Console · Migration Tool
The checks
Last updated 18 November 2026
The checks
After the load, every check is first tried on something known to be wrong, which it must catch. A check that does not catch it gives no verdict rather than a pass. Every count is stated out of its total, never on its own.
| Check | Shown as | Passes when | Tried first on |
|---|---|---|---|
| PARITY | Records | Every row loaded is identical on the destination, dates included. A row kept at its own id also keeps its own external id, where the new site keeps it, and PARITY compares that too. Rows loaded before release 0.6.0 never had theirs put back, so they differ until a final sync loads them again. | A row with one value changed compares unequal. |
| COVERAGE | Source records accounted for | Every record the source holds is loaded, matched to one the destination already had, or left behind with a reason. | A record in none of those is counted as missing. |
| INVENTORY | Everything the source holds | Everything the source holds is accounted for, item by item and kind by kind: pages, partials, layouts, forms, queries, notifications, policies, asset files, constants, tables, people and records. Each is on the destination, left behind with its reason (an asset file unticked on Transform, Assets, so it is not copied, for one), the platform's own, for you to set up, or history. Only what sits in an Insites module's own folder (modules/insites_...) counts as the platform's own. Anything else is missing, by name, and so is any kind the tool does not copy yet, such as AI embeddings. Custom user profile types are copied: each is on the destination under its Insites name (modules/ins_profiles/...), or left behind with its reason. A migration read before release 0.8.7 did not copy them, so there they are counted as for you to set up, never missing, and the To do list has a job to recreate them in the site's code. Any other gap still fails the check. Something the load copied that is not on the destination now is named as removed since the load, with a To do job to keep the new site as it is or put it back; until then it still fails the check. A code file kept as the new site has it, which the new site does not have, counts as left behind with the reason given when it was kept. A count the old site could not give, such as its constants when they could not be read, fails by name: whether there are any was not checked. | The same count against the destination with one file taken away names exactly that file. |
| USERS | People | Every person is on the destination with their external id and created date. Someone who had an account there before the migration, such as whoever set it up, is given the created date they have on the source; if the site will not change it, that is said on its own line and does not fail the check. Anyone else who differs can be accepted as they are on the Result step, with a reason. A person missing from the destination cannot be accepted. | A person missing from the destination is reported by email. |
| PROFILES | People's details | Every person's profiles are on the destination with the same values. Each kind of user profile is counted on both sides, under its old name when the move renamed it, so a kind that did not come across shows as, for example, 34 on the source and 0 on the destination. | A profile with one value changed compares unequal. |
| OWNERS | Who each record belongs to | Every record still belongs to the same person. Each row's owner, and every value in a column that names a person by number (such as user_id), names the same person on the destination. People get new numbers on Insites, so they are compared by email or external id, never by number. A row owned by someone the migration does not copy must arrive with no owner, and is counted as such. The plan names those owners before anything loads. | A made-up number on the destination must not name the source's person. |
| PASSWORDS | Passwords | Each person the load created carries the password they have on the source. If fewer could be checked than the load created with a password, the check says how many and fails; a person the load created who is not on the destination is named. | A scrambled password with one character changed compares unequal. They are compared in memory and never read out. |
| FILES | Files open | Every uploaded file a row names answers on the destination. A file in a column the source keeps private opens only through a signed address, as on the source, and the line says how many were private. A private file that opens without a signature on the destination fails, by name. A file the old site cannot serve either, at its plain or its signed address, had nothing to copy: the check accepts it, names it under Accepted with that reason, and leaves it out of the count. | A made-up path must not answer. |
| ASSETS | The site's own files | Every asset file of your own (styles, scripts, images) is on the destination at the size the platform records. A file whose size was not recorded, such as a Webflow file, whose listed size is often wrong, is checked by presence, and the line says how many. A file you unticked on Transform, Assets is not copied and not looked for: the check names it with its reason, and it never fails the check. Each file the migration copied must also answer where the pages ask for it (a file that does not answer is tried twice, for up to 20 seconds each, and then named as not answering, so one file that hangs cannot use up a whole step; a file it has just been given can be slow to serve), and be served with the type its extension needs: a .js file as application/javascript, a .png as image/png, and so on. A font is accepted under any of its common types, such as application/x-font-woff, because a browser uses a font whatever its type. A file served as the wrong kind fails, with what it is served as, because a browser ignores it with no error on the page: a script served as an image never runs. Copy again sends it again with the right type. A file whose extension has no fixed type, such as a .glb, is counted on its own and its type is not checked. If the destination's file storage cannot be reached at all, the check says so and fails. | A size one byte larger compares unequal. A script served as an image fails the type check, and a picture served as one passes. |
| SCHEMA | Databases and fields | Every table the load carries a definition for is on the destination with its label and groups, and every column with its type and every setting: its label, form control, dropdown options, whether it is required, its group, order and tooltip, what it points at and its default. An upload column's settings count too, so a column whose files are private on the source and public on the destination is named. | The same comparison with one column's label changed reports exactly that column. |
| REFERENCES | Links between records | Every reference a row holds names a row that is on the destination, in the table it points at. Runs where the plan numbered the rows, such as a Webflow collection's items, and rewrote each reference to the new number. | A made-up id is not found. |
| SOURCE | Records, read again from the source | The source, read again at the check, matches the destination row by row and field by field, each row found by its id on the source. PARITY compares the destination with the copy the plan was built from; SOURCE compares it with the source as it is now. For a source that is quick to read again, such as Webflow. | A row with one value changed compares unequal. |
| CONTENT | Pages say the same | Every page rebuilt from a website says the same on the destination as on the old site: the words a visitor reads, compared whole. A page the old site was too busy to serve counts as not compared, and fails, rather than being left out. When the old site gives no page to compare at all, the check is not checked, not failed. | A page with one word changed compares unequal, and the same words in different markup compare equal. |
| ADDRESSES | What pages load | Nothing a rebuilt page or the stylesheets it links loads still comes from the old site's own hosts, and every file a page loads from the destination answers. A file whose fingerprint (the code the page lists so a browser can check the file) could not be checked, because its contents could not be read, is named, and so is a page that loads files from the new site's storage when the tool does not know that storage's address. | A page naming one file on the old site's CDN is reported, and one naming the destination's is not. |
| SRCSET | Smaller images for phones | No page rebuilt from a website offers smaller versions (srcset) of fewer images than the old site, read from the pages Pages say the same already reads. A phone picks a small copy from srcset and downloads the full image without one. Advisory: it is reported and never decides the verdict. | A page with one image's srcset dropped is flagged, an empty srcset does not count, and an img written inside a script is not an image. |
| SEO | SEO tags | Every page that answers on both sites prints every SEO tag the old site printed (its title, its description, robots and keywords meta tags, its Open Graph and Twitter tags, and its canonical address), with the same value. The old site's own address in a value counts as the new site's. A tag only the new site prints is counted, never a failure. Advisory: it is reported and never decides the verdict. | A page whose meta description went missing and whose og:title changed is flagged, and a tag only the new site prints is not. |
| FORMFIELDS | Form fields | Every field the old page's forms ask for, a file upload included, is on the new page's form, posts somewhere, and takes a file where the old one did. Each form rebuilt in the form builder saves every field to a column of its table. Read from the pages Pages say the same already reads. A field the new form adds, and the platform's hidden fields, are not counted against it. | A careers form whose resume upload was left off is flagged, and so is a field drawn posting nowhere. |
| BOUND (advisory) | Fields that change the page | Every field of a collection the old page shows (12 characters or more) is printed from its row on the new page, so changing it in the admin changes the page. A field written in as fixed text is named, and so is a list that reads nothing from its table, which is a fixed copy that never shows a new item. A value can also be in a page's own words by chance, so a person reads each one. | A slider heading written in as fixed text, the same on every slide, is flagged; the same page printing it from its row is not. |
| PAGEIDS (advisory) | Page ids for animations | Every page moved from Webflow carries the old page's Webflow page and site ids, which Webflow's animations start from. A page carrying another page's id can keep content hidden that should fade in. | A page carrying another page's id is flagged, and a page that never had one is not compared. |
| TRACKERS (advisory) | Analytics and tracking tags | Every analytics or tracking tag the old page loads (Google Analytics, Tag Manager, Google Ads, the Meta pixel, Hotjar, LinkedIn, Microsoft Clarity) is on the new page with the same id. A Universal Analytics (UA-) id is named with a note: Google switched it off in 2023, so leaving it out may be right. | A page whose Google Analytics tag was dropped is flagged, and an id in the page's words rather than its scripts is not a tag. |
| PAGELINKS (advisory) | Links on every page | Every link a new page makes to the site's own pages opens. A link that fails on the old site too is said, and not counted: the move did not cause it. At most 400 links are asked for, and the result says how many there were. | A link to one of the site's own pages is read, and a link to another site, to a place on the same page, or to an email address is not. |
| FORMLAYOUT (advisory) | Form spacing | On every page whose form has a captcha, read by the tool's browser at 1280 and 375 pixels wide on both sites: the space between the captcha and the Submit button, and the captcha's height, are within 8 pixels of the old site's. A form whose captcha does not show is named. Up to 12 pages are read. | A phone form whose captcha sits right on Submit (0 pixels against 36) and is 84 pixels tall against 59 is flagged on both counts. |
| PHONEWIDTH (advisory) | Pages on a phone | Nothing on a page runs off the side of a 375 pixel wide phone that did not on the old site. Read on the pages with forms first, then on pages spread across the site, 12 in all. | A page 140 pixels wider than a phone is flagged, and one the old site already ran off the side by as much is not. |
| FORMCRM | Forms add CRM contacts | Each form the tool rebuilt that asks for an email address adds a CRM contact for each enquiry, with the name, email and phone number the form asks for. A form you unticked on the plan is named with your reason and not counted against it. | A form whose CRM contact leaves out the cell number it asks for is flagged, and so is a form with no contact step. |
| SENTFILES | Files sent through forms | Every file people sent through a form before the move, such as a resume, opens on the new site. A file still on the old site fails: it is lost when the old site is canceled. A file that gave no answer is counted as could not be checked. | A resume on the old site's form uploads is found as a sent file, and a website address someone typed into a form is not. |
| LINKS | Links to the old site | For an Insites site moved to Insites: nothing in the moved code or records still names the old site, and every file address the move carried over opens on the new site. It reads every page, partial, layout, email, SMS, API call, form and query, a page's redirect, each one's settings (such as a page's Open Graph image), and every text field of every record, for the old site's file storage (uploaded files and the site's own files) and its own address. A custom domain is not counted, because it moves to the new site at cutover. An address on the old site fails, and so does a file address on the new site that does not open. Each one is listed with every place it is, and the To do list gets one job listing them all. An address that gave no answer is counted as could not be checked. The tool cannot read a constant's value, so an address held in a constant is not looked for. | An address on the old site's storage is found, in a page's settings as well as its text; one on another site on the same stack (/instances/5880 beside /instances/588) and one on another site, such as an image on an outside CDN, are not. |
| TRANSLATIONS | Translations | For an Insites (previous platform) or Insites site: every translation the copied code names ({{ 'about.title' | t }}) is held on the new site, in every language the old site holds it in, so no page shows the name of a translation where the old one showed words. It reads the new site's translation files (app/translations) from its export of its code, the way insites-cli pull does. A translation the old site does not hold in any language showed as its name there too: it is named, and is not counted against the migration. | A translation held in a translation file is found, and one held in none is not. |
| HOLD | Emails and outside connections | Only on a migration whose plan held its emails and outside connections until cutover. While they are held, a note, never a failure: how many are held on the new site, and any that send anyway because someone switched them on there. Once they are turned on, when and by whom, and any still held. See Test on a live site safely. | A held email reads back with the trigger condition that never sends, and one put back without it. |
| REDIRECTS | Redirects | Every old address the plan made a redirect for answers with a redirect to its page, in one hop, whether it was written to Settings > Redirects or as a page that redirects. A copied 404 page is given the line that hands an unmatched address to Settings > Redirects, so those redirects answer instead of 404. Where the page's code decides where it goes, the address answers as it does on the old site: the same status, to the same place. | An address no redirect was made for does not answer with one. |
| CODE | Code | Every file of your code that was copied reads back identical. A file over 4 MB, which the Console proxy will not return, is read whole from the site's export of its code and compared like any other, and the line says so. Only a file the export cannot return either has just its settings compared, and is counted as could not be checked: it fails the check, and is never counted as matching or as different. | A file with one character added compares unequal. A partial whose file name starts with an underscore matches when its address differs only by that underscore. |
| CONSTANTS | Settings and keys | Each constant copied holds the source's value, compared by fingerprint, and none of the keys issued to one site, such as the Console sign-in key, holds the old site's value. A setting every site on the same Console shares, such as the Console address, may rightly hold the same value on both, and is not flagged. If the source's constants could not be read when it was captured, the check fails and asks for the source to be read again. A constant you chose that neither site holds is said on its own line, not counted as matching. | Two different values give different fingerprints. |
| SITE (advisory) | Pages answer the same | Every page answers the same on both sites: status, title, and no Liquid error the source does not show. Again as each test account. A page the old site did not answer, or one that answers 404 on both sites, was not compared, and is counted as could not be checked. | A stock error page cannot pass for the real one. |
| SCREENS (advisory) | Pages look the same | Every page looks the same on both sites: under 5% of what the page shows differs. The share is of the page's content, every pixel that is not its background on either side, so a page with no stylesheet reads as very different even though most of both pictures is white. A changed date is about 1%. A site with more than 200 pages has a sample of 200 photographed, spread across the whole site, and the result says how many pages the site has. A page that could not be photographed is counted as not compared, and the check does not pass. | An image with a quarter painted over must differ. |
| LOGS (advisory) | The new site's log | The new site logged no errors since the load began: during the load, and while the checks before it opened every page. Each kind of error is listed once, with how often, on which pages and in which file, and with whose it is: yours to fix, one a To do job sets right, one the old site's own log has too, or one in a module Insites provides (see the marks). A form with no reCAPTCHA keys or a partial that was not copied shows up here first. A form with no reCAPTCHA keys is expected until it is set up again. An error the migration did not cause is accepted, with its reason under Accepted: one the old site's log has too, one in a module Insites provides, the new site's own database connection dropping for a moment (the database says it lost the connection, and no page or file is named; a page's own error that mentions a connection stays yours), a form on a page that is searchable on the old site too (the site logs that each time the page is saved), and one a To do job not yet ticked sets right, such as the reCAPTCHA keys. An error is Fixed since when every time it was logged came before the tool last copied its page's code (Copy this one again, Copy all again, Copy the old site's version again, or the load), and this check opened that page on the new site and it answered with no error. It is listed with when it was logged and which copy came after, and not counted. An error logged after that copy, on a page this check did not open, on a page that still fails, or with no page named, is still counted. A change you deploy to the new site yourself is not a copy the tool knows of. The result says how many of the logged errors were fixed since, out of all of them. The check passes when every error is accepted, so a new one stands out. If the new site's log cannot be read, or was read only in part, the check says so and does not pass. | One error entry and one information line group as exactly one error. |
| ROLLBACK | Rollback | After a rollback, nothing the load wrote is live and every changed row is back as it was. | A row counts as restored only when identical to its before-image. |
SCREENS can use AI to read a pair of screenshots when the pictures differ. This is off unless the tool's owner turns it on. Each picture sent is no larger than 1568 pixels on its long side and is of a public page, taken signed out, so it shows only what any visitor sees. It is sent with the page's address and nothing else. Keys, passwords and your data.
Advisory checks inform and never decide the verdict: a page that shows today's date differs for an honest reason. Each is shown with what differs so you can decide.
When every check has run, the migration reads Verified if every check that decides the verdict passed, and Verified, with problems if any failed. Both mean the checks ran to the end; only the first means they passed. The cutover holds back its live steps while a check fails. Statuses.
A page that does not answer is asked for again, twice, a few seconds apart. One that still does not answer is a page the check could not check: it fails the check and is counted, and the check says it did not answer and why. It is never left out of the count, and two sites that both failed to answer are not the same answer. Check again once the site is back.
When a check could not measure
A check that could not read what it needs does not pass. It says what it could not read: a list the old site could not count, storage that did not answer, a log that could not be read, a page that was not photographed. Nothing is counted as matching unless it was compared. This holds for INVENTORY, PASSWORDS, ASSETS, CONSTANTS, CONTENT, ADDRESSES, LOGS and SCREENS. LOGS and SCREENS are advisory, so for them it is a note and does not decide the verdict. Fix what the check names, or wait for the site to answer, then press Check again.
A check that could not measure is Not checked: not a pass, and not a failure. That is a check that could not read most of what it compares, and found nothing that differs in what it did read. For example, the old site gave no answer for a night, so none of the 29 pages could be read; it refused the tool's requests on 24 of 26 pages; or it sent every page to its other address (with or without www), so no page was compared where it was asked. It does not count against the migration, so on its own it never makes the migration read Verified, with problems, and it is never offered Copy again. It is listed under Worth a look with what could not be read and why, and a Check again button that runs the checks again. A single item that differs keeps the check a failure, so a real difference is never hidden. This holds for CONTENT, ADDRESSES, SITE (and SITE signed in), SCREENS and every other check that counts what it could not read. A verification from before release 0.8.6 reads the same way.
When bot protection in front of the old site turns a page request away (a 403), SITE, CONTENT and the old-page reads behind LOGS read that page again through the tool's browser, the same one that takes the SCREENS pictures. Each check says how many pages it read that way. Only pages are read like this, never the old site's data, at most 40 in one verification, one at a time. A page the browser could not read either is Not checked, and the reason says who turned the tool away. If it is the old Insites hosting itself, that is on the Insites side, not the site owner's: carry on, there is nobody to ask. Otherwise ask the site's owner, or whoever runs that protection for them, to allow requests carrying the X-Insites-Migration-Tool: 1 header, then press Check again. You can carry on either way: Not checked does not count against the migration.
How the tool introduces itself. Every request the tool makes to a site sends the User-Agent insites-migration-tool/0.8.8 (+https://insites.io), with the tool's version in place of 0.8.8, and the header X-Insites-Migration-Tool: 1. The tool runs on servers with no fixed address, so a host cannot allow it by address: allow that header or that User-Agent instead. The tool only reads the old site. It never changes or deletes anything there.
One item a check could not read is said apart from one that differs: could not be checked, never does not match. A code file over 4 MB that the site's export could not return either, a page the old site did not answer, a file that could not be asked for, a page whose stylesheet could not be read. CODE, SITE, CONTENT, ADDRESSES, REDIRECTS and FILES count these on their own. When only a few items could not be checked and most matched, each still fails its check, so the migration reads Verified, with problems until it is resolved, and each is named with why on the Result step and in the Verification log. Each has a plain page of its own under Things that need you, and a job on the To do list. Look at it on the new site yourself. For a page the export could not return, Copy again will not help: open the page on the new site and check it by eye.
When a check finds a difference
PARITY and CODE show exactly what differs, not just which row or file. A row lists each field side by side: what it holds on the old site, and what it holds on the new one. A file shows the settings that differ and the lines around the first change, numbered. A check lists up to 500 of what failed and describes up to 200 field by field, and always says how many failed in all.
Dates count. A row keeps the date it was created and last changed on the source, and PARITY compares both, except for a row that had to move to a new id, where only the values are compared. A date is compared as a moment in time, so the same moment written two ways, with or without milliseconds or with a time zone such as +10:00, matches. A date with no time zone, or text that is not a date, is compared as text. USERS compares people's dates the same way.
In your words. On Result, a table puts each count from What it holds, such as contacts, companies and opportunities, beside the same count read from the new site now. It is a note and does not decide the verdict. The new site can hold more, because anything it held before the load is counted too. A count the tool could not read shows ?, never 0.
On Things that need you, the step after Result, the failing checks are a short list of problems, each with one button for all of it: Copy all again for rows, code, the rows whose file does not open and the site's own files; Accept all as missing to add your own reason for files the old site cannot serve either (the check already accepts them, so they are worth a look, not a thing that needs you); Accept as they are for people who were already on the new site. Open a problem to look through it five at a time, and copy or accept one item at a time there. Copying again writes each item exactly as the old site has it, reads it back, and says whether it now matches. What it replaced goes into the rollback record first, so a rollback still puts it back. The checks themselves stay as they were until you press Check again.
A row that is not on the new site at all, whose id a row of another table already holds there, is copied under a new id, the way the load copies a row whose id is taken, and it keeps its old id as its external id so a later load finds it. Pressing again finds that copy and does not make a second one. Copying a row again also puts back the row's own external id, which writing it had replaced, and the read-back compares the external id the way PARITY does, so a row that says it now matches also passes PARITY. Where a row now is goes into the run's record of new ids and into the migration's ledger, so the next night's sync does not list it as missing again.
By why they failed. When the items of one problem failed for more than one reason, the problem lists them by reason, each with how many and its own Copy these again: for the site's own files, not on the new site, a different size there, served as the wrong kind of file, or listed but not answering; for uploaded files, the upload failed, the old site could not be read, or they do not open on the new site; for records and code, not on the new site or different there. Each button copies only the items with that reason, so you can copy the missing files without sending again the ones that need a fix of another kind. Copy all again still copies every one. Only an operator or an owner can copy again. Roles.
Accept, with a reason. Each item on the Not confirmed list on Result has Accept, with a reason. The reason goes in the report, and an accepted item no longer holds back the final sync. The list's header counts the same items as the cards, for example “20 items in 2 checks”. The warning on Cutover names the failing half of each check, for example “19 of 27 do not open”. Only an operator can accept an item. Roles.
Point them at the copies. When links on Verify lead to files that never reached the new site, press Point them at the copies. It copies each missing file from the old site's storage, reading its whole name (a name with a space, such as “Google Logo.png”, included), points the links at the copies and checks the links again. A name written in the page's text with &, such as “Logo Design & Rebranding.webp”, is asked for as the file it names, “Logo Design & Rebranding.webp”. It works on the loaded run, with no final sync. The message afterwards says what that press did, then what is left, counted over the addresses the card counts now, for example “Copied 15 files and fixed 15 links. 4 of 28 links still do not open: the old site does not serve these 4 files either, so there is nothing to copy.” A press that finds the work already done says when it was done and by whom. It shows the success tick only when something was fixed and nothing is left. Otherwise it is a warning that names what to do: Accept, with a reason on Result, or Go ahead anyway on Cutover, with Open Result beside it. It stays until you close it. When the old site does not serve a file either, the links card says so beside it and offers Accept, with a reason rather than this button. On Before cutover, the links job lists its addresses under their own heading, with whole file names, and points at this button.
Not sure what to do about one? Press Suggest a fix under it: AI reads what that check found and says what is wrong, why, and the change, and offers one of the buttons above when one fits. Suggest a fix.
A few settings are set by the new site itself, whatever the migration writes, so CODE does not compare them: copying again could never change them. Today that is one: a form's background job priority, which Insites always runs at the default. A value that differs is always shown exactly: empty text reads “” (empty text), nothing reads (empty), and a number and the same digits as text say which is which. When every code file still listed was copied again and still differs, the problem says copying again will not help, and offers Keep the new site's version first.
Keep the new site's version
Sometimes a code file differs because your team changed it on the new site on purpose, for example a fix from QA. Copying it again would undo that fix. Press Keep the new site's version on the file, or Keep the new site's version of all on the problem, and add a reason if you like. The file's card turns green and says who kept it, when and why, and the decision goes into Activity and the report.
- It is no longer a problem. Press Check again: CODE lists the file as kept, with what differs, and does not fail on it.
- Nothing writes over it. Copy all again leaves it out, Copy this one again is not offered on it, and a final sync or a nightly sync does not write it.
- You can undo it. Press Copy the old site's version again on the file. The keep is removed and the old site's version is copied over the new one, as Copy this one again does.
How many were confirmed
The Verify result opens with two figures, each saying what it counts: records copied, out of the records on the old site, with how many were new, changed, unchanged and left behind under it; and items checked passed, out of every item checked, with how many failed, were accepted with a reason and could not be checked under it. A record is a row of data; an item is anything a check looked at, records, code files, site files, pages and links alike, so the two figures are never added together. The email and the report open with one count: how many of the items that should be on the destination were checked there and match. Each check adds how many it looked at and how many match: rows (Rows), people (People), their details (People's details), owners (Who each record belongs to), uploads (Files open), code (Code), asset files (The site's own files), constants (Settings and keys), passwords (Passwords), pages (Pages answer the same), tables and columns (Tables and columns), references (References between rows), rows read again from the source (Rows, read again from the source), and a website's pages by their words (Pages say the same), what they load (What pages load) and its old addresses (Redirects). For example: 15,120 of 15,127 confirmed: 4 do not match and 3 could not be checked.
- Always a count out of a total, never a percentage. Each line names what was checked, and its result says it in words: 26 of 26 match, 101 of 101 there, or 72 of 73 match with 1 could not be checked under it.
- Does not match and could not be checked are kept apart. An item that differs is a difference. An item the check could not read was not compared at all, and is never counted as matching or as different.
- Each item not confirmed is named, with why, under the count, with a link to its check in the Verification log.
- What stays behind on purpose is not counted. Tables and code left behind with a reason, the platform's own code, history and what you set up yourself are counted by INVENTORY instead, which says whether every item the source holds is accounted for.
- Green only when every item checked matches, everything is accounted for, and no check that decides the verdict failed. Amber when anything is short, whether it does not match or could not be checked, or a check failed without a count. Red when under 95% are confirmed. The color never decides the verdict: the checks do.
The Verification log
Checking a migration again adds a result; it never replaces one. The Verification log, the last Verify step, lists every verification as Verification 1, Verification 2 and so on, newest first, each with when it ran, who ran it and every check in full. The PDF report lists them too. The plan has the same kind of record, the Plan log, with Plan 1, Plan 2 and so on. Once a later sync has loaded onto the same destination, the earlier migration is not checked again: it points at the later sync, whose copy of the source is current.
Checks on the new site, around the cutover
Launch runs its own checks on the new site, apart from the verification. Each looks at the site the way a visitor or a search engine does, and changes nothing. Before the domain moves, it looks at the new site's own address. After the domain moves, it looks at the domain. They are listed under Checks on the new site on the Before cutover and After cutover steps. Check again runs them, the watch runs them once a day, and the technical report lists the latest result of each.
- Search engines can find it. Search engines can crawl the site. /robots.txt answers, is plain text, does not shut every search engine out of every page, and names a sitemap on the same address. The sitemap answers and is a sitemap. Every address it lists is on the same address, answers, or redirects once to a page that answers, and is not kept out of search by a robots meta tag or an X-Robots-Tag header. It lists the home page and not the not-found page, and about as many addresses as the migration put there. It opens up to 200 of them each time and says how many it opened. Before the domain moves, a site address kept out of search engines is expected, so it is shown For information, not as a failure. So is a new site that matches an old site with nothing to tell search engines: if the old site had no robots.txt, or its robots.txt named no sitemap, the new site is not failed for matching it. The check fails only where the new site lost something the old one had. While the watch runs, a Crawlable check that passed and now fails is emailed straight away.
- Old addresses redirect. Listed only when a page's or a collection item's address changed on the move, so the Redirects step added a permanent (301) redirect from its old address to its new one. Before the domain moves, it reads the new site's Settings > Redirects and its list of pages, with the new site's key, and each old address you kept ticked must be on one of them as a permanent redirect to its new address. The notes say how many of them are on Settings > Redirects. A list that holds none of the pages the redirects lead to is not counted as read. After the domain moves, it asks each old address on the domain, as a visitor would, and each must answer 301 or 308 with its new address. Every failure names the old address. The watch does not run it: press Check again once the domain has moved. If the new site refused a redirect when it was written, the To do list has a job for it.