
Methods: ISO/IEC 17025. Metrology institutions include Germany’s Physikalisch-Technische Bundesanstalt (PTB), the UK’s National Physical Laboratory (NPL) and the US National Institute of Standards and Technology (NIST).
Learn the baseline variation first
Repeat the unchanged setup before comparing alternatives. Include removal and reinstallation if those operations are part of normal use. This distinguishes instrument repeatability from assembly repeatability. Where multiple operators are involved, assess whether the procedure produces comparable results between them. An instruction that depends on an unrecorded personal technique is not yet a stable method.
Use a sequence that helps reveal drift. Alternating baseline and modified configurations can be more informative than testing all baseline trials first and all modified trials later. Randomisation may reduce some order effects, while blocking groups comparable conditions together. Neither technique eliminates the need to log conditions or makes an unsafe trial acceptable.
Treat the track as a changing environment
Track testing combines hardware with push performance, driving, crew posture, ice state and weather. Record the variables that can plausibly affect the outcome, including relevant timing splits. A complete run time can hide changes occurring in different parts of the run. A split time provides a more local observation, but does not automatically identify the physical cause.
Location is part of the context, not a universal correction factor. The naturally refrigerated Olympia Bob Run St. Moritz–Celerina and venues such as the Whistler Sliding Centre, Altenberg and Winterberg present different operating environments and track geometries. A comparison observed at one venue should not be presented as established across all venues without further evidence.
Venue context: Olympia Bob Run St. Moritz–Celerina, Whistler Sliding Centre, Altenberg and Winterberg. Venue references do not imply participation in our testing.
Report uncertainty without hiding the result
Measurement uncertainty describes the dispersion associated with the quantity being measured, based on the available information and measurement model. It is not the same as a mistake, nor is it necessarily equal to the scatter of repeated observations. Calibration, environmental effects, processing and model assumptions can contribute even when repeated readings look very consistent.
State how uncertainty was evaluated and what interval or coverage statement is being reported. Keep uncertainty in a single measurement distinct from uncertainty in a difference between configurations. Shared error sources can be correlated; treating every contribution as independent may distort the conclusion. When the data do not distinguish the alternatives, report that limitation rather than translating a small numerical ranking into a performance claim.
Primary guidance: JCGM 100 (GUM) and JCGM 200 (VIM), published through the Joint Committee for Guides in Metrology at the BIPM.
Leave an auditable decision trail
Predefine invalid-run criteria where practical and retain an exclusion log. Equipment faults, missed procedures and unusual environmental events should be documented with reasons, not silently removed. Keep raw data separate from processed data and record the processing steps. An independent reader should be able to understand how a graph was produced from the original observations.
The final report should state the tested conditions, the comparison, the uncertainty, the decision and the limits of transfer to competition. Sometimes the correct decision is to repeat the experiment or improve the measurement method. That is not a failed programme. It is a more reliable outcome than manufacturing a performance story from evidence too weak to support one.