How technical debt sinks a codebase

The Tyranny of Small Steps (also called creeping normalcy or the boiling-frog trap) describes a system that degrades through increments each too small to justify resistance, while the cumulative effect is catastrophic.

In the model, see the edge from shortcuts taken to detection threshold: it is weak and negative. No single shortcut moves the needle enough to set off the alarm — so the governing response never fires. The decision you actually face is never "should I ruin the codebase?" It is always "should I skip this one test?" And the honest answer, this one time, is usually yes.

While the alarm sleeps, a reinforcing loop R is compounding. Shortcuts raise code complexity, which raises technical debt, which eventually drags down delivery speed. Slower delivery widens the gap to the roadmap, which raises delivery pressure — and pressure produces more shortcuts.

The vicious part is the direction: debt makes you slower, slowness makes you more rushed, and rush makes more debt. The team digs faster the deeper it gets. Note the delays on the complexity and speed edges — the pain shows up long after the shortcut that caused it, so cause and effect feel unrelated.

Run the simulation. For a long stretch delivery speed barely moves while technical debt climbs steadily in the background. Then speed drops sharply. Nothing changed in the last week — what changed is that accumulated complexity finally crossed the point where every change has to fight the mess.

This is why post-mortems are so disorienting: people look for the catastrophic commit and never find it, because there was none. The balancing "alarm" loop B — threshold finally breached, shortcuts finally curtailed — does eventually fire, but on a long delay and only after the damage is done. A governor that activates after the collapse is not a governor.

You cannot fix this one shortcut at a time, because per-shortcut is exactly the resolution at which the problem is invisible. The leverage is to change what gets measured — lower the detection threshold by making the aggregate visible.

Track total debt as a first-class number: complexity trend, change-failure rate, lead-time drift. Give debt an explicit budget the way you budget features, and review it on the portfolio level, not the pull-request level. The point is not to ban shortcuts — sometimes the deadline really does win. The point is to stop judging them individually, where each is innocent, and start governing them in aggregate, where the truth lives.