The Forgetting Curve

Forgetting Is the Default #

I used to know the code to my parents' alarm well enough that my fingers moved before my mind caught up. The body remembered. The conscious mind followed. That kind of knowledge feels permanent while you have it, as if it has crossed some invisible threshold from recall into reflex, from memory into identity. And then one day it doesn’t come. You stand at the keypad waiting for muscle memory to surface, and nothing arrives. The code still exists, unchanged. What has degraded is access. You have not lost the information all at once. You have simply stopped rehearsing it, and in the absence of rehearsal the mind reallocates space without asking permission.

Organizations do this constantly, though the consequences are harder to see because the forgetting rarely announces itself as loss. We prefer a cleaner story, one where knowledge disappears because someone leaves. The engineer resigns, the context departs with her, a strange retry delay gets interpreted as a bug and removed, and the outage returns as if to remind everyone what that delay had once absorbed. That narrative is comforting because it localizes the problem in a departure. Departures are legible. They have dates, farewell messages, and backfill requisitions. They suggest procedural remedies.

The more unsettling version is that forgetting begins long before anyone resigns.

It begins at 2 a.m., staring at a migration dashboard glowing green except for one red cell. Millions of rows transferred overnight. Most checks passed. One discrepancy small enough to ignore and large enough to notice. The room was tired. The alternative was rollback, delay, explanation, reopening work everyone had already emotionally closed. I remember squinting at it and saying “close enough,” like a mechanic kicking a tire he has already decided is fine, testing for reassurance more than truth. I told myself that if it mattered, someone more technical would object. I told myself the monitoring would catch anything serious. I told myself momentum had value. What I was really protecting was the story that we were competent and on schedule. The red cell became tomorrow’s problem, which is another way of saying it became no one’s problem at all.

The relief I felt when the room agreed was genuine. I did not recognize it at the time as a signal. Relief often marks the moment we stop interrogating something that deserves one more question.^[1]

Six months later a 0.3 percent reporting discrepancy surfaced in a quarterly review. Nothing catastrophic. Just enough to be embarrassing. The old database had been decommissioned. The timing aligned with the migration but could not be proven. We corrected forward. We labeled it “lessons learned,” which allowed us to acknowledge the issue without reopening the origin story. The organization continued. Revenue did not collapse. No one lost their job. It would be too neat to say that the red cell caused the discrepancy. I cannot prove that. What I can say is that I did not revisit my own relief until much later, when enough time had passed for the narrative to feel less fragile.

Forgetting rarely feels like deletion. It feels like efficiency.

A retry delay that looks irrational until you remember the cascading outage three years earlier. An internal tool sunset by committee because no one can articulate why it exists, only to discover later that it had quietly been preventing audit exposure. A roadmap that drops a failed initiative and reintroduces the same idea under a different label once the emotional temperature has cooled. The archive grows thinner. The narrative grows smoother.

Complex systems generate more context than they can carry. People rotate. Incentives change. What felt urgent one quarter feels excessive the next. Survival requires simplification. Not every scar deserves preservation. Not every workaround deserves immortality. If every painful lesson retained equal weight, organizations would become so cautious that they could not act. Perfect memory would not produce better judgment. It would produce hesitation layered on hesitation until nothing moved.

A ticket marked complete because reopening it would require admitting the underlying tradeoff remains unresolved. A metric definition changed quietly because the previous one no longer flatters current performance. A postmortem that records what failed but not what was believed at the time, leaving future readers with answers divorced from the reasoning that produced them. Answers survive. Reasons fade.

Answers without reasons are brittle. They appear durable until someone new asks why and no one can reconstruct the emotional landscape that made the decision feel obvious.

Earlier in my career I kept a private decision log. It was not institutional. It was a habit. I wrote down where calls came from, what constraints shaped them, which assumptions we were relying on. The habit faded. I told myself I would formalize it when things slowed down, but quiet rarely comes, and by the time it does the sharp edges have blurred. The log still exists somewhere, last modified before two reorganizations and one executive turnover. Every so often a new hire stumbles across it and thanks whoever left it behind. I do not tell them it was me. Partly because it feels self-congratulatory, but mostly because the version of me who kept it stopped for reasons that are harder to explain. It became uncomfortable to write down context that made our current decisions look less inevitable.

A two-paragraph note once prevented us from dismantling a system we barely understood. It included a crude diagram and a sentence explaining why an inelegant solution had been chosen over a cleaner one under specific constraints. It was not formatted. It did not conform to any template. It survived because someone wrote it while the reasoning was still fresh. That scrap carried more weight than the official documentation repository we had spent months refining.

Scraps matter because they preserve friction. Systems of record tend to preserve conclusions.^[2]

Around the same time as our migration, a music-streaming company operated out of the same building. I built playlists there slowly over months, organized by mood and season and the quiet of Sunday mornings. They became a map of my attention. When the company announced it would shut down, users were given three weeks to export their data. I read the email. I noted the deadline. I assumed tomorrow would resemble today. I did not export the playlists. I do not have a principled reason for that decision. Laziness is one explanation. Another is a quiet belief that platforms endure because they feel ubiquitous while you are inside them. I treated continuity as default.

The app stopped opening.

The building remained. The fluorescent lights hummed. A new logo replaced the old one. The playlists did not disappear in drama. They simply became irretrievable.

Institutions behave similarly. They repaint the lobby. They rename the initiative. They adjust the framing. The infrastructure continues to hum.

The most disorienting form of institutional forgetting is reinterpretation. A decision once framed as risk mitigation becomes, two quarters later, evidence of over-caution. A conservative hiring plan becomes lack of ambition once cash stabilizes. The story adjusts to align with current incentives. The fear that shaped the original choice dissipates as the people who felt it rotate out. What remains is behavior without justification. A delay in the code no one would defend from first principles. A process everyone follows and no one can fully explain.

There is a temptation to solve this through documentation. Write everything down. Capture rationale. Institutionalize decision logs. I have advocated for this approach. I have built templates and delivered presentations on its importance. Over time I have become more aware that documentation is also narrative control. The act of writing freezes one interpretation of events while marginalizing others. Every postmortem reflects what the organization is willing to admit about itself at that moment, authored by people whose incentives are not neutral.^[3]

Documentation is curated memory. It is necessary and incomplete at the same time.

Forgetting clears noise and erases scars in the same motion, protects morale and protects reputation, enables reinvention and repeats mistakes with better branding, and the distinction between those outcomes rarely announces itself when the decision is being made.

Someone eventually asks the same question again.

When they do, the system will either reconstruct the answer and call it iteration, or it will encounter a fragment that shortens the path. The system does not care whether we remember. That part belongs to us.