The distortion is not caused by dishonest people, which is why it survives every attempt to fix it by demanding honesty.
A supervisor knows a job will finish in eleven days. Asked, he says nine or ten. He is not lying. He has a plan for making up the difference, the fuller answer would require explaining a problem that is probably fixable, and he has observed that the shorter answer produces a shorter conversation. Every element of that is reasonable, and the number that leaves his mouth is wrong in a specific direction.
The distortion compounds by layer
One adjustment of that size is trivial. The problem is that the same adjustment happens at every level, in the same direction, and the errors do not cancel.
The supervisor rounds nine percent toward optimism. His manager consolidates four such reports and applies his own adjustment, for his own good reasons. The division head does the same. By the time a figure reaches somebody four layers up, it has passed through four rounds of small kindnesses, each individually invisible and all pointing the same way.
The person at the top is not receiving a slightly optimistic picture. They are receiving a systematically wrong one, and the error scales with the height of the organization. This is the reason senior leaders are so frequently surprised by things that were widely known three levels down for months.
What actually gets filtered
Not facts, mostly. Facts are hard to conceal and dangerous to misstate. What gets filtered is everything softer, and the soft material is where the useful information lives.
Uncertainty disappears first. A forecast held with genuine doubt is reported as a number, because the doubt is difficult to convey and invites a conversation nobody has time for.
Timing shifts next. The problem is mentioned after it has been contained rather than while it is still frightening, which means you hear about it at the point where your options have narrowed.
Alternatives get compressed. The option the person expects you to reject is described accurately and briefly, and brevity does the work of dismissal without anyone having to argue for it.
And attribution softens. A failure caused by one person becomes a process issue, because naming the individual would be unkind and the process framing is defensible.
You set the filter strength
The degree of filtering is not fixed. It is calibrated to you specifically, and it is set by a small number of observed events rather than by anything you have said about wanting the truth.
What people are pricing is the cost of delivering unwelcome information. That cost is inferred from what happened to the last person who did it: whether the questioning went on longer, whether the tone changed, whether the messenger acquired ownership of the problem, and whether it came up again three weeks later.
That last one deserves attention because leaders do it without noticing. A person brings you a problem and you make them responsible for solving it. This is often correct. It is also a price, and if it is charged every time, the rational response is to bring you fewer problems.
Three mechanisms that partly work
None of these eliminate the filter. Each reduces it measurably, which is the realistic objective.
Ask for the range, not the number. What is the earliest this finishes and what is the latest, if the things you are worried about happen? The second half of that question grants permission to describe the worry, which the direct question does not. It also produces a spread, and the width of the spread is information you would otherwise never receive.
Go to the source occasionally. Not to check on managers, and not as an ambush. Walk the site, sit in on the work, talk to the people doing it. What you are collecting is not a corrected number but a sense of whether the picture you hold from above resembles the situation on the ground. Announce that you do this and why, or it will be read as surveillance and will make the filtering worse.
Ask what would have to go wrong. A general invitation to raise concerns returns nothing, because volunteering a concern requires the person to decide it is worth your time. Naming the question makes answering it the expected behaviour rather than an act of initiative.
The response that costs the most
Everything above can be undone in about ninety seconds.
Somebody tells you something you did not want to hear. Your first reaction, which is largely involuntary and entirely visible, is the single most informative event in this whole system. It is observed closely, it is discussed afterward, and it sets the price for everybody who hears about it.
Nothing you say later adjusts that price. A leader can hold a meeting about the importance of candour in the morning and reset the filter to maximum by four in the afternoon with an expression, and the meeting will be remembered as evidence that the stated position and the real one differ.
Which means the useful work here is not a system, a policy, or a survey instrument. It is what your face does when the news is bad, and whether the person who brought it had a worse afternoon than they would otherwise have had.
Edited by Patrick J. Wolf, PhD