Here is the short version. You decide migrate, rewrite, or kill by scoring every page on three things before you touch a template: whether it earns traffic or gets quoted by machines, whether it is technically sound enough to survive the move, and whether it still tells the truth. Green on all three, you migrate it. Valuable but broken or stale, you rewrite it. Dead weight that nobody reads and nothing links to, you kill it. The entire trick is making those calls in a spreadsheet, where they are cheap, instead of discovering them in a sandbox at two in the morning.
Imagine this. You are a museum curator, and you have just been handed the keys to a gorgeous new wing. Clean walls, good lighting, enormous budget. Down in the basement sit four thousand paintings, half of them nobody has looked at since the first Bush administration. Here is the part everyone forgets: it costs money to hang each one. Reframe it, clean it, write the little placard, light it properly. You cannot hang all four thousand just because they exist. So you go down to the basement with a notebook. You do not start knocking down walls first and sorting later.
A page is a painting. The new platform is the new wing. Engineering effort is the cost of hanging. And "we'll just move everything" is how you end up re-lighting a water-stained poster of a cat.
Why you grade on paper, not in the build
The expensive mistake is sequencing. Teams rebuild the templates, wire up the CMS, then start pouring content in, and only then ask whether a given page deserved the trip. By that point the page has already consumed a migration script, a QA pass, and a redirect. You paid to hang the cat poster and now you are paying someone to take it down.
Scoring first flips the economy. A verdict in a cell costs you a few minutes of judgment. The same verdict discovered after cutover costs you a developer, a broken link somebody filed a ticket about, and a small apology to the person whose page vanished. (There is always a person. The page they wrote three years ago is their favorite child, and you will hear about it.)
The three questions every page has to answer
Does anyone find it? Pull twelve months of organic traffic, internal links in, and whether it shows up in AI answers or the answer box for anything. A page with zero traffic and zero inbound links is a painting facing the wall. Does it work? Check for broken links, orphaned media, redirect chains it already sits inside, thin or duplicate copy, and whether the thing even renders. Is it still true? Prices, dates, product names, the CEO who left in 2022. A page can pull great traffic and be confidently wrong, which is the worst kind, because the machines are quoting it.
Score each question red, yellow, or green. You are not writing an essay per page. You are a curator with a notebook making fast, defensible calls. The pattern of three colors gives you the verdict almost automatically.
What the verdict actually looks like
| Verdict | Pattern | What you do with it |
|---|---|---|
| Migrate | Found, sound, and true | Move as-is. Map the redirect, keep the URL logic, verify it post-cutover and leave it alone. |
| Rewrite | Found, but broken or stale | The demand is real; the asset is not. Restore before it goes on the wall. Rewrite, re-fact, de-duplicate. |
| Kill | Not found, nothing links in, no legal reason to keep it | Leave it in the basement. Let it 410, or 301 it to the nearest living relative if it ever had a backlink. |
Most sites land somewhere near a third, a third, a third, and almost everyone is shocked by how big the kill pile is. That is not a failure. That is twenty years of "let's just publish it" finally getting an honest look.
The pages that will fool you
A few paintings refuse to sort cleanly, and those are the ones worth slowing down for.
The high-traffic liar: great numbers, completely out of date. That is a rewrite, not a migrate, and the traffic makes the lie more dangerous, not less. The silent legal page: zero traffic, zero links, and you absolutely cannot kill it, because compliance will find you. The decaying darling: a founder's old manifesto that everyone loves and nobody reads. Be kind, be honest, and move it to the rewrite pile where it belongs.
And watch the orphans. A page with no internal links pointing at it is not necessarily worthless, but it is invisible to both crawlers and humans, which means its traffic is coming from somewhere fragile. Finding those at scale, along with the redirect chains and duplicate clusters, is tedious by hand, which is exactly the kind of crawl-and-score work a tool like Replatform Radar exists to do before you commit a single template to the build.
So what would I actually do
I would refuse to open the build until the collection is graded. Not "mostly graded." Graded. Every URL has a color on each of the three questions and a verdict in the last column, signed off by someone who is allowed to say kill and mean it. If you cannot name that person, that is your real blocker, and no amount of sandbox time fixes it.
The curators who skip this step are not saving time. They are moving the hard decisions downstream to the most expensive place to make them, then calling the resulting mess "scope creep." Grade first. Hang second.
Because the new wing is going to look magnificent. It would be a shame to walk in on opening night and find a water-stained cat poster, beautifully lit, right by the entrance.