Every company keeps two cultures. One is written: the handbook, the values page, the benefits summary. The other is enacted: what people actually did, this quarter, in this building. When the two diverge, and they usually diverge, only one of them governs, and it is never the written one. Culture is not a text. It is the observed record of behavior, and the most revealing entries in that record concern rights that were available but risky to use.
The leave nobody took
The cleanest specimen comes from Semco in the 1980s, as Ricardo Semler has told it. The company granted two weeks of paid paternity leave to every new father, decades ahead of its market. Not one of them took it. The reason was not the policy, which was real and funded; it was the fear of what the boss would think. From that episode came a rule worth the whole HR literature: a benefit nobody dares use is not a benefit, it is decoration. The handbook said two weeks; the culture said zero; the culture won without a fight because no fight was ever visible. Every untaken week was a private decision by a man reading the room correctly.
The modern analogue is unlimited PTO, a handbook sentence of maximal generosity whose enacted value depends entirely on what happened to the last person who took five consecutive weeks. Written policy is an option; culture is the option’s exercise price, and the price is set by observed consequences, not by the text.
Why documents lose to incentives
Employees are not cynics; they are Bayesians. They weigh a sentence in a handbook against the evidence of their own eyes, and the eyes carry more data. The meta-analysis of employee silence by Hao and colleagues (168 samples, 63,134 employees) shows the mechanism at scale: silence tracks inhibition-type antecedents, fear above all, and spreads into worse attitudes, well-being and performance. A company can print “speak up” in three languages, but the fate of the last messenger is the version everyone reads. The same asymmetry explains why aspirational handbooks fail so quietly. No one announces disbelief in the values page. They simply behave according to the incentive gradient, and the gradient is set by what was rewarded, punished and tolerated in living memory, the enacted culture, which compounds silently while the written one gets redesigned.
When the handbook is real
The exception proves the rule by inverting it. GitLab runs handbook-first: a public document that would exceed two thousand printed pages, which the company treats as the operating system of its all-remote work, asserting that asynchronous collaboration is impossible without documentation strong enough that colleagues can self-serve answers instead of waiting for someone to wake up. Whatever one thinks of the model, notice why this handbook is not fiction: it is operational, not aspirational. People consult it to do today’s work, so when it drifts from reality the drift is discovered within hours and patched like a bug. A handbook read daily cannot stay false for long. A handbook read at onboarding can stay false forever.
The family-rhetoric connection is direct. Companies that describe themselves in kinship terms produce the most aspirational handbooks, because the documents are written in the same register as the speeches. And they suffer the widest gaps between text and behavior, because kinship language suppresses exactly the feedback that would reveal the gap: nobody files a complaint against the family.
That suggests the general repair. First, measure enacted culture directly: uptake rates of every right the company claims to offer, parental leave taken, vacation actually used, flexible arrangements exercised, upward feedback submitted, published internally with the same seriousness as revenue. A right with near-zero uptake is a finding, not a footnote. Second, make usage safe by making it observed at the top: the benefit a senior leader visibly uses is repriced for everyone below within a week, which is the cheapest culture intervention that exists. Third, write less and operationalize more; every sentence in the handbook that no process touches is a candidate for either deletion or implementation.
The audit question for any organization fits in one line: do not tell me what the handbook permits, tell me what the last three people who used its boldest clause are doing now. Their trajectory is the culture, printed in the only ink employees trust, which is consequence.
Sources
- Hao, Zhu, He, Duan, Zhao & Meng / Journal of Business and Psychology (2022). When Is Silence Golden? Meta-analysis of 168 samples and 63,134 employees: silence is driven by inhibition-type antecedents (fear, lack of safety) and predicts worse attitudes, well-being and performance. https://link.springer.com/article/10.1007/s10869-021-09788-7
- GitLab (2026). Handbook-first and asynchronous work (all-remote handbook). Company-stated practice: a public handbook exceeding 2,000 printed pages as the operational foundation of async work, on the claim that async is impossible without strong documentation. https://handbook.gitlab.com/handbook/company/culture/all-remote/asynchronous/