The Instant Observation Concept (Snap Reading)
Record a state at a defined instant and distinguish that observation from continuous duration measurement.
Updated:
A snap reading records what happens at a selected instant, not what happened before or what someone appears about to do. Accumulating readings under an appropriate design allows sampling to estimate time proportions for a defined population. It complements timing specific operations.
Define the unit and instant
Choose a person, machine or workstation as the observation unit. A round of ten workstations is not a simultaneous photograph: each reading happens at a different time. Keep planned and actual times and check whether route order favours particular states.
For example, a person holds a tool while awaiting authorisation. Do not classify direct work merely because the tool is in their hand: apply the observable rule and verify causes separately. If visibility is insufficient, record a missed reading rather than reconstructing what might have happened.
Classify without judging
Define exclusive categories with examples: operation, preparation, transport, material waiting, equipment waiting, breaks and other. Conversation may be coordination; a pause does not establish fatigue or lack of discipline. A taxonomy improves reproducibility but cannot guarantee no ambiguity. Pilot it and compare observers.
There is no universal five- or ten-second duration ensuring a pure reading. The state refers to the selected instant; annotation may take longer provided it does not reconstruct a sequence.
Randomness, independence and precision
Select instants with known probabilities within included shifts and days. Randomisation reduces convenience selection but guarantees neither independence nor removal of reactivity. Nearby rounds and people observed together may share conditions.
For a proportion under the simple independent-Bernoulli model:
n = ceiling[z²p₀(1−p₀)/e²]
With z=1.96, p₀=0.5 and a target absolute margin of five percentage points, plan 385 readings; for three points, 1,068. Confidence belongs to the interval procedure; it is not the probability that each reading is correct. The later approximate margin is z√[p̂(1−p̂)/n]; near extremes, consider Wilson, and adapt analysis for dependence.
Simulated example
Of 400 valid readings, 220 show operation, 60 preparation, 50 waiting, 40 transport and 30 breaks/other. Operation is 55%, with an approximate normal 95% margin of 4.9 percentage points under simple assumptions. The remaining 45% is not removable savings. Investigating the 50 waits requires distinguishing supply, authorisation and equipment.
Increasing sample size reduces random uncertainty; it does not correct an excluded night shift or a rule that misclassifies waiting.
Observer reactivity and safety
Explain the purpose and train recorders. If familiarisation is used, agree on its scope before seeing results; do not automatically discard initial days or apply percentage corrections for a presumed Hawthorne effect. There is no universal week after which it disappears.
Observe from authorised positions without entering hazardous areas to meet a schedule. Minimise personal data and review access, retention and legal basis.
What can be concluded
Wrench time is a study-defined maintenance direct-work proportion, not a general efficiency judgement. OEE needs availability, performance and quality: compatible time records, ideal cycle time and total and good units are required. A snapshot of an active machine does not provide all components.
WorkSamp supports observation planning and recording. Review the design and data as well as the calculation.