Scratch Work vs Final Proof — The Messy Exploration That Writes the Clean Version

Textbook proofs read as if a genius paused mid-morning, wrote them down in order, and moved on. That impression is an illusion. Almost every direct proof you read was the final edit of much messier paper — a scratch sheet full of guesses, dead ends, backward reasoning, and underlined target expressions. Separating the scratch work from the published proof is one of the most useful mental habits you can build, because it means you stop feeling like a failure every time your first draft does not look like the textbook.

The split-screen

Drag the slider: on the left you see scratch work for the claim "if n is odd, then n^2 is odd." On the right you see the clean, published direct proof. Both were written by the same person, five minutes apart.

step:

The same proof, twice. Left: scratch — target-first, in shorthand, full of question marks and arrows. Right: final — hypothesis-first, in full English, with justifications. The two columns are tightly linked, but they do *not* look alike.

Five things scratch work does that the final proof does not

Starts from the target. Scratch paper almost always begins with "I want to show \dots." You write the conclusion at the top and then ask what is sufficient to get there. The final proof flips this — it starts from the hypothesis and ends with the conclusion, because that is the logically valid direction of argument.
Lives in shorthand. "odd = 2k+1", "group?", "yes", "works". These are notes to yourself, not to a reader. The final proof writes everything in complete English sentences so that a reader with no prior context can follow.
Contains dead ends. Scratch work may include an attempt you abandoned ("try substituting n = 2k - 1 instead?"), a pattern you spotted but did not use, or an algebraic detour. Dead ends belong on scratch paper; they do not belong in the published proof. The final proof is the successful path only.
Uses backward reasoning. Scratch often reads "to show n^2 is odd, I need n^2 = 2m + 1. Where could that m come from?" That is working backward from the target. Backward reasoning is not a valid proof structure — but it is an excellent way to discover a proof. The final version is forward only.
Annotates with why. Scratch paper is full of reasons written in the margins ("this is the def of odd," "I want 2 times something," "this works because of closure of \mathbb{Z}"). The final proof embeds only the load-bearing reasons and trusts the reader with the obvious ones.

Five things the final proof does that scratch work cannot

Reads linearly from hypothesis to conclusion with no detours, no question marks, no backward arrows.
Uses complete sentences with subjects and verbs.
*Introduces each new variable with a clean "let m = \ldots" phrase and never with a "suppose I let m" aside.
Includes justifications only where a reader could get lost — not everywhere.
Closes with a clear conclusion sentence and a \square or "done" mark.

Why the two documents differ: scratch work is for finding the proof. The final version is for verifying and communicating the proof. Different purposes, different formats. A finished proof written backwards would not be a proof — it would be a list of sufficient conditions. A scratch sheet written in final-proof style would be unusable as a working document.

The common mistake and its fix

Mistake: students see a clean proof in a textbook, try to produce one on first attempt, fail, and conclude they are "bad at proofs."

Fix: allow yourself two stages. Stage one is scratch — target-first, shorthand, backward if needed. Stage two is copying the successful path of your scratch into clean final form.

Scratch (stage one):

show: 4 | a² if a is even
target: a² = 4·(?)

a even → a = 2k
a² = 4k² = 4·(k²)  ←  there it is, k² is the ?
take m = k², a² = 4m, done

Final (stage two):

Suppose a is even. Then a = 2k for some integer k. Squaring, a^2 = (2k)^2 = 4k^2. Let m = k^2, which is an integer. Then a^2 = 4m, so 4 \mid a^2. \square

Same proof, two documents. The scratch was built backward from the target; the final version reads forward from the hypothesis. The time between writing stage one and stage two: under a minute.

When a proof feels hopeless — almost always a scratch-work problem

If you sit down to prove a statement and immediately try to write a clean final proof, you will usually freeze. The blank page wants a first sentence, and the first sentence of a clean proof is "Suppose \ldots" — but you have no idea yet what comes next, because you have not explored the problem.

The fix is to refuse to write a clean proof until you have scratch work that produces one. Let scratch be ugly, tentative, target-first, shorthand, full of detours. Only once the scratch reveals a viable forward chain do you sit down to write the clean version. Most "I can't prove this" moments are actually "I skipped scratch" moments.

What readers see vs what writers do

Published proofs are misleading because they hide their own history. Read enough of them and you start to believe proofs spring fully-formed from mathematicians' heads. They do not. Every working mathematician has notebooks full of scratch work that will never see publication. The clean proof you read is the summary; the scratch paper was the work. Once you have internalised this, you stop comparing your first draft to the final draft in a textbook — and you start treating your own scratch sheets as evidence of a working proof, not of a failure to produce one.