---
title: "The Weight of Time in Systems That Don't Forget"
description: "A database without TTL is a promise to carry everything forever. Every decision not to delete is a decision to carry the past."
author: "Anderson Henrique"
date: "2026-05-09T19:20:00Z"
updated: "2026-05-12T14:05:52.054528Z"
category: "philosophy"
tags: ["philosophy","data","architecture","time","forgetting"]
canonical: "https://www.ntlabs.dev/en/blog/weight-of-time-systems-that-remember"
locale: "en"
---

There's a quiet idolatry among engineers for the data that never gets lost. Immutable logs, audit trails that live forever, soft delete instead of delete. We treat system memory as a virtue — the more it remembers, the more serious it looks.

But memory has a price. Every row that isn't deleted needs to be indexed, replicated, backed up, scanned for compliance, and eventually migrated when the schema changes. The cost isn't in the row — it's in the obligation the row creates. An audit log from five years ago joins every new backup. An event table that was never truncated determines the performance of queries that didn't even exist when the first event was written.

The irony is that most of this memory is never read. Logs sit in S3 waiting for an investigation that never happens. Chat sessions from 2024 occupy rows in Supabase that nobody queries anymore. And still we hesitate to delete, because "what if it's useful someday?" — the same sentence that fills basements.

Perhaps the maturity of a system is measured by what it chooses to forget. Not in the sense of carelessness, but in the sense of priority: anonymous sessions expire in 30 days, non-auditable events in 90, operational logs in 180. What remains is what we chose to carry.

There's a line from Italo Calvino that keeps coming back: "memory only matters if it holds together the impressions of the outer world and those of the inner one." The rest is weight. And systems, like people, walk lighter when they learn to let go.