Does my dog actually know the command?

Lola's file carries the concept's clearest statement, and it is addressed to a family rather than to staff. The lead trainer spends several minutes of a first in-program lesson taking apart the treat-in-the-pocket sit, and the punchline is not about the treat at all: the dog learned the reach, because the reach came first. He generalises it - dogs attend to the first thing in the chain - and then draws the conclusion for the handler's own body, which is that you have to discipline yourself not to do anything before the word. By the second lesson the same session produces three separate live instances: the retractable line's constant tension, an accidental hand wave that has got itself conditioned onto the release, and the leash tug becoming the real cue for every movement command. The second-lesson version is the sharpest one in the corpus because it names the failure that follows - a dog who is great on leash and then sits there blank the moment there is no line, because "the command, the cue is the feeling of the leash." Lola's file also supplies the stage past help: a dog can have no help at all and still be answering a word that carries nothing, at which point "it's not really a command, it's a suggestion."

The household with two dogs is where the concept turns into a pass/fail standard rather than a warning. The handler catches his own habit - leaning down and pointing at the ground - and gives it up on the grounds that the dog will learn the point instead of the word, then states the replacement standard in the most usable sentence in the corpus: hands in his pockets, say sit, and she sits. Elsewhere the same file holds the whole gradient: a down that still needs "training wheels" because he pulls her down while saying it; a dog put at "like 50%" on a word because a barely perceptible tug was still in the picture; the moment the tug is deliberately withheld and the word works anyway; and finally the finish line called as it is crossed - no point, no gesture, just the word. And the last piece of help gets faded too. The handler leaves the room, shuts the door, listens, watches through the gap, and returns to correct the break - so the correction lands while the dog believes she is unobserved, which is what turns a rule about this room into a rule.

Bo's file supplies the instrument. Instead of describing the dog as trained or untrained, the handler walks through the behaviours one at a time and records which ones still have an aid attached: the down needs no aid, no negative reinforcement and no pile of treats; the sit the same; loose-leash walking nothing at all; the place cot is the one thing still in the negative-reinforcement range. The absence of help is itself the grade - that is what 'he knows it' is being made to mean. Bo also gets the most uncomfortable version of the test, aimed at the trainer rather than the dog. Twice in one session the trainer removes himself: once taking everybody out of the room, once leaving only his microphone behind, on the stated worry that people obey him because he is standing there. The debrief reports the answer including the part that complicates it.

With a green dog the concept shows up as three separate tests and one argument. The test that matters most is the out-of-sight place: the handler leaves the dog's line of sight entirely, goes inside, and verifies through a window rather than by peeking - and Bullet passes it in one session and fails it three times running in another, with neither the trainer nor the notes reconciling the two. The second test is off leash versus on leash, where he isolates the fault himself: the heel lags only off leash. The third is the smallest and the most honest - he notices that he looked at the dog while giving the release and says so rather than counting the rep as clean. The argument is about luring: a dog staring at a raised toy is chasing the reward rather than attending to where his body is, which is precisely the criterion this handler wants - and he grants, in the same breath, that luring works.

The nervous dog supplies both the best self-check and the most dramatic failure. The self-check is geometric: the ball rides in the pocket on the far side of the handler's body, so that a dog coming for the ball would have to cross him to get to it - which means the heel that just happened was the word. The same file records the habit of spelling commands out loud between handlers so that discussing the plan does not cue the dog. The failure is the cot. Kodak's sit is excellent on the place cot and collapses the moment he is off it; a second staff member describes it in ordinary words - he will sit, lie down and go on place there, and anywhere else he is a different dog - and the session ends with the hardest shutdown anyone has seen from him rather than with an escalation. Kodak is also where the withdrawal curve gets named: ugly, smoother with help, ugly again, and that second ugly stretch identified as an extinction burst rather than as damage.

Sage's program is the only place in the corpus where help is written down. Every skill in every setting is graded each session on a six-point scale that runs from doesn't know it, through a lot of help, active guidance, light help and verbal cue, to verbal and reliable - and help is defined explicitly as anything that indicates or guides the dog into position, hand signal, leash, collar tap or the handler's body, which is what makes a grade mean something. Her file also holds the stage language: she knows the word but does not own it, because she is still escaping the pressure rather than avoiding it, and the analogy given is a friend taught a new word on a hike who cannot define it ten minutes later. The remaining help is measured in fingers - a finger hooked in the collar with a slight downward touch, named as very light help, with verbal-only stated as the next step - and the test is run by taking the leash out of the picture mid-session.

On a severely fearful hundred-pound dog the concept becomes a gate on correction rather than a quality standard. The handler routes the leash behind his own back specifically so the leash cannot be what puts Finn in the heel position, then states the criterion for knowing a word as latency: keep doing it until the response gets quicker and quicker until it is instantaneous, and only then is punishing a refusal honest - with the word 'punishing' immediately qualified. The same session draws the distinction most owners and many trainers collapse: pushing the dog's hindquarters down is not teaching, because what it teaches is inevitability rather than meaning. Later, the release word gets the same diagnosis from the other side - he is still reading the handler's body for it, so the word has to be made to carry it.

Louie's file adds a distinction the existing page does not make. The problem is not always that the leash is doing the work; sometimes the leash is simply one of the conditions the dog has filed the rule under, so the exemption generalises the instant the equipment comes off. Two handlers spell that out as a two-way conversation in his own living room - a dog concluding that he only has to listen while he is on the leash - and then test it by taking the leash off and asking for place. The purest instance is on a walk: he answered when he hit the end of the line and otherwise did as he liked. And the scaffolding end is recorded without embarrassment on his first heel session, where the leash guidance is named as present and the handler declines to bank it as progress.

Luna's sessions hold both ends of the same lever. At the scaffolding end the leash pull is paired with the word on purpose and to saturation, with the payoff named in advance: do it a billion times and the one time you say the word without the pull, she does it anyway. At the fading end the handler caps his own use of the pull-down while he is using it, on the explicit ground that anything that depends on a leash is worthless off leash. Her place work runs the same logic on a bigger unit - forced downs are abandoned because being placed into a position is not the dog doing it, and the criterion becomes a down offered at her own volition.

A large, tool-driven case gives the concept its most granular record and its subtlest form. The granular record is a down tracked by what got removed: full leash pressure, then jiggling the line, then merely picking the line up and pointing, then the word alone - each step named on camera in the order it happened. The subtlest form is a sentence almost nobody says out loud: the down has to work without a leash attached that the dog feels the handler might use. Zeus also supplies the other way a handler fools himself, which is not help at all. He knew sit and treated it as optional - "maybe if I feel like it" - and on the heel he knows both the position and the obligation and simply does not want it, which the handler classifies as a reinforcement problem rather than a teaching one.

With a ninety-six-pound adolescent the hidden help turns out to be the trainer. At the go-home lesson he says that the dog will catch on to him too quickly for any test to mean anything, hands the long line to the owner, and asks her to stop following him around so that there is nothing else in the picture - and in the same session he names the limit of his own work, that the ball gets a speed out of the dog that his recall never has. The owner lessons run the concept in the other direction: he corrects arm movements live, because a repeated gesture becomes part of the command permanently, and he catches the same fault in himself when a lifted arm nearly becomes the cue for sit. The down is help used openly and named as being reduced rather than pretended away.

Stormie's file names two forms of help nobody else in the corpus names. The first is the handler's reach for the remote: if the hand only arrives on the button when a stimulation is coming, the dog learns the reach, so the hand stays on the button permanently. The second is the leash kept on indoors, so that avoidance is not an available answer while the owners are learning their timing. Her own commands are graded against the same bar by the handler working her - a down taken on physical guidance is named as unfinished work, twice, including once to nobody at all - and the rep he actually wants is the one she gives with no collar behind it.

The protection dog's record supplies the metric. Progress on the implied stay is reported rep to rep by how much help was needed - multiple corrections, then a couple, then none - and the way to reach none is a deliberate pause: give the cue and wait a second before any leash input, so the dog gets the chance to choose right. Her heel is the one command still carrying a prop, and the prop is named as training wheels with a stated plan to test without it, while everything else is reported as verbal. And the correction ladder is built precisely so the collar never becomes the cue: it answers an ignored correction, never produces the command, "so then she's not reliant on feeling the e-collar to do the command". The bar for a behaviour counting at all is that no leash is needed.

Dutchess's record holds both ends and one honest exception. At the zero end the handler guides her bodily into the down and says four times over that she does not know it, so there is no ambiguity about what the help is for. At the go-home end he explains to the family that visual aids are training wheels and get faded because the command has to be a hundred per cent verbal - and then keeps the point for place, knowingly, because place needs it. The same lesson warns them that the command will have to be reinforced again on their own cot in their own room, because the dog reads the whole setup and not just the word, which is hidden help stated as a transfer problem.

King's file is where the ethics of help get answered rather than the mechanics. A senior trainer, teaching a junior one, meets the charge that a lure is cheating with a question about a four-year-old and a bicycle: you use the training wheels, and pretty soon you take them off. He then gives the operating rule in one line - start with lots of help, push on, and when confusion appears go backwards, add help, clarify, and progress again - which is the only statement in the corpus that treats adding help back as a normal move rather than a retreat. The same file carries a live example of help nobody planned: the second dog in the session complies beside the person she is attached to and ignores the trainer, and it is that handler who names it.

How do I tell whether he knows the word or just knows my hands?

Take everything out of the picture and ask once. The corpus's own version of this is a sentence you can hold yourself to without buying anything: hands in your pockets, say the word, and the dog does it. If you want to be stricter, put whatever he works for in the pocket on the side away from him, so that if he were coming for the ball rather than the word he would have to cross your body to get it - that is the check one handler runs on his own heel work, and he says out loud afterwards that he did not use the ball as a lure. If the behaviour disappears the moment your hands are still, you have your answer, and it is not a bad answer: it means you know exactly what to work on next.

Is using a treat or a leash to show him cheating?

No, and the trainers answer this directly rather than defensively. The standing reply is the bicycle: you do not refuse a four-year-old training wheels on principle, you put them on and then you take them off. What makes help legitimate in this corpus is that it gets named while it is being used and scheduled to shrink - "He has a little bit of help with the down" followed immediately by "but he'll require less and less help every time" - rather than quietly becoming part of the cue. The failure is not the help; it is help that nobody was counting. One handler goes further and adds help back on purpose when confusion appears, then starts progressing again, which is the only thing in the corpus that treats going backwards as a normal move.

He does it on his bed but not anywhere else. Is that hidden help?

Yes, and it is the version owners are least likely to spot, because from outside it looks like a dog who knows the command. In one session the sit is described as excellent on the cot and "trash" off it, and the second person in the room describes it in ordinary language: bring him this way and he will sit, lie down, go on place, no trouble - anywhere else he is a different dog. The cot was doing work nobody had counted as work. The same problem is what the handover warning about your own cot in your own room is about: the dog learned a setup, and a different cot in a different spot reads to him as random. The fix is not more pressure, it is more places.

It got worse when I stopped helping. Have I broken something?

Probably not - that shape is described on film as the normal one. The sequence named is ugly at first, smoother as help is added, and ugly again when help is taken away, and that second rough patch is read as the dog making a last-ditch attempt to find the exit that used to work. The instruction that follows is unusually blunt about the doubt owners feel at exactly that moment: do not give up there, and do not conclude you are hurting him or scarring him, because what you are actually doing is clarifying. The honest caveat is that the corpus does not tell you how to distinguish that patch from a dog who has genuinely had enough; on the same dog, a fortnight later, the same picture was answered by stopping. If you cannot tell, stopping costs you a session and pressing costs you the dog's willingness, which is not a symmetrical bet.

When is a command actually finished?

Three criteria show up, and they stack. The first is that no help is present at all - no aid, no negative reinforcement, no pile of treats - which is what one handler means when he says a dog knows his downs really well. The second is speed: keep repeating until the response gets quicker and quicker and eventually becomes instantaneous, and only then is it fair to treat a failure as a refusal. The third, and the one most easily skipped, is that the dog should not be merely compliant - the target named elsewhere in the corpus is the correct position and a good emotional state at the same time. A dog who needs a little pressure still knows the word without owning it: he is escaping the pressure rather than avoiding it.

He behaves for the trainer and not for me. Whose problem is that?

It is treated as a measurement problem before it is treated as your problem. Two handlers in this corpus decide that they themselves are the help and remove themselves on purpose. One says plainly that the dog will catch on to him too fast for a test to mean anything, hands the long line to the owner, and asks her to stop following him around so there is nothing else in the picture. Another leaves the room twice in one session - once taking everybody with him, once leaving only his microphone - because he worries that people listen to the dogs' handler rather than to the dog's family. Both report the result including the parts that do not flatter them. So the gap is expected, it is named out loud, and closing it is what a handover session is for.

Teaching the reward instead of the word: reaching for the treat before saying the cue, fifty times, until the reach is the command and the empty-handed cue does nothing.

Adding a hand signal by accident and then owning it forever, because the gesture arrived before the word.

Leading the movement commands with the leash or the hand, so that the dog is answering tension rather than 'come' - and then finding he sits there blank the first time there is no line.

Reaching for the remote only when a correction is coming, so the reach becomes the cue and the dog listens when he sees your hand move.

Assuming the command transferred, when the dog learned the trainer's cot in the trainer's spot - the family's own cot in their own room reads as a new thing.

Reading the dog's obedience for the trainer as the dog's obedience, when part of what is holding the behaviour together is the trainer standing there.

Quitting during the rough patch when the help comes off, which teaches the dog that the old escape route still works.

He sits the moment your hand goes near your pocket, and does not sit when your hands are empty.

He listens on the leash and does as he likes off it - or, more precisely, he listens the moment he feels the line and not otherwise.

He does it perfectly in one spot - on his bed, in the kitchen, on the mat - and not two feet away from it.

The command works while you are looking at him and stops working when you turn away or leave the room.

It works when the trainer, or your partner, or whoever usually handles him is standing there, and falls apart when it is your turn.

He only holds a position on the leash you are holding, and slides out of it the moment there is nothing attached that he thinks you could use.

He knows it, plainly and visibly, and does it when he feels like it - which is a different problem and not this one.