Keep delay dispersion with the input excursion
A comparator output responds after the differential input has crossed its comparison condition. The input excursion beyond that condition affects the propagation delay. The ADI explanation treats overdrive as a material part of a delay description, so a nominal delay detached from its input conditions is an incomplete basis for comparison.
Dispersion describes how that delay varies as the input excursion changes. This is a different question from asking for one representative delay at one stated condition. A device description can therefore include a nominal delay and a separate dispersion description without either being redundant. One concerns a point of operation; the other concerns the behavior across the specified change in excitation.
For a meaningful comparison, keep the excursion and the type of delay statement visible together. The cited TLV360x tables also distinguish overdrive and underdrive conditions, reinforcing that the input history belongs with the timing entry. This guide does not rank devices from a curve or assign a timing budget. It explains why a seemingly faster nominal entry and a more consistent response across input conditions are not the same property.
Retain the input common-mode operating point
Differential input excursion and common-mode voltage describe different aspects of the input. The excursion concerns the difference between the inputs, while common mode concerns the level at which those inputs operate. A comparison that preserves the differential condition but changes the common-mode operating point can still change the delay behavior being described.
The ADI article discusses named rail-to-rail examples in which different input transistor pairs operate over different common-mode regions. Transitioning between those regions can affect propagation delay. This provides a physical reason to retain common mode rather than assuming that a shared input excursion always produces the same timing description.
The explanation remains specific to the examples and mechanisms discussed. It does not establish that every comparator uses the same input-stage arrangement or that delay changes monotonically as common mode moves toward a rail. When reading a timing entry, retain its input operating point alongside the excursion and classification. That separates the question of how far the input moves from the question of where the input stage is operating, without converting an illustrative graph into an unverified numerical rule.
Locate the output crossing under its capacitive load
Propagation delay is defined using an output reference crossing, rather than waiting for the output to reach its final settled value. The ADI discussion makes this distinction explicit. A crossing time and a complete transition or settling description therefore answer different questions, even when both are observed on the same output edge.
Output loading also belongs with the crossing definition. Charging or discharging a capacitive load is part of producing the output transition, and the cited TLV360x prose ties its transient characteristics to a stated load. A delay entry obtained with one load should not be treated as a description of every possible receiving circuit.
Read the reference crossing, direction and loading context together before comparing delay statements. This preserves the boundary between an internal comparison event and the externally observed edge at its defined level. It does not require selecting a load or predicting a receiver response. The useful distinction is that reaching a specified reference point under a stated capacitive load is not synonymous with complete output settling, and neither description alone proves a timing result for an assembled circuit.
Separate a short input pulse from a single-edge delay
A short input pulse poses a different question from the propagation of one isolated edge. The TLV360x table includes a parameter named minimum allowed input pulse width, with a stated input excitation and an output-pulse retention condition. That condition must remain part of the parameter description rather than disappearing behind the word minimum.
The reported entry in the cited table is in the typical column. Its parameter name does not turn that typical entry into an unconditional guaranteed minimum. This is a specification-classification distinction as well as a pulse-response distinction: the name identifies the question, while the column and conditions identify the kind of answer supplied.
Keep this pulse condition separate from a nominal edge delay. An output can have a delay description without that description establishing how every arbitrarily short input pulse will be represented. Comparing the separate pulse parameter preserves the relationship between input duration and the retained output pulse. It avoids deriving a pulse threshold from another timing entry, copying a typical number as a guarantee, or claiming successful event detection for conditions that the document does not establish.
Sources and further reading
Bounded ADI propagation-delay prose and TLV360x SNOSDB1E conditions. No graph values, numerical recommendations, guaranteed typical limit or completed timing measurement is presented.