Here is the rule. A movement earns a tenure floor of six to eight exposures before you are allowed to judge it. After that, if you get three consecutive sessions at matched RPE with no added reps and no added load, it is dead. Cut it. Those two numbers, six to eight and three, are a defensible choice, not a derived constant. A coach running seven exposures and a four-session stall window is not wrong. What is wrong is having no number at all.
Open a program that has been running for a year and count the exercise list. The ones that land in my inbox run fourteen movements on a given mesocycle, sometimes more. Now ask how many of those fourteen have been audited, meaning someone looked at the trend line and decided whether the movement is still doing work. The honest answer, most of the time, is five or six. The rest just accumulated. Somebody added a cable pull-through in March because a session felt stale, and it has been sitting on the sheet since, never removed, never defended, never once asked to justify the slot it occupies.
The sheet grows by one interesting movement per interesting week and never shrinks, because shrinking requires a decision and growing only requires an impulse. It is a drawer nobody has cleaned out.
An exercise is an instrument, not a stimulus
The confusion starts at selection. When you add a movement to a program, you are not adding "stimulus" in some vague sense. You are adding an instrument, and an instrument's whole job is to produce a reading. Load at matched RPE, or reps at matched load. One number, trending, session over session.
If you cannot name that number before the movement goes on the sheet, you did not select an exercise. You filled fifteen minutes because the session needed something there. The two failures are the same failure, just at opposite ends of the lifespan: no criterion at entry means no criterion at exit, which means the movement just sits there until someone gets bored of it, which is not a training decision. It is a mood.
Tenure follows from this directly. The first handful of sessions on any new movement are motor learning, not adaptation you can trust. The bar path is unfamiliar, the stabilizers haven't sorted out their job yet, and the RPE you report on session two is partly a report on how foreign the movement still feels. A movement rotated out every block, which is a pattern I see constantly in programs built around "variety," never survives long enough to produce a trend at all. It gets replaced while it is still noise. Rotation feels like intelligent programming. Mechanically, it is a way of never having to audit anything, because nothing ever sticks around long enough to be auditable. A conjugate system is the exception that proves the rule: it rotates the variation and audits the pattern, so the squat to a box and the squat to a lower box are two readings of one instrument. Rotation without that pattern-level ledger is just churn.
The criterion, and what it requires
Watch for three consecutive sessions where the number does not move, same reps, same load, same effort. Two sessions is noise, a bad night's sleep, a stressful week at work leaking into the bar. Three in a row is a pattern. Three is where the cost flips. Cutting after two risks discarding a working movement over one bad week. Waiting past three buys a confirmation you already have, and pays for it with a month of a slot that is not reading anything.
This entire rule is unusable without calibrated RPE. If your RPE 8 drifts session to session, a flat line on the page might be a real stall or might just be your RPE scale wandering, and you cannot tell the difference from the log alone. Calibrate the scale first. The kill criterion assumes you already know what an 8 feels like on that specific lift, in that specific gym.
It also leans on a clear definition of "adding," which is where double progression does the work. On an accessory sitting in a 3x10-12 range, adding means the bottom set climbs from 10 to 11, not that the session felt harder. I have written that I stop caring about the exact rep on high-range accessory work, and that still holds inside a session. It stops holding the moment you are judging tenure: the audit needs one number recorded the same way every week, even on a lift you otherwise run by feel. Without a progression scheme that names its own ladder, you have no way to say whether a number moved or you just interpreted a decent day generously.
Diagnose before you cut
A stall is a symptom, and the first instinct, pulling the movement, treats a symptom as if it were the diagnosis. Before you cut anything, ask why the number stopped moving.
Sometimes fatigue upstream ate it. The row sits fourth on a pull day, behind a top-set deadlift and two heavy accessory lifts, and by the time the row comes up the lifter has nothing left to give it. The row isn't the problem. The slot is. Move it earlier in the session, or lighten what comes before it, and you may get three more months out of a movement you were about to discard.
And sometimes the block is the reason. Accessory numbers flatten in the back half of an accumulation phase because that is what accumulated fatigue does, and they flatten through an intensification because the attention has moved to the main lifts on purpose. Three flat sessions inside a fatigued stretch is the program working. Count the stall window against comparable weeks: week two of one block against week two of the next, post-deload against post-deload. A criterion that reads week six against week two is reading fatigue, not the movement.
Sometimes the rep range drifted without anyone noticing. A 3x8-12 lift quietly turns into a 3x8-8 habit because the lifter stops pushing past the bottom of the range once it feels comfortable there, and a range that never gets attempted at its top can never progress past it either. That is a discipline problem wearing the costume of an exercise problem.
Fix the slot. Reorder the session, redistribute the fatigue, hold the lifter to the actual top of the rep range, and only then re-run the clock on the kill criterion.
If you fix the slot and the number still does not move, cut the exercise. Do not swap it for a close cousin in the same block, a different row variation for a row that stalled, a different single-leg press for a single-leg press that stalled. That is not an audit, that is a substitution, and it carries the same structural weight problem forward under a new name while letting you feel like you did something. Cut, then go back to the reason the slot existed. If the upper back was the gap, the gap is still there, and the answer is a movement that loads it differently enough to read differently: a chest-supported row for a barbell row that stalled, not the next barbell row variation. If the slot had no gap behind it, leave it open and let the next block decide what, if anything, goes there.
The one exemption, and where it stays honest
Not every exercise on the sheet should be judged by a rising number, and pretending otherwise is where this rule gets brittle. Some work is there as insurance, not progression. Cuff work, ab work, rear delts, loaded carries for general trunk durability. You are not trying to add five pounds to a face pull every month. You are dosing it, and judging it by whether the shoulder that used to flare up in week ten of a peak has stopped flaring up at all.
This matters because a rigid progression audit, applied without exception, deletes exactly the exercises that are working best. A shoulder that goes quiet in week twelve of a peak never shows up as a progressing number on a page. An audit that only looks for rising lines will cut that movement for stalling, right as it is doing its entire job.
The fix is a declared exemption, not a loophole discovered after the fact. At the moment you add a durability movement, say out loud, or write on the sheet, that this slot is insurance and will be judged by symptom absence, not by load or reps. Cap it at two slots per block, maximum. And never grant the exemption retroactively. A movement that stalled under the normal criterion does not get reclassified as "insurance" after the fact just because you like it and do not want to cut it. Retroactive exemptions are exactly how a fourteen-movement program gets back to fourteen movements: every stalled exercise eventually gets waved through under an insurance label invented after the stall, and the sheet never shrinks.
There is a second honest complication. A rising number on an accessory tells you that you got better at the accessory. It does not, by itself, tell you the accessory is why your squat moved. Carryover to the competition lift is a real and separate question, and this post does not answer it. The stall rule governs how long a movement stays once it is in the program. Carryover governs whether it should have been chosen in the first place. A lifter can run a row for eight months with a textbook rising number and a squat that never benefited from it, and the kill criterion alone will never catch that, because it was never built to.
This is also where this post sits next to the case for accessory work generally. Accessories are not optional for raw lifters made the selection argument: you choose accessory work from bar path breakdowns, sticking points, and asymmetries, because the big three do not train the tissue that holds them up. That argument never said when a chosen accessory should leave. This is the removal condition that argument was missing.
Put the exemption and the carryover caveat aside, and the cost of skipping all of this is the same cost every time. Carry fourteen movements with no criterion on any of them, and you are not training with more coverage. You are training blind on fourteen things instead of reading six honestly. The decision space everyone spends their attention on, which exercise to add, was never the expensive decision. The expensive decision is the one nobody writes down: when it leaves.