How to Revise a Lyric Without Killing It

Decide before you touch anything how much you are allowed to change — a light pass might be one line in ten; a rebuild might be half the song. Then treat that number as a ceiling with a floor of zero. Nothing about setting a budget obliges you to spend it. The failure mode is editing to feel busy: you sat down to revise, so you revise, and a working line gets replaced by a slightly different working line minus the accident that made it alive.

We learned this the hard way — our own refine engine originally told itself the amount changed "must match" the intensity the writer picked, and it dutifully sanded strong songs to hit the number. The fix was one sentence: changing a line that already works is a failure, not a win. Adopt that sentence.

Before a revision pass, mark the lines that are off-limits — the ones that made someone look up, the hook, the line you would defend in an argument. Physically mark them: highlight, brackets, anything. The point is not sentimentality; it is that mid-revision you will not reliably remember which lines were load-bearing, because by the fourth pass everything sounds the same to you.

And when you revise again — a second pass on the revised draft — carry the locks forward. The sneakiest revision damage happens on pass two, when the protections you set on pass one silently lapse and the line you saved last week gets rewritten this week.

Real revision is local and uneven. Nobody improves a song by changing 30% of every section; you improve it by rebuilding the dead verse and leaving the great one alone. Rank your sections honestly — which one would you skip if it were someone else's song? — and concentrate the whole change budget there.

The exception that outranks everything: a singability break (a line you physically cannot deliver in one breath, a stress landing on the wrong syllable) gets fixed wherever it appears, first, before any taste-level edit. A clumsy line is a defect; a plain line is a choice.

After a pass, do the accounting: put the old and new versions side by side and count what you actually kept, changed, and cut. Not what you meant to do — what the page says. Writers systematically misremember their own edits in both directions: "I barely touched it" after gutting the bridge, and "I rewrote the whole thing" after swapping six words.

The honest diff answers the only two questions that matter: did the weak part actually get stronger, and did anything I loved get lost in the churn? If a locked line went missing, restore it — deliberately deciding to break a lock is fine; discovering you broke one by accident is the exact failure the diff exists to catch.

Some drafts are finished before you revise them, and the craft is recognizing it. The signs: your edits have started trading one adequate word for another; you are changing lines to feel ownership rather than to fix a defect; the version from two passes ago keeps sounding better than the current one. That is the draft telling you it is done.

We went as far as teaching our refine engine to refuse: at light intensities it may hand a strong lyric back essentially unchanged and say so. Give yourself the same permission. "This is already strong — ship it" is a legitimate result of a revision session, and the writers who can say it keep more of their best lines than the ones who can't.

When you genuinely can't tell whether a draft needs another pass, get a cold read — run it through the Crucible and let an adversarial panel tell you whether the weakness you suspect is real before you start cutting.