Resistance is evidence that something is being contested. It is not, by itself, a verdict.
Your engagement survey came back favorable. Somewhere in the same building, a change program from last year is not in force: announced, trained, dashboarded, and worked around, its predecessor workflow still running underneath like a river under a road. Both facts are true at once. Most executives are reading only the first.
Be fair to the survey, because the critique that discards it overshoots. An engagement survey measures sentiment, and sentiment is worth measuring; a workforce that reports feeling heard is telling you something real. What the survey does not measure is operational: whether a specific change landed, whether anyone below the announcing level modified it, whether it survived contact with the daily work after the mandate expired. Those are different facts, and no cross-tab recovers them. One caution on trend lines. Comparison weakens when the stakes of answering change. People answer differently in a reorganization year than in a calm one, and the instrument cannot tell you whether the mood moved or the caution did. That does not make surveys worthless. It makes them the wrong instrument for this chapter's question.
This chapter's condition needs an operational definition. Human Experience is whether the people who carry a change can actually carry it: adopt it into the real work, challenge the parts that are wrong, modify it through some channel that answers, and improve it after it lands. Carry, not accept. Acceptance is a sentiment and shows up in surveys. Carrying is behavior and shows up in operations, and the gap between the two is where deployments go to die.
So the evidence this chapter teaches you to find is behavioral, and it forms a chain. A change was announced. Someone below the announcing level raised an authorized objection or proposed a modification. A decision was made, for reasons written down. The change then landed: the old way stopped, the new way ran and kept running afterward. Not every link is visible in ordinary records, and where one is missing or cannot be inspected safely, the probe returns No reading. One complete chain is evidence about one change, not proof of a company-wide adoption system or a full Human Experience reading. What it does is separate an announcement dashboard from evidence that correction and landing occurred. That distinction is the whole problem.
The two failures look opposite
The condition fails in two directions, and they look like opposites from the top floor.
The first is the familiar one: a system that contests nearly every change. Announcements produce counter-memos. Pilots produce grievances. Every adjustment is renegotiated with every affected party, and the organization's energy goes into the negotiation instead of the work. Executives call this resistance and buy change management for it. The diagnosis skips something. People also resist sound systems, for reasons that are theirs and often reasonable: privacy, workload, incentives that punish the new behavior, distrust earned by the last three programs, timing, risk they can see and the announcer cannot. Resistance is evidence that something is being contested. It is not, by itself, a verdict on the change or on the people.
The second failure is quieter and less diagnosed: a system that carries every announcement unchanged. No objection, no modification, no friction, every program landing on schedule. Look. I have misread that silence. I once treated the absence of objection at announcement as evidence the work had changed, then learned it showed only that no objection had reached the room. That is weak evidence that the work changed. I had measured the meeting instead of the change's afterlife. Smooth execution can sit on top of silenced correction. That is the mistake to guard against.
These are directions of risk, and they coexist. One function can be contesting everything while another swallows everything, inside the same company, under the same values statement. So do not ask which kind of company you run. Ask where each risk lives.
The healthy evidence sits between the two ends, and it is specific: a change was challenged or modified by someone below the level that announced it, the reasoning behind the decision was recorded, and the change then landed and stayed. Challenge without landing is the first failure. Landing without challenge is possibly the second. Challenge, decision, landing: that sequence, found in the records, is what a working correction system leaves behind.
When green is dangerous
The most useful symptom in this chapter is a deployment that looks successful while the old workflow stays alive underneath it. The new system is live, the metrics report usage, the project closed on schedule, and the previous method keeps running: the spreadsheet maintained on the side, the workaround passed to new hires in a lowered voice. A surviving duplicate is worth investigating, though on its own it settles little. It can mean the official version is not carrying the real work. It can also mean transition overlap, a compliance requirement, poor fit, missing functionality, habit, or unclear authority over which system wins. The duplicate narrows the search. It does not name the cause, whatever the dashboard says.
Handle symptoms with discipline, because each has rival explanations. Bad output can be a data problem. A stalled handoff can be a design problem. A deleted check can be an efficiency improvement or a disaster in waiting. No downstream change after a big deployment can mean the deployment was absorbed, or that it was pointless. Symptoms narrow the search. They do not name the cause, and treating any one of them as a verdict repeats the survey mistake with better-looking evidence. The operational signals worth collecting, none sufficient alone: sustained voluntary use after the launch push ends, duplicate workflows, the depth of the exception queue, and what the receiving roles actually do with the system's output.
Two public cases mark the boundaries of what such evidence can say. The Permanente Medical Group reported that 7,260 physicians used ambient AI scribes across more than 2.5 million patient encounters in a 63-week evaluation, with physicians reporting better patient interaction and work satisfaction. One distribution detail is worth the whole report: the top third of users accounted for 89 percent of activations. That is large-scale deployment with highly uneven carrying, and the report itself calls for more research on why. Resist the inference the number invites. The lighter users did not necessarily reject the tool, fail to benefit, or share one condition. Unevenness is a finding about where to look next, and that is all it is here.
The other case is the dark boundary, and it stays strictly inside the government record. In its 2020 resolution, the United States Department of Justice said that from 2002 to 2016, pressure to meet unrealistic sales goals led thousands of Wells Fargo employees to open millions of accounts or products under false pretenses or without consent. The bank admitted that Community Bank leaders knew of unlawful and unethical sales practices and failed to take sufficient action. The record says information reached senior leadership through multiple channels while leadership minimized the problem as individual misconduct instead of the sales model. The supported lesson is narrow and chilling: information can travel all the way up while the operating model does not move. Signals arriving is not correction happening. I assign the case no grade and no framework verdict, and it needs none to make the second failure vivid.
Follow one change after the announcement
The method is chronology, and the schematic can show its shape. Suppose the disputed-renewal workflow gets an authorized change: a revised refund policy, or a new escalation rule for exceptions. The announcement has a date, and it implies an end state: from next month, cases like this go there, under that rule. The change then has an afterlife, and the afterlife is the evidence. Was the rule modified between announcement and landing, and by whom, with what authority? Did the reason for any modification get written down when it happened? Is the old escalation path still being used after the launch period ended? The schematic answers none of these. It only shows what a complete answer would contain.
Now run the chronology on something real: one announced change, from the past year or so, that touched the workflow you selected in chapter nine. Build its afterlife record from ordinary, authorized operational records: the announcements, tickets, procedure versions, and decision notes your role can properly see. Start with the promise: the announced date and the announced end state. Then the present: is the change in force, partly in force, withdrawn, or not in force at all? Then the middle, which is where the condition shows itself: whether the change was modified along the way, which role had the authority to decide that, whether the reason was written at the time or reconstructed later, and whether the workflow the change replaced is still active past the mandate. Last, from the same ordinary records, whether anyone below the announcing level raised an objection or proposed a modification, and what decision followed. Record the role, the channel, and the outcome. Do not record the person. And keep it clear of organizing: union activity, collective bargaining, and other protected concerted action are not a data-quality signal to be traced and smoothed. They are a right, and nothing in this probe reaches them.
If the trail runs dry, say so, formally. No documented change touching your workflow, no ordinary records of its path, no safely inspectable modification at role level: any of those returns No reading for this probe. Write none found and stop. Do not go looking in grievance files, personnel records, resignation histories, or protected channels to fill the gap; those records are protected for reasons senior to this exercise, and a probe that needs them has exceeded its authority. And do not ask colleagues to disclose protected events or report outside their normal channels so your page can be complete.
The artifact is one change-afterlife record: promise, present state, modifications, decision reasons, duplicate status, and the objection chain at role level. It is a probe. One change's path cannot grade your organization's Human Experience, and a single smooth landing proves as little as a single rough one. What the record shows is narrower and harder: whether, in one traceable instance, your organization could hear a correction and still land a change. That capability is what everything in this chapter turns on.
The person-side consequence, briefly, and without costumes. A Blank Collar working inside a sound correction system challenges a defect through the channel that exists for it, improves the change instead of silently absorbing it, and gets a reasoned decision back. That is not heroism and should not have to be. Needing heroism to file an objection is itself a reading of the condition.
The condition that moves when you look at it
A warning that belongs in the chapter rather than a footnote: measuring this condition can damage it. Human Experience is unusually sensitive to the act and manner of observation. A data lineage rarely changes because you traced it. People can, and how you ask often alters what you find. Ask a team whether they feel safe challenging decisions, in a meeting, with their manager present, and you have not measured their safety. You have tested it, publicly, and everyone in the room now holds one more data point about what happens to candor here. The condition moves when you look at it. How you look is part of what you find.
Hence the standing rule this chapter has already used: where an artifact and an interview can answer the same question, prefer the artifact. The change-afterlife record exists precisely because records do not get braver or more careful depending on who is asking. And hence a harder rule for managers with good intentions: do not run disclosure exercises. A well-meant session inviting people to name what went wrong, who was ignored, or when they stayed silent can identify respondents and expose them to consequences you do not control, and intent does not indemnify anyone. If your organization needs to hear dangerous truths, that need is real, and it routes through authorized protected channels built for it, not through a workshop.
Repairs, where the record shows the correction system failing, land on leadership systems and stay there: decision records that show reasons, protected channels that actually answer, response times that get measured, consequences that follow findings. None of that repair is aimed at a named employee, and no list of resistant individuals is ever a legitimate output of this condition. The moment Human Experience work produces a list of people, it has become the thing it was supposed to detect.
Landing rates from the past do not determine the next change's fate, and this book will not pretend otherwise.
Which leaves one question unexamined, and it is the loud one. Everything so far has asked whether the organization's base can carry work and correction. The remaining question is the amplifier itself. A system can make a workflow faster, more consistent, and more confident while leaving open whether the result was correct, whether it transfers beyond the cases it was tuned on, and whether it is worth scaling. Speed is visible. The other three are not. They are the next chapter.