The Cycling That Wastes What the Capacity Gained
Why this matters
Every start costs something before it delivers anything. Metal has to change temperature, a coil has to wet up, a loop has to charge, a burner has to bring an exchanger up. During that stretch the equipment is drawing full input and putting out less than its rating, sometimes far less. Add capacity and the on-cycles get shorter, so the same fixed cost per start gets divided into a smaller and smaller amount of useful output. That is how a system with more capacity delivers less of what the customer feels. This article is about the accounting, and about the one measurement that tells you whether cycle length is your problem at all.
Before you make anything cycle on purpose
You will be starting and stopping equipment repeatedly to get this number, and that is where techs create hazards they did not intend.
Do not induce cycling by interrupting a safety. Never jumper, defeat, or "test" a limit control, a flame safeguard, a low-water cutoff, a high or low pressure switch, or a relief device to force a start or stop. Cycle the equipment from the operating control, the thermostat, or the service test mode only.
Rapid restarts on a compressor risk starting under a pressure differential and can stall or overheat the motor. Respect the equipment's anti-short-cycle delay rather than defeating it, and leave at least the manufacturer's stated minimum off-time between starts.
Watching a start means standing near equipment that is about to move. Keep sleeves, straps, meter leads and hands clear of belts, sheaves and fan blades before you call for the start, and if you need a reading from inside a moving assembly, stop the equipment and isolate it under 29 CFR 1910.147 for the mechanical and stored energy rather than reaching past a guard.
If your logging requires a probe inside an enclosure with live parts, that is energized electrical work under 29 CFR 1910.333(b)(2). Establish an electrically safe working condition where you can, and prove your instrument live-dead-live per NFPA 70E-2021, 120.5 before trusting any reading you take.
Hot delivered fluid burns on contact. Fit a temperature probe into an existing well or strap it to the outside of the line with the line at rest; do not open a pressurized or hot fluid path to insert one, and relieve pressure and verify zero at a gauge before removing any plug.
The per-start transient, and how to measure it
The transient is the time from start command to the point where delivered output stops rising. It is not the same as time-to-first-output, and it is not a number you can look up: it depends on mass, on charge, on how long the equipment sat off, and on the condition it started from.
Measure it as its own quantity:
- Let the equipment sit off long enough to be at its normal off-cycle condition. Ten minutes off is a different starting point from four hours off, so record which one you used.
- Log delivered output temperature at the point the load sees it: supply air a few feet downstream of the coil or exchanger, or supply water at the outlet, sampled every 10 seconds.
- Start it and log until the trace flattens.
- Call the transient over when delivered temperature reaches within 10 percent of its eventual steady value. State that convention alongside the number, because a tech who used "within 2 percent" will get a longer transient on the same machine and the two figures will not compare.
Write it down as a pair: transient in minutes, and the off-time it started from. One without the other is not reusable.
Rating standards already account for this loss. Equipment ratings published under AHRI test procedures carry a cyclic degradation term precisely because measured cyclic output falls below steady-state output. You are measuring the field version of a number the industry already agrees exists.
The gate
Per on-cycle, at the run fraction the system actually sees: if the measured transient is more than one third of the observed on-time, cycle length is the constraint.
Unit of analysis is one on-cycle, sampled at the mild-weather condition where the system spends most of its hours, not at design. This is a single gate. Do not pair it with a cycles-per-hour threshold; a system with a long transient is hurt at four cycles an hour, and a system with a negligible transient is fine at eight.
When it triggers, the step is to lengthen on-time to at least three times the measured transient, which puts productive output at two thirds of the cycle or better. On a differential-based control, on-time scales roughly with the differential, so raising the differential by a factor lengthens both on-time and off-time by roughly that factor and halves the cycle rate when you double it. The cost is proportional: room swing tracks the differential, so doubling it roughly doubles the swing the occupant feels. Thermal mass and any cycle-rate or anticipator setting modify both, so verify with a log rather than assuming the factor landed.
The productive fraction is (on-time minus transient) divided by on-time. Run both of the following through that expression and through the gate above.
Case one: the system where cycle length is the whole problem
A hydronic system with a high-mass boiler and a long distribution loop. Complaint is that the far end of the building never comes up, on a system that satisfies its own controls.
Measured from a 30-minute off-time, supply water reaches within 10 percent of its steady value at 4.0 minutes. Logged on-time at the current outdoor condition is 6.0 minutes.
Gate: 4.0 / 6.0 is 67 percent, well above one third, so it triggers. Productive fraction is (6.0 - 4.0) / 6.0 = 0.33. Two thirds of every on-cycle is spent getting the loop to temperature, and the far emitters, which are last to see hot water, are getting the least of what remains.
Step: target on-time of at least 3 x 4.0 = 12.0 minutes. That is double the current 6.0, so widen the control differential by roughly a factor of two, for instance from 1.0 degree F to 2.0 degrees F, and expect the space swing to roughly double as well.
Result at 12.0 minutes on-time: productive fraction is (12.0 - 4.0) / 12.0 = 0.67. The productive fraction exactly doubled, from 0.33 to 0.67, with no change to capacity, fuel input, or a single component.
The trade is real and has to be said to the customer before you make it: they will feel a wider swing between calls. On a space with high mass and slow response, that swing is mostly invisible. On a light space with a lot of glass, it is not, and the honest answer there is that the fix is a capacity or control-strategy change, not a differential change.
Case two: the same gate, the opposite call
A low-mass electric resistance heater in a small space. Same complaint shape from the customer, "it cycles constantly."
Measured from a 30-minute off-time, delivered air reaches within 10 percent of steady value at about 20 seconds, which is 0.33 minutes. Logged on-time at the current condition is 6.0 minutes, the same as case one.
Gate: 0.33 / 6.0 is 5.6 percent, far below one third, so it does not trigger. Productive fraction is (6.0 - 0.33) / 6.0 = 0.944. Cycling is costing about 6 percent of delivered output.
The call is to leave the differential alone. Widening it to chase a 6 percent loss buys back at most a few percent of output and pays for it with double the temperature swing, which is the thing the customer is actually noticing. On this system the complaint about constant cycling is a complaint about audibility and swing, and the productive answer is either a control with a tighter differential and a higher cycle rate, or a conversation about what normal operation sounds like.
Same measured on-time, same complaint wording, opposite recommendation, and the transient is the only input that differed. That is why the transient is worth the twenty minutes it costs to measure.
Stored is not lost, and lost is not stored
The productive fraction is an output accounting, not an energy accounting, and conflating the two will make you overstate the case to a customer.
Energy that went into heating mass during the transient is stored. A cast-iron boiler, a heavy exchanger, a masonry surface, a full loop of water: much of that comes back out during the off-cycle as the mass gives it up to the space. It arrives late and it arrives where the equipment is rather than where the load is, but it is not gone. The cost is distribution and timing, not fuel.
Energy that went into changing a state that then reverses is lost. The clearest case is a cooling coil that wets up during the run and then re-evaporates that condensate back into the airstream during the off-cycle when the blower keeps turning. The latent capacity delivered during the run is handed straight back. That one is a genuine loss, and it is why off-cycle blower strategy matters more on short cycles than on long ones.
A jacket or flue that keeps losing heat to outdoors during the off-cycle turns stored into lost, at a rate set by the insulation and by any standby draft path through the equipment. So the same transient costs more on a unit in an unconditioned space than on one inside the envelope, and that is a condition worth naming before you promise a number.
What this gate does not decide
It does not diagnose why the on-time is short. Short on-time comes from oversized capacity, from a control differential set tight, from a sensor in the supply stream that satisfies before the space does, from a limit tripping, or from a load that genuinely is small right now. The gate tells you whether shortening matters, not what caused it. The sibling article on oversizing covers the sizing path to a short on-time.
It does not apply cleanly to modulating equipment. A unit that ramps its output down instead of stopping has traded cycle losses for part-load efficiency effects, and the relevant question becomes whether the load sits below the minimum firing or minimum stage. Measure the minimum stage output against the load before you go looking for cycling losses on that machine.
It does not override a manufacturer's minimum on-time or minimum off-time. Those exist for oil return, motor cooling, purge, and ignition sequencing. Where the equipment's documentation sets a floor, the floor wins and your target moves up to meet it.
Verifying you changed the right thing
Re-log after the change, at a comparable outdoor condition, from a comparable off-time. Both halves matter: comparing a post-change log taken on a mild evening against a pre-change log taken on a cold morning is comparing two different loads and will show you whatever you want to see.
Confirm three things in the new log. On-time reached the target you set, not just "longer." The transient did not grow, which it will if you also changed something on the supply side. And the space swing landed where you predicted, because if it doubled and you told the customer it would not, the next call is about that.
Then check the complaint, not the numbers. If the far end of the building still does not come up with a productive fraction at 0.67, cycle length was not the constraint after all and the log you just built is the evidence that clears it, which is worth as much as a fix.
References
- AHRI equipment rating test procedures, for the cyclic degradation term applied to rated part-load performance
- 29 CFR 1910.147 for mechanical isolation and stored energy; 29 CFR 1910.333(b)(2) for energized electrical work; NFPA 70E-2021, 120.5 for live-dead-live proving
- Manufacturer documentation for minimum on-time, minimum off-time, and anti-short-cycle delay values
- See related: Why Oversizing Hurts; What Changes Inside a System During a Long Run Cycle