Redundancy Is the Only Reliable Draft Strategy

No engineer designs a hospital with one generator. You design it with two, sometimes three, on the assumption that the one you are counting on will fail at the exact hour you need it most, and I did not carry this assumption into my drafting life until I lost forty pages to a corrupted file two days before a deadline.

The forty pages were not backed up anywhere except the laptop that produced them, because I had treated the manuscript the way beginning engineers treat a single load path: as something that only needs to work, not something that needs a second way of working in case the first one fails. I rewrote those forty pages from memory in three days, worse than the original, and the worseness taught me more than the loss did, because I could feel exactly which sentences I had lost the reasoning behind and was now only reconstructing the shape of.

The backup generator draft

I now write what I call a backup generator draft for any scene I consider load-bearing to the book, meaning any scene where a later chapter depends on something specific happening in it. The backup is not a full alternate draft of the whole manuscript. It is a second, shorter version of that one scene, written from a different entry point, usually starting later in the action or from a different character's awareness of events, filed away and never looked at again unless the primary version fails a reader test in revision.

Here is what this looks like in practice. A scene in my current manuscript has a father telling his daughter, for the first time, why he sold the family compound. The primary version runs long, four pages, building to the confession through an argument about an unrelated debt. The backup version is one page, starts after the confession has already begun, and simply shows the daughter's reaction to information the reader has to infer. When my writing group found the primary version overwritten and slow, I did not spend a week trying to fix four pages with the theory of what should have worked. I pulled out the backup, which had been sitting unread for two months, and it solved the pacing problem the primary version had, because it had been built on a different assumption about where the tension actually lived.

Why this is not the same as just revising

Ordinary revision starts from the failing draft and cuts or expands it, which means every fix is downstream of the same original decisions that produced the failure. A backup generator draft is written from a different starting assumption, before you know the primary version has a problem, which means it is not contaminated by the same blind spot. When the funeral scene between the two sisters I wrote about elsewhere kept reading as flat across three revision passes, all three passes were revising the same original entry point, the moment the sisters sit down to talk. My backup for that scene started five minutes earlier, at the sister who is owed money quietly counting out cash before her sister arrives, and that different entry point is what eventually made the scene work, not a fourth attempt to fix the entry point I had already tried three times.

The obvious pushback

You could say this doubles the drafting labour for every important scene, and that a writer with limited time should spend that time getting the one version right rather than writing two mediocre attempts at the same beat. I take this seriously because it is true that the backup drafts are usually worse in isolation than a fully polished primary draft would be. What redundancy buys is not two excellent options, it is insurance against the specific failure mode where your only version has a blind spot you cannot see from inside it, because you wrote it from the one entry point your instincts reached for first. A hospital's second generator is not expected to run the building more smoothly than the first one. It is expected to exist for the hour the first one does not start.

What I actually keep

I do not backup every scene. I reserve the method for the four or five scenes per book that later chapters structurally depend on, the same threshold I use for a tariff sheet or a maintenance ledger. Most scenes only need one honest attempt. The load-bearing ones get two, written far enough apart in time and approach that if the first one is quietly broken, I have a second one that was never exposed to whatever assumption broke the first.