Synopsis.
Name
Belt and Braces. Someone adds a backup, and it is a reasonable thing to add: a second pump behind the first, a relief valve behind the interlock, another guard behind the guard, more defence in depth, because a system with a spare is safer than a system without one and everyone knows it. A component added in good faith, obviously sound to the one who adds it, because they can see exactly the failure it catches, the specific gap it covers, the single point it removes. And the reliability figure agrees: add an independent backup in parallel and the number the designers compute climbs, two nines becoming three, three becoming four, each addition reading on the page as more safety, because that is what the arithmetic of parallel redundancy says, and the arithmetic is correct for parts that fail independently. And here is the thing about a backup: it does not sit apart from the system, it sits inside it, wired to the same bus, cooled by the same water, maintained by the same crew, watched by the same operators, resting on the same assumptions, so the more backups there are the more interactions there are, and each interaction is a path the designers did not draw, a new way for one part to reach another that was not in the diagram. And the more backups there are, the more they share: a common power supply, a common calibration, a common flawed premise, a common flood or fire or dust, so the independence the arithmetic assumed is not there, and one cause reaches all of them at once, the second pump dead of the same water that killed the first. And the more backups there are, the less each guard watches, because there is another guard behind them, so responsibility diffuses across the redundant set, each trusting the spare to catch what they miss, and the operators, knowing a backup exists, take the risk they would not take without it, running closer to the edge because the edge has a net. Each addition is locally rational: it removes a real single point, it reads as more safety, the designer is right that the specific failure it catches is caught. But past a point, each further backup adds more coupling than cover, more shared cause than independent catch, more diffused vigilance than watchful redundancy, so the true reliability of the system, the number no one is computing, has already turned down while the number everyone is computing still climbs. Nobody erred. Nobody was reckless. The backup was added in good faith, the arithmetic was done correctly, the defence in depth is real defence, and every addition removed a genuine gap. It is only that a backup in a coupled system is not the independent part the arithmetic assumed, so the redundancy that raises the figure is the same redundancy that couples the parts, shares the cause, and diffuses the care, and past the peak the belt and the braces fail together of the one thing that was never redundant.
Arc
The bend is this: adding a redundant backup to raise reliability adds complexity, shares a common cause, and invites shirking, so past a point each further backup lowers the true reliability it was added to raise, while the figure that counts it still climbs. It runs on additions that are each locally rational. A backup added in parallel reads as more safety, and by the arithmetic of independent redundancy it is: the computed figure climbs with every spare, so adding one is the obvious, warranted move and refusing one looks like accepting a known single point (Sagan, on redundancy as the reflexive answer to risk, more security forces read as more security). So the addition is necessary-seeming and not careless, and this is what makes it a trap rather than a fault, because each backup removes a real gap and the defence in depth is real defence. And each addition opens interactions the designers did not draw. A backup does not sit apart but inside the system, wired and cooled and maintained and watched with everything else, so the more parts there are the more paths there are between them, and each new interaction is a way for one part to reach another that was not in the diagram (Perrow, on interactive complexity and tight coupling, the added component multiplying the unforeseen interactions that produce normal accidents). The complexity each backup adds is a new surface for failure, and the surface grows faster than the cover. And the supposedly independent backups share a common cause. The second pump draws the same power, the same crew maintains both, the same premise underlies each, the same flood reaches all, so the independence the arithmetic assumed is not there, and one cause fells the redundant set together (Sagan; Perrow, on common-mode failure, the shared dependency that defeats parallel redundancy at a stroke). The belt and the braces are not two chances but one, wearing the face of two. And the backup invites the very shirking it was meant to cover. Because another guard stands behind each guard, responsibility for catching the failure diffuses across the redundant set, each trusting the spare, and the operators, knowing a net exists, run closer to the edge than they would without it (Latane and Darley, on the diffusion of responsibility across a group; Sagan, on social shirking and risk compensation, the backup relaxing the vigilance and inviting the risk it was added to guard). The redundancy that was to add care subtracts it, and the net that was to catch the fall invites the leap. And the figure climbs while the reliability falls. The number the designers compute rises with every independent parallel unit, so each addition reads on the page as more safety, but the true reliability, which the added coupling and common cause and shirking are quietly lowering, has past a point already turned down, and no one is computing that number (the deceptive metric rising as the system it measures declines). The arithmetic that counts the backups cannot see the coupling they bring, so it reports more safety exactly where there is less. None of it is an error or bad faith. The backup was added in good faith, the arithmetic was done correctly, the defence in depth is real, and every addition removed a genuine single point, so no one erred and the reliability was lost to the structure and not to anyone's fault. What makes a backup read as safety, that it adds an independent chance in parallel, is exactly what the coupled system denies it, because a backup wired and cooled and crewed and trusted with the rest is not the independent part the arithmetic assumed. Count the true independence before adding: trace whether the backup shares power, crew, premise or environment with what it backs, price the complexity each addition brings, and remove the coupling rather than adding the spare, and the turn can be caught before the figure and the reliability part ways: but this means refusing a backup that reads as more safety, spending to find the shared cause that no incident has yet exposed, arguing against defence in depth that everyone can see is prudent, and treating a climbing reliability figure as suspect precisely because it climbs, against every signal that another guard is safer, so the reliability is held only by paying, continuously, the cost of counting an independence that is invisible until it fails.
Stakes
Whether a system can be made safer by adding to it, when each addition reads as more safety and the coupling it brings is invisible until it fails. Every plot but the last moves the same way: toward the backup added, which the designer is right to add and the figure right to reward, and away from the independence counted, which would refuse a spare that reads as prudent and price a coupling no incident has yet shown. The production cannot make the backup the independent part the arithmetic assumed, because a component wired and cooled and crewed and trusted with the rest shares the cause it was meant to cover, so the redundancy that raises the figure is the redundancy that couples the parts and diffuses the care: no one erred and every addition removed a real gap. It can only show that the backup was added in good faith and the arithmetic done correctly and the defence in depth is real defence, that past a point each further guard adds more coupling than cover and more shared cause than independent catch and more shirking than vigilance, and that the one move that holds the reliability, counting the true independence and pricing the complexity and removing the coupling rather than adding the spare, must be made by paying, continuously, the cost of refusing a backup that reads as safety and doubting a figure that climbs, so the system is defended in depth and every addition is sound and the reliability it was raising is the thing it has begun to lose.
Estate
khai-misfits: the house that holds this production and answers for the whole run.