Derating produces a table of numbers that all look the same, which is why a bad derating analysis is so hard to spot in review. Every entry below yields a report that passes.
| Pitfall | What it looks like | The guard |
|---|---|---|
| Nominal stresses | The check run at typical line, typical load, room ambient | Worst case on every axis: max line, max load, hottest qualified environment, end-of-life tolerance |
| Board ambient used as part temperature | Every part on the card at 55 °C because the enclosure is | A temperature per part, from a thermal model; θja and the local ambient, not the specification's |
| Ratios checked, temperature not | Every stress ratio green, no junction temperature anywhere in the report | Both axes per part, both in the deliverable |
| No rule set named | "Derated to 50 per cent" with no document behind it | The standard, its revision and the application class, recorded once in the policy |
| Two rule sets mixed | A subsystem checked to one document, the rest to another, results merged | One rule set per assembly, recorded; never merged into a single verdict |
| Rating taken at the wrong temperature | A 0.25 W resistor treated as 0.25 W at 100 °C | The rating curve, not the headline: above the knee the part's own rating is already falling |
| Contact ratings read across load types | 5 A resistive contacts carrying 3.2 A into a capacitive load | Ratings are per load type; a motor, lamp or capacitive load has its own number |
| Tantalum treated as an ordinary capacitor | A voltage ratio checked, the circuit-impedance condition ignored | Read the family rule, not just the ratio column |
| Transients folded into the steady-state check | Inrush counted as normal current, or excluded and never analysed anywhere | Steady state here, transients in their own analysis, both referenced from the report |
| The derating report and the prediction disagree | One at worst case, the other at nominal, nobody comparing | One operating point per part, feeding both |
| Violations sized rather than typed | "Only 4 per cent over" treated as nearly compliant | The stress factor is nonlinear; classify by axis and fix, not by margin |
| Waivers without mechanisms | A disposition reading accepted, engineering judgement | The mechanism, the evidence and the named authority, and the mode raised in the FMECA |
| Coverage never checked | A parts list of 400 and a derating table of 340 | Reconcile against the BOM; the missing parts are usually the late additions |
| Frozen at first issue | The report dated at PDR, the thermal design changed twice since | Re-run on every change to dissipation, ambient, duty cycle or thermal interface |
Two of these deserve a closing sentence each. Using the board ambient as the part temperature is the one that produces the most confident wrong answer, because it makes the second axis disappear: every part comes out at the same temperature, every temperature check passes, and the analysis silently becomes a stress-ratio spreadsheet. The worked example's regulator is the illustration, and it is not an exotic case: a linear regulator dropping a couple of watts in a small package is on most boards ever built.
And the frozen report is the most common. Derating is the analysis whose inputs move most and whose document is revised least, because nothing about a parts list looks out of date. The trigger for a re-run is not a design review; it is any change that moves a dissipation, an ambient, a duty cycle or a thermal path, and most programmes have several of those between issues of the report.