A kettle develops a fault. The kitchen circuit's breaker should open, the kettle should die, and nobody else in the building should ever know. Instead the distribution board's incomer trips, a whole floor goes dark, and somewhere a designer's drawing shows a perfectly compliant installation doing exactly what it was never checked against doing.
That failure has a name, a section of the standard, and a set of rules almost nobody demonstrates. Selectivity is the arrangement where a fault is cleared by the nearest upstream device and only that device, and Section 536 of BS 7671:2018+A4:2026 is where it lives.
What Section 536 actually asks
The section's framing is the part most designs miss. Regulation 536.1 states plainly that it does not set requirements for selecting any single device; it sets requirements for selecting devices to achieve coordination between them, and it says in terms that the requirements cover continuity of supply. Regulation 536.3 then requires the mutual interaction of devices to be considered so they do not adversely affect the safety of the installation, across short circuit, overload, and residual current conditions.
In other words: every device on your schedule can be individually correct, and the schedule can still fail Section 536, because 536 is a question about the relationships.
The myth of ascending ratings
The habit that stands in for selectivity on most projects is picking ratings that ascend towards the origin. A 32 A final, a 100 A sub main, a 200 A incomer, and the assumption that bigger upstream means slower upstream.
In the overload region that instinct roughly holds, because thermal curves scale with rating. In the short circuit region it collapses. A solid fault presents the same thousands of amps to every device in series with it simultaneously, every one of them is into its instantaneous region, and which opens first is decided by mechanism speed and let through energy, characteristics that live in the manufacturer's data and nowhere on your drawing. Two breakers whose ratings differ by a factor of three can race, and the upstream one can win.
The four permitted ways to prove it
Regulation 536.4.1.2.1 for overload and 536.4.1.3.1 for short circuit both give the same closed list of verification routes: a desk study using the product standards and the manufacturer's published data, software tools where the manufacturer provides data for that specific use, testing to the applicable product standard, or the manufacturer's own declaration.
Read the list again for what is not on it. Habit is not on it. Ascending ratings is not on it. The standard names its acceptable evidence, and all four routes lead through manufacturer data, which is the point: selectivity is a property of specific device pairings, not of ratings.
The desk study route carries its own discipline. For overload comparisons using time current curves, the reference ambient temperature and the load conditions before the overcurrent both have to be accounted for. For short circuit comparisons using energy limitation curves, the voltage the curves were drawn for matters.
Total, partial, and the selectivity limit current
Short circuit selectivity frequently is not all or nothing. A device pairing can be selective for fault currents up to a specific value, the selectivity limit current, published by the manufacturer, and non selective beyond it. The standard recognises this directly: selectivity may be total or partial.
That turns the design question into a comparison the drawing alone cannot answer: is the prospective fault current at this point of the installation below the pairing's selectivity limit? Below it, the kettle takes out the kettle. Above it, the kettle takes out the floor, compliantly.
RCDs: the one place the rule is simple
Regulation 536.4.1.4 gives residual current selectivity the section's most concrete test, and it has two conditions that both must hold: the upstream RCD must be a selective type, type S or time delayed with an appropriate setting, and its rated residual operating current must be at least three times the downstream device's.
A 30 mA device downstream of a 100 mA device is a ratio of 3.3 and passes the second condition; if the upstream device is also type S, the pair coordinates. The same 30 mA downstream of another instantaneous 30 mA is a race, every time, whatever the drawing implies. Which is precisely what happens on installations where an up front RCD and RCBO finals were specified separately, each correct, never coordinated.
A detail from implementing this
Selectivity is on CurrentProof's roadmap and it is not in the product today, and the reason is the same honesty rule the tool already applies to disconnection times. Proving selectivity between two specific devices requires those manufacturers' curves and selectivity tables, exactly as the standard's four verification routes insist. Shipping a selectivity verdict without that data would be manufacturing precision, and Section 536 itself says the evidence has to come from the manufacturer.
What the standard's structure makes clear, and what the tool will follow when it ships, is that the check belongs at the pairing level: this upstream device against this downstream device at this fault level, answered from data, not from the comforting slope of ascending ratings.