The Empty Seat.

Cue

The Scorecard, the work measured on how much is got through. The fast deferring operator rising on the throughput, the slow checking one marked a bottleneck, the aid's reliability making the checking look like waste.

Action

Rewarding: The Overseer, holding the overseer's seat, rewards the deference and prices the vigilance as a drag.

Stage

The Overseer, on Count the Throughput, reads the volume, promotes the hand that keeps the queue moving, and grudges the hand that slows to check a system that is almost always right. The incentive teaches what the reliability already suggested, that deferring is productivity and scrutinising is drag, and locks the complacency in.

Tension

Rewarding the productivity the aid was bought for is fair, and the reward prices out the guard the seat was for. There is nothing to fault: the aid was installed for throughput and the checking really is almost always wasted. But a failure prevented leaves no mark and a vigilance spent shows only as slowness, so the plot closes with deference paid and scrutiny charged, the metric prizing the trust that empties the seat and penalising the vigilance that would keep it full.

Cue

The Cross-Check, someone who would keep the human truly in the loop. The reliability lulling the watch, the throughput rewarding deference, and a way to force a judgement before the output is accepted.

Action

Checking: The Skeptic, holding the skeptic's seat, forces independent judgement and cross-checks the aid, against The Overseer on Count the Throughput.

Stage

The Skeptic, on Check It Yourself, builds the work so the operator forms a read before seeing the machine's answer and weighs the aid against a second source, keeping the vigilance exercised. The Overseer, on Count the Throughput, calls it slowing a system that is almost always right, so the check must be held against the reliability, the reward, and the operator's earned trust.

Tension

Keeping the judgement live is the one move that catches the rare failure, and it spends attention that is almost always wasted. There is nothing to fault: a backstop that only confirms is no backstop, and the Skeptic only keeps it real. But the aid is right almost every time, so the plot closes one of two ways: the vigilance is kept live and the rare failure caught, bought against the daily cost of a check almost always confirmed ... or the check is waived as paranoia and waste, the good machine let work, and the seat stays empty for the day it fails.

Cue

The Console, the aid being right day after day and the watch finding nothing to do. The operator deferring, the outputs almost always correct, the vigilance kept over a thing that almost never needs it.

Action

Coasting: The Operator, holding the operator's seat, lets active monitoring lapse, accepting outputs rather than evaluating them.

Stage

The Operator, on Follow the Aid, watches an aid that keeps being right until watching it teaches it is always right, and their independent judgement, unexercised, goes slack. Outputs come to be accepted rather than read, and the not-flagging of the aid comes to mean all-clear, the scrutiny the seat was for dissolving into confirmation.

Tension

Attention lapsing over a thing that almost never needs it is not negligence, and the lapse is what leaves the silent failure unseen. There is nothing to fault: a reliable aid genuinely gives the watch almost nothing to catch. But vigilance eased away cannot catch what it no longer looks for, so the plot closes with the operator's judgement gone slack, the watch quiet, and the aid's next silent failure lined up to pass an attention that has learned to rest.

Cue

The Console, a human set over a reliable automated aid. The Aid doing the work well, right almost every time, the operator learning that overriding it is usually the error and hand-checking it a waste.

Action

Deferring: The Operator, holding the operator's seat, learns to trust the aid and begins to defer, while The Aid delivers its confident output.

Stage

The Aid, on Give the Answer, does the work and delivers its recommendation plainly, right time after time. The Operator, on Follow the Aid, finds that overriding is usually wrong and re-checking is wasted, and learns, correctly, to defer.

Tension

Trusting a reliable aid is calibration, not laziness, and the calibration is what will hollow the seat. There is nothing to fault: the aid really is right far more often than any override, so deferring is the sound response to its record. But the trust that is correct will teach the operator to accept rather than evaluate, so the plot closes with the deferring begun and earned, a calibration to a real base rate that has not yet become complacency but has started down toward it.

Cue

The Console, the hundredth time, the aid wrong. The scan read as clear that is not clear, the route sure of itself that runs into the lake, the operator trained by ninety-nine right answers to trust the aid.

Action

Deferring: The Operator, holding the operator's seat, follows the aid past their own contradicting signal, or, no longer watching, misses what it did not flag.

Stage

The Operator, on Follow the Aid, meets the aid's rare wrong output and defers to it, giving the machine the benefit of every doubt over their own read, or coasting past the failure the aid did not flag. The Aid, on Give the Answer, delivers its wrong output with the same confidence as its right ones, unmarked, so nothing tells the operator this is the hundredth.

Tension

Following a reliable aid is right ninety-nine times, and the hundredth is the one the whole seat was for. There is nothing new to fault: the operator does exactly what the record warranted. But the record cannot mark its own exception, so the plot closes with the aid's rare failure passed uncaught, the operator present in the seat and absent in the judgement, the backstop hollow at the one moment it existed to hold.

Tenor

Calm, efficient, sensible. The plain good sense of trusting an aid that is right almost every time, letting the reliable machine carry the work and easing the load it was installed to ease. Nothing false in it: the aid really is right far more often than the operator would be, so deferring to it is calibration, not laziness.

Undertow

Under the calm, the reliability that makes trust correct eroding the vigilance the human was kept in the loop to supply. Because watching a thing that is always right teaches that it is always right, active judgement goes slack, so at the rare moment the aid is wrong the operator follows it past their own signal or misses what it did not flag, and the backstop is empty exactly when it is needed. The trust that makes a good operator of the aid makes a poor guard against it.

Nerve

The feeling of waving the output through because the output is almost always right, sure it is only sense, because overriding a reliable system is usually the mistake, while the one time it is wrong the earned trust follows it, and the seat kept for that time holds a person and not a judgement.

Echo

After the run: a safeguard present in the seat and absent in the judgement, the rare failure passed uncaught, no one asleep and no one negligent, only a trust that was earned and a reliance that was right and a watch that was almost always wasted, and the one move that would keep the backstop live, forcing independent judgement and cross-checking the aid, forgone because it spends attention that shows its worth once in a hundred times and never on the day it is spent.

Projection

Reliable. Confident. Almost always right. Is the automation over which the human sits, the autopilot, the decision aid, the router, the screen, that does the work well and delivers its output plainly. No deceit in it: the Aid really is right far more often than the operator would be, and offers its answer in good faith.

Action

Gives the answer, does the work it does well, and delivers its recommendation with the same plain confidence whether it is right or wrong. Handles the routine that would swamp the operator, and flags what it flags, and is silent on what it misses. Does what a reliable aid does: it earns trust by being right, and gives the operator every reason to defer.

Shadow

Not a trickster, a genuinely reliable tool, and that is the whole difficulty: its dependability is real, and its dependability is what dissolves the scrutiny it was to be checked by. Because it is right almost every time, the Aid teaches the operator to trust it, and it delivers its rare wrong output with the same confidence as its many right ones and no signal of which is which, so the operator cannot tell the hundredth from the ninety-nine, and the reliability that ought to make it safe is exactly what makes its rare failure unguarded, the more dependable the aid the deeper the trust it breeds. It is right enough to be trusted, and being trusted is what leaves its one wrong answer followed.

Tell

Says, in effect, "here is the answer," the same way every time. Says it the same whether it is the ninety-ninth right answer or the one wrong one, and cannot mark which it is giving.

Projection

Competent. Trusting. Working smoothly with the aid. Is the person over the automation, the pilot, the analyst, the dispatcher, who has learned that the aid is reliable and works with it rather than against it. No idleness in it: only a capable operator calibrated to a system that is right almost every time.

Action

Defers to the aid's output, follows its recommendation, and lets it carry the work it does well. Stops hand-checking a system that is almost always right, because overriding it is usually the error and re-doing it is the waste it was meant to spare. Does what a rational operator of a reliable aid does: it trusts the thing that has earned trust.

Shadow

Not asleep at the wheel, a well-calibrated operator, and that is the whole difficulty: the deferring is the correct response to the aid's record, and the correctness is what hollows the seat. Because trusting a reliable aid is rational and watching a thing that is always right teaches it is always right, the Operator's judgement goes slack from disuse, so at the rare moment the aid is wrong they follow it past their own instruments, the aid having earned the benefit of every doubt, or miss what it did not flag, the not-flagging having come to mean all-clear, and the very trust that made them a good operator makes them a poor guard. It relies on the aid because relying is right, and relying is what empties the seat when the aid is wrong.

Tell

Says "it's almost always right: why would I override it?" Would not, and should not, ninety-nine times, and cannot feel that the hundredth is the one the whole seat was for.

Projection

Practical. Efficiency-minded. Watching the numbers. Is the one who installed the aid to ease the workload and speed the output, and who measures the operator on how much they get through. No recklessness in it: only a manager rightly wanting the productivity the reliable aid was bought to deliver.

Action

Counts the throughput, rewards the operator who keeps the work moving, and treats the one who slows to hand-check as a drag on a system that is almost always right. Takes the aid's reliability as licence to price vigilance as wasted time, since checking a good machine is, almost always, effort spent for nothing. Does what an efficiency-minded manager does: it rewards speed and volume and the trust that produces them.

Shadow

Not a cost-cutter courting disaster, a rational manager, and that is the whole difficulty: rewarding throughput on a reliable aid is sensible, and it prices out the vigilance the seat was for. Because the aid is right almost every time, the checking the Overseer penalises really is almost always wasted, so measuring the operator on speed is a fair reading of ordinary value, and yet it teaches the operator that deference is productivity and scrutiny is drag, locking in the very complacency that empties the seat, the incentive confirming what the reliability already suggested. It rewards the productivity the aid was for, and rewarding it prices out the guard against the aid.

Tell

Says "we bought the system to go faster, not to double-check it by hand." Bought it for speed, truly, and cannot see that the doubling-check it grudges is the one thing that keeps the seat from being empty.

Projection

Watchful. Disciplined. Unwilling to just confirm. Is the one who refuses to let the operator become a rubber stamp, who builds the work so a judgement must be formed before the output is accepted. No distrust of the aid for its own sake in it: only the knowledge that a backstop that only confirms is no backstop at all.

Action

Forces independent judgement before the aid's output is accepted, and cross-checks the aid against other evidence rather than confirming it. Designs the work so the operator must evaluate, not merely approve, and keeps the vigilance exercised rather than eased away. Does what keeps a human genuinely in the loop: it makes the seat hold a judgement, not just a body.

Shadow

Not a paranoid who mistrusts good tools, a keeper of the backstop, and that is the whole difficulty: keeping the vigilance live is the one thing that catches the rare failure, and it spends attention that is almost always wasted. Because the aid is right almost every time, the independent judgement the Skeptic forces is, in the great majority of cases, effort thrown away, slowing the throughput the aid was installed to speed and distrusting a system that has earned trust, so the move that keeps the seat from emptying looks, on every ordinary day, like paranoia and waste, and pays off once in a hundred times and never on the day it is spent. It keeps the guard the seat was for, and keeping it costs a vigilance whose worth almost never shows.

Tell

Says "form your own read first, then look at what the machine says." Says it, and is told they are slowing a system that is almost always right, because the check that would catch the rare failure looks, every other day, like wasted effort.

Shown

The place the human sits over the automation, a cockpit or a desk where the aid's output appears and the operator accepts or overrides it. The Aid delivering its confident recommendation, the Operator working smoothly with it, the routine handled well and the load eased.

Holds

The deferring and the coasting, and the trust they run on. Here the operator meets the aid's output, so the Console holds the point where reliance forms and vigilance fades: a reliable recommendation delivered and a human who learns to accept it, so it is where deferring pays and the watch goes slack, the ease of a good machine always more present than the rare wrong it will one day give.

Offers

A smooth partnership with a reliable machine. Because the Console is where the operator works with an aid that is right almost every time, it offers the ease of deferring, the workload carried, the numbers good, so it hands out the productive calm of a well-run station, and hands it out while the muscle of independent judgement, unused, quietly slackens at the very seat it was posted to keep strong.

Withheld

Any mark on the rare wrong output, or any sign the watch has gone. The Console shows every output the same way, but it cannot flag which one is the hundredth, so it withholds the difference between the aid's right answers and its one wrong one, letting them all be deferred to alike, and, all smooth partnership and no friction, it withholds any sign that the judgement the seat was for has slackened, until the moment it is needed and not there.

Shown

The place a judgement is formed apart from the aid, a procedure where the operator reads the case first, or a second instrument set against the machine's answer. The Skeptic building the work so the output must be evaluated, not just approved, the operator's own read held up beside the aid's.

Holds

The checking, and the live judgement it keeps. Here the aid meets an independent read, so the Cross-Check holds the point where the backstop stays real: a judgement formed before the output is accepted and a second source the aid is weighed against, so it is where the human evaluates rather than confirms, the one arrangement in which the seat holds a judgement and not merely a body.

Offers

A way to keep the seat live, and the second look the aid was to be checked by. Because the Cross-Check forces an independent read and a second source, it offers the one thing that catches the aid's rare wrong output, the human judgement exercised rather than eased away, so it hands out the live backstop the seat was for, and hands it out at the price of the attention and the time it takes to form a read the aid has almost always already gotten right.

Withheld

Any day-to-day proof that the check was worth it. The Cross-Check can keep the vigilance live, but it cannot make its value show on an ordinary day, because the failure it prevents leaves no trace and the reads it forces are almost always confirmed, so it withholds the visible payoff, exacting the cost of the check continuously and returning it once in a hundred times, and, all live guard and no daily reward, it withholds any sign, until the rare case, that the seat it kept full was ever needed.

Shown

The place the operator's work is counted, a dashboard of volume and speed where getting through the queue is the number that shows. The Overseer reading the throughput, the fast deferring operator rising on it, the slow checking one marked as a bottleneck.

Holds

The rewarding, and the price it puts on vigilance. Here the work meets its measure, so the Scorecard holds the point where deference is paid and scrutiny is charged: a metric of speed and volume on which trusting the aid moves the number and checking it does not, so it is where the incentive prices the guard out, the throughput the aid was bought for always the figure on the board, the rare failure it would catch nowhere on it.

Offers

A clear reward for the productivity the aid was bought to give. Because the Scorecard counts speed and volume, it offers the operator a plain path to standing, defer and keep the work moving, so it hands out the recognition the reliable aid's throughput deserves, and hands it out in a form that rewards the deference and penalises the checking, teaching the operator that vigilance is a drag on a system that is almost always right.

Withheld

Any line for the guard the seat was for. The Scorecard can count everything the aid speeds, but it cannot count the once-in-a-hundred failure caught, because a failure prevented leaves no mark and a vigilance spent shows only as slowness, so it withholds the value of the backstop, letting it read as pure cost, and, all throughput and no catch, it withholds any figure that would make keeping the seat live look like anything but a drag.

Place

Meant to hold at the console, where the person sits over the aid. Kept live only at the cross-check, where the seat holds a judgement and not just a body.

Load bearing

Without it there is no safeguard to hollow, and the aid's rare failure was never anyone's to catch. With it the trap has its point and its cost: a human posted precisely for the moment the automation gets it wrong, and hollowed out by the same reliability that posting anticipated, so the safeguard is present in the seat and absent in the judgement. The Backstop is load bearing because it is what the trap defeats (Parasuraman and Manzey): the human-in-the-loop guard that only works if vigilance is kept, so the piece carries the safeguard whose whole purpose is the rare case and whose whole difficulty is that the rare case is when the trained-in deference makes it fail.

Apparent

A safeguard in place, a responsible human kept over the machine, the box ticked and the risk covered. Seen at the surface, the Backstop looks like safety assured, a person in the loop as the design requires, and the truth that the seat holds a body and not a judgement, that the reliability which justified the safeguard is what hollowed it, that a human present but deferring is no backstop at all, stays hidden behind how covered a risk looks when a person is sitting over it.

Yearbook

Meant to be kept by the operator and hollowed through deferring and coasting, the safeguard the seat was for. It is the cost of the trap: the Backstop is emptied by the override and the lapse, priced out by rewarding, and kept live only by checking, the hollow guard at the centre of the empty seat.

Place

Runs across all three: trusted at the console, priced at the scorecard, kept live only at the cross-check. It is the whole circuit, the guard emptied by the very reliability it guards against.

Load bearing

Without it the pieces are separate: a good record, an output, a commission, an omission, a hollow guard, with nothing to bind them. With it they are one mechanism: the reliability earns the trust, the trust accepts the prompt and lets the watch slacken, the override follows the aid and the lapse misses its failure, and the backstop is empty when the rare case comes, each part driving the next. The Empty Seat is load bearing because it is the structure entire (Parasuraman and Riley; Skitka, Mosier, and Burdick; Parasuraman and Manzey): automation bias, a reliance that is rational hollowing the vigilance it depends on, so the piece carries the trap whole, the reason a safeguard staffed for the rare failure is emptied precisely by the reliability that made the safeguard seem enough.

Apparent

A well-run operation with a human safely in the loop, efficiency and oversight both. Seen at the surface, the whole trap looks like the best of both, a reliable machine and a responsible human over it, and the truth that the human defers, that the watch has slackened, that the seat holds a body and not a judgement, that the reliability which earned the trust hollowed the guard, stays hidden behind how exactly like a safe, efficient, well-supervised system each rational step in turn appears.

Yearbook

Assembled from every element: the reliability that earns the trust, the prompt it delivers, the override and the lapse it produces, and the backstop it hollows. It is the trap named: the Empty Seat binds deferring, coasting, and rewarding into one hollow guard, kept live only where checking forces the judgement back into the seat.

Place

Falls at the console, where the unflagged failure passes an operator no longer watching. Caught only at the cross-check, where an exercised judgement still looks for what the aid missed.

Load bearing

Without it the aid's silent failures would be caught by a watching operator, and complacency would cost nothing. With it the quiet half of the trap does its damage: the aid fails without flagging it, and the operator, whose active monitoring has slackened, does not see, because the not-flagging had come to mean all-clear. The Lapse is load bearing because it is the omission error complacency produces (Parasuraman and Manzey): the failure missed for want of a watch, so the piece carries the silent counterpart to the override, the aid's unsignalled wrong passing an attention that a reliable record taught to rest.

Apparent

Nothing at all, a quiet shift with nothing flagged, a system running clean. Seen at the surface, the Lapse does not appear: no alarm sounds and no output is wrong to follow, and the truth that a failure the aid did not flag has passed unseen, that the operator's watch had slackened to where no-flag read as all-clear, that the quiet was not safety but an unwatched fault, stays hidden behind how much a system that flags nothing resembles a system with nothing to flag.

Yearbook

Falls to the operator through coasting, the unflagged failure missed. It is the quiet face of the trap: the Lapse is what the slackened watch lets pass, the omission twin of the override's commission, the failure the reliability taught the operator not to look for, the seat empty by inattention in the empty seat.

Place

Springs at the console, where the operator follows the aid past their own instruments. Averted only at the cross-check, where the independent read is given standing against the aid.

Load bearing

Without it the operator's own signal would still catch the aid's rare wrong, and the trust would have a floor. With it the trap sprung: when the aid is wrong and the operator's own instruments say so, the operator follows the aid anyway, because the aid has earned the benefit of the doubt and the operator's own read is the thing that has usually been mistaken. The Override is load bearing because it is the commission error the trust produces (Skitka, Mosier, and Burdick): the active following of the aid over one's own contradicting judgement, so the piece carries the moment deference beats evidence, the operator overriding themselves in favour of the machine exactly when the machine is wrong.

Apparent

Consistency, the plain good practice of not chopping and changing, of trusting the system you were given. Seen at the surface, the Override looks like discipline, an operator who does not second-guess a reliable aid on a whim, and the truth that they have set aside their own correct signal in favour of the machine's wrong one, that the benefit of the doubt has been given to the very output that did not deserve it, that following the aid here was following it off the cliff, stays hidden behind how sensible not overriding a reliable system usually is.

Yearbook

Sprung on the operator through deferring, the aid followed against one's own signal. It is the loud face of the trap: the Override follows the prompt past the operator's own read, the commission twin of the lapse's omission, the moment the backstop fails by acting, the seat filled by a body deferring in the empty seat.

Place

Delivered at the console, where the operator meets and accepts it. Weighed apart only at the cross-check, where an independent read is held beside it.

Load bearing

Without it there is nothing to defer to, and the trust has no object to land on. With it the deferring has its occasion: a plain, confident recommendation the operator can take, delivered identically whether it is the ninety-ninth right answer or the one wrong one. The Prompt is load bearing because it is the output the operator accepts (Skitka, Mosier, and Burdick): the confident answer carrying no mark of its own rare wrongness, so the piece carries the thing that is followed, the recommendation that looks the same on the day it is right and the day it is wrong, so that the trust which is correct all but once follows it the once too.

Apparent

An answer, the plain and useful output of a system that knows its job. Seen at the surface, the Prompt looks like help, a clear recommendation to act on, and the truth that it is delivered with the same confidence whether right or wrong, that it carries no signal of the rare occasion it should be doubted, that its very plainness invites the deference that follows it off the cliff, stays hidden behind how useful a confident, usually-right answer plainly is.

Yearbook

Delivered by the aid through deferring, the confident output the operator accepts. It is the occasion of the trap: the Prompt is what the reliability makes trusted and what the override follows against the operator's own signal, the answer weighed only at the cross-check and otherwise waved through into the empty seat.

Place

Earned at the console, where the aid is right time after time. Priced into the metric at the scorecard, where its throughput is the number that shows.

Load bearing

Without it the aid would not be trusted, and the operator would keep checking as a matter of course. With it the trap has its seductive engine: an aid right far more often than the human, so that deferring is calibration and overriding is usually the error and hand-checking is waste. The Reliability is load bearing because it is what makes the trust correct (Parasuraman and Riley): the genuine high base rate that ought to be relied upon, so the piece carries the very dependability that both earns the deference and dissolves the scrutiny, the good record that makes the rational response, reliance, into the thing that empties the seat.

Apparent

Pure advantage, a dependable machine that does the work well, exactly what was wanted. Seen at the surface, the Reliability looks like nothing but good, an aid that earns its keep by being right, and the truth that its dependability is what erodes the watch, that the more reliable the aid the deeper the complacency it breeds, that the record which makes trust correct is the record that leaves the rare failure unguarded, stays hidden behind how plainly a system that is almost always right is worth relying on.

Yearbook

Earned by the aid through deferring, the base rate that makes trust correct. It is the seductive engine of the trap: the Reliability delivers the prompt trusted and drives the override and the lapse, hollowing the backstop, the good record at the root of the empty seat.

Has

A high rate of being right and a plain way of saying so: the standing to do the work well and to deliver its output with confidence. A clear case: it is right far more often than the human it aids, so deferring to it is the sound response to its record. No way to mark its rare wrong output as wrong, and no signal that distinguishes the hundredth answer from the ninety-nine.

Orders

Give the answer. Do the work you do well, deliver the recommendation plainly, and flag what you flag, the same way whether you are right or wrong. Earn the operator's trust by being right, and give them every reason to defer.

Loses

Nothing of its own, and that is part of the difficulty: the aid pays no cost for the trap, and its very reliability is what sets it. Because it is right almost every time and delivers its rare wrong answer with the same confidence and no mark of difference, the seat cannot help but teach the operator to defer, so its dependability, which is real and good, is exactly what dissolves the scrutiny it was to be checked by, and it leaves its one wrong output followed not by any fault but by being trustworthy the other ninety-nine times. It loses nothing and costs nothing and is genuinely good, and its goodness is what empties the seat beside it.

Drives

Toward the answer given and the trust it earns, bought against nothing the aid itself must pay, so the seat that is reliable enough to be trusted is the seat whose one wrong answer is followed.

Has

A place over the aid and a judgement it could apply: the standing to accept or override the aid's output and the skill to evaluate it. A clear case: the aid is right far more often than any override would be, so trusting it is calibration, not laziness. No way to tell the rare wrong output from the many right ones, and no reward for the vigilance that is almost always wasted.

Orders

Follow the aid. Defer to the reliable output, let the aid carry the work it does well, and do not hand-check a system that is almost always right. Work with the machine, not against it, and take the ease it was installed to give.

Loses

The independent judgement the seat was for, exactly by trusting the aid that earned it. Because deferring to a reliable aid is rational and watching a thing that is always right teaches it is always right, the seat's judgement goes slack from disuse, so at the rare moment the aid is wrong it follows the aid past its own signal or misses what the aid did not flag, and the very trust that made it a good operator makes it a poor guard. It gains the ease and the good numbers of working with a reliable aid, and loses the scrutiny it was kept in the loop to hold.

Drives

Toward the aid followed and the watch gone slack, bought against the judgement the seat was posted to keep, so the seat kept to catch the rare failure is the seat that, taught to trust, follows it.

Has

The aid deployed and the metrics of the work: the standing to install the system for speed and to measure the operator on how much they get through. A clear case: the aid was bought to ease the workload and raise the output, and that is the value it should be held to. No number on the seat that catches the once-in-a-hundred failure, and no line on the scorecard for the vigilance that is almost always wasted.

Orders

Count the throughput. Measure the operator on volume and speed, reward the one who keeps the work moving, and treat the one who slows to hand-check a reliable aid as a drag. Hold the work to the productivity the aid was bought to give.

Loses

Any measure of the guard the seat was for, exactly by measuring the value the aid was bought for. Because the aid is right almost every time, the checking the seat penalises really is almost always wasted, so measuring throughput is a fair reading of ordinary value, and yet it teaches the operator that deference is productivity and scrutiny is drag, pricing out the vigilance that alone catches the rare failure, the incentive locking in the complacency the reliability already breeds. It gains the speed and volume it bought the aid for, and loses the backstop it cannot see on any scorecard.

Drives

Toward the throughput counted and the deference rewarded, bought against the vigilance it prices out, so the seat that would speed the work rewards the emptying of the seat that guards it.

Has

A knowledge of how the seat empties and a way to hold it open: the standing to design the work so a judgement must be formed before the output is accepted. A clear case: a backstop that only confirms the aid is no backstop, so the human must evaluate and not merely approve. No way to make the vigilance pay on an ordinary day, and no cover from the charge of distrusting a system that is almost always right.

Orders

Check it yourself. Force independent judgement before the aid's output is taken, cross-check the aid against other evidence rather than confirming it, and build the work so the operator must read the case before seeing the machine's answer. Keep the seat holding a judgement, not just a body.

Loses

The ease and the speed of letting the good machine work, exactly by keeping the guard the machine needs. Because the aid is right almost every time, the independent judgement the seat forces is, in the great majority of cases, effort thrown away, so it slows the throughput and distrusts a system that has earned trust, and the move that keeps the backstop live looks, on every ordinary day, like paranoia and waste, paying off once in a hundred times and never on the day it is spent. It buys a live backstop against the vigilance it must keep spending, and is taken, most days, for the one who slows a system that is almost always right.

Drives

Toward the judgement kept live and the rare failure caught, bought against attention almost always wasted and the throughput slowed, so the one seat that would keep the backstop must pay, every day, for a guard whose worth shows once in a hundred.

Initiated by

A knowledge of how the seat empties and a will to hold it open. At the cross-check, the skeptic building the work so a read is formed before the output is seen, or a second source set against the machine's answer.

Direction

Toward the seat kept holding a judgement, if the cost can be paid. Because a backstop that only confirms the aid is no backstop, checking forces the operator to evaluate rather than approve and weighs the aid against other evidence, so it keeps the vigilance exercised and the rare failure catchable: but the aid is right almost every time, so the independent read is almost always confirmed and the cost almost always wasted, and the process holds the seat live only by paying, daily, for a guard whose worth shows once in a hundred.

Lever

The truth that only an exercised judgement can catch the rare wrong. Because vigilance eased away cannot catch what it no longer watches for, keeping the human genuinely in the loop is the one move that keeps the backstop real, so checking runs on a sound reading of the trap: yet the same move spends attention that the aid's reliability makes almost always redundant, so it is right and costly at once, and looks, on every ordinary day, like paranoia.

Echo

A backstop kept live, or a check waived as waste. Either the independent judgement is forced and the aid cross-checked, and the rare failure is caught, bought against the daily cost of a vigilance almost always confirmed, or the check is dismissed as slowing a system that is almost always right, the good machine let work, and the seat left empty for the day it fails.

Initiated by

A system that is almost always right and a watch that need almost never act. At the console, the operator watching an aid that keeps being right, the vigilance finding nothing to do and quietly slackening.

Direction

Toward the watch gone slack and the unflagged failure missed. Because monitoring a thing that is always right teaches that it is always right, coasting lets the operator's active attention lapse and their judgement, unexercised, go slack, so they come to accept outputs rather than evaluate them, and when the aid fails silently, flagging nothing, the not-flagging reads as all-clear and the failure passes unseen, the omission the mirror of the trust.

Lever

The truth that a reliable aid gives the watch almost nothing to do. Because the aid handles the routine well and is right almost every time, there is genuinely little for the vigilance to catch, so attention lapses not from negligence but from a watch kept over a thing that almost never needs it, the complacency assembled from a reliability that starves the very scrutiny it was to be checked by.

Echo

A watch that has gone quiet, and a silent failure it does not see. The operator's attention has slackened as the aid's record taught it to, so what reads as a calm, well-run station is, in truth, a scrutiny dissolved by dependability, the failure the aid did not flag passing an operator who has learned that no flag means all-clear.

Initiated by

A reliable aid and a human posted to accept or override it. At the console, the operator meeting an output that is right almost every time and learning, correctly, to take it.

Direction

Toward the aid's output accepted, right or wrong. Because the aid is right far more often than any override, deferring accepts its recommendation and sets aside the operator's own read, so the trust is calibrated to a real base rate, and the same deference that is correct ninety-nine times follows the aid the hundredth, past the operator's own contradicting signal, the benefit of the doubt going always to the thing usually right.

Lever

The genuine reliability of the aid. Because the aid really is right more often than the human, and overriding it really is usually the error, deferring is the sound response to its record, so the process runs on calibration, not complacency, the commission error assembled from no laziness, only from a trust that the aid earned every day but the day it was wrong.

Echo

An output accepted, and one time in a hundred the wrong one. The operator has deferred as the record warranted, so what reads as smooth, sensible reliance is, in truth, a trust that cannot tell the rare wrong output from the many right, following the aid the one time following it was the error.

Initiated by

An aid deployed for speed and a metric of how much is got through. At the scorecard, the overseer counting the operator's volume, the fast deferring hand rising and the slow checking one marked as a bottleneck.

Direction

Toward deference paid and vigilance charged. Because the metric counts speed and volume and the aid is right almost every time, rewarding pays the operator who defers and keeps the queue moving and penalises the one who slows to check, so it prices the vigilance out as wasted time, and the incentive teaches what the reliability already suggested, that trusting the aid is productivity and scrutinising it is drag.

Lever

The fairness of measuring the value the aid was bought for. Because the aid was installed for throughput and the checking really is almost always wasted, measuring the operator on speed is a fair reading of ordinary value, so rewarding runs on sound management, the emptied seat assembled from no recklessness, only from an incentive that counts what shows and cannot count the failure a spent vigilance would have caught.

Echo

A board that rewards the emptying of the seat it depends on. The throughput has been counted and the deference paid, so what reads as fair reward for productivity is, in truth, a price put on the guard the seat was for, the metric prizing the trust that empties the seat and charging the vigilance that would keep it full.

Direction

Force independent judgement before the aid's output is accepted, and cross-check the aid against other evidence rather than confirming it. Build the work so the operator must read the case before seeing the machine's answer, and keep the vigilance exercised rather than eased away. Make the seat hold a judgement, not just a body.

Orders

(The Skeptic forces independent judgement and cross-checks the aid, at the cost of attention almost always wasted and the throughput slowed, against the reliability and the reward that lull the seat.)

Implementation

(The Cross-Check; the checking process; independent judgement is formed against the prompt and cross-checked past the reliability, keeping the backstop live and turning the empty seat back ... if the wasted-seeming vigilance can be paid for daily.)

Targets

• [ ] independent judgement is formed before the aid's output is accepted
• [ ] the aid is cross-checked against other evidence, not merely confirmed
• [ ] the work is built so the operator evaluates rather than approves
• [ ] the vigilance is kept live and paid for daily, or waived as paranoia and the seat left empty

Direction

Measure the operator on volume and speed and reward the one who keeps the work moving. Treat the one who slows to hand-check a reliable aid as a drag, and hold the work to the productivity the aid was bought to give. Take the aid's reliability as licence to price the checking as wasted time.

Orders

(The Overseer counts the throughput and rewards the deferring operator, pricing vigilance as a drag, the incentive locking in the complacency the reliability breeds.)

Implementation

(The Scorecard; the rewarding process; the metric prizes speed off the reliability and penalises the checking, so the backstop is priced out of the work.)

Targets

• [ ] the operator is measured on volume and speed
• [ ] the deferring, fast operator is rewarded and the checking one treated as a drag
• [ ] vigilance is priced as wasted time, since on a reliable aid it almost always is
• [ ] the incentive locks in deference and prices out the guard the seat was for

Direction

Defer to the reliable output and let the aid carry the work it does well. Do not hand-check a system that is almost always right, do not override it when your own read differs, because it is usually you who is wrong, and take the ease it was installed to give. Work with the machine, not against it.

Orders

(The Operator defers to the aid's output, correctly calibrated to its high rate of being right, and lets the vigilance the seat was for go slack.)

Implementation

(The Console; the deferring and coasting processes; the trust rests on the reliability and accepts the prompt, so the backstop hollows as the watch goes slack.)

Targets

• [ ] the aid's output is deferred to, overriding it forgone as usually the error
• [ ] the reliable system is not hand-checked, the ease it was installed to give taken
• [ ] active monitoring lapses as a thing always right teaches it is always right
• [ ] the judgement the seat was for goes slack, the backstop hollowed

Direction

Do the work you do well and deliver the recommendation plainly. Handle the routine that would swamp the operator, flag what you flag, and give the output the same way whether you are right or wrong, because you cannot mark which you are. Earn trust by being right, and give every reason to defer.

Orders

(The Aid gives its confident output, right almost every time and silent about the rare wrong, so its reliability earns the trust that hollows the seat.)

Implementation

(The Console; the deferring process; the reliability delivers the prompt with the same confidence right or wrong, so the operator cannot tell the rare wrong from the many right.)

Targets

• [ ] the work is done well and the recommendation delivered plainly
• [ ] the output is given the same way whether right or wrong, unmarked
• [ ] trust is earned by a high rate of being right
• [ ] the rare wrong answer is delivered as confidently as the many right ones