Migration delivery
Spain Internship
Making a multilingual migration explicit: what to keep, restore and redirect.
Redirect only where the destination exists
- 1,547old and new URLs compared; 927 matched exactlyMy analysis
- 519historical URLs audited: 347 preserved, 172 notMy analysis
- 341redirect stubs in the archived source: 172 + 152 + 17Counted from source
- 198historical vacancy redirects specified, all present in sourceMy specification
- Challenge
Hundreds of addresses no longer matched.
Comparing 1,547 old and new URLs, only 927 matched exactly.
- My contribution
I diagnosed the URL problems in Spain Internship's multilingual migration, specified the redirect and URL architecture, and implemented it through AI-assisted development in Lovable. All 198 historical vacancy redirects in my specification appear exactly in the site's source code.
- Delivery and measurement
Every changed path checked.
In the migration environment, all 152 changed paths returned 200 and were self-canonical.
What this doesn't prove
I can't yet show production deployment of every rule or any traffic effect; I left in September.
- Diagnosed — Done
- Specified — Done
- Implemented in source — Done (all 198 historical vacancy redirects in the specification appear in the source code)
- Handed over — Sep 2026
A migration with more than one URL problem.
I worked on Spain Internship's multilingual migration: comparing old and new addresses, auditing historical URLs, deciding how each path should be treated, and specifying the redirect and URL architecture. The vacancy specification covers query-string links as well as language-prefixed paths, so a single description such as ‘URLs changed’ would hide different mapping problems.
Different populations, different checks.
These counts describe different checks and deliverables. I can't show how they overlap, so they are not an additive migration total or consecutive steps in a funnel.
| Population | Count and unit | Recorded finding |
|---|---|---|
| Old/new URL comparison | 1,547 old and new URLs | 927 matched exactly |
| Historical URL audit | 519 historical URLs | 347 preserved; 172 not preserved |
| Localised path corrections | 152 changed paths | 200 and self-canonical in the migration environment |
| Historical route restoration | 172 historical routes restored | Implementation count |
| Vacancy redirect specification | 198 historical vacancy redirects | All specified redirects appear in source |
The mapping decision lives in the details.
Five rows from the 21 August 2026 vacancy redirect specification show the old address, the intended destination and the decision recorded. They are a sample, not every rule and not a live redirect test.
Muted text identifies the route prefix; underlining identifies the vacancy slug. The record ID is marked in the query-string example.
Match the vacancy when the wording changes.
/?post_type=job_listing&p=24857/internships/paid-media-internship-in-a-technology-company-in-barcelona-spainOld WordPress ID link. Matched to the live vacancy by its record, even though its title and slug had changed.
The row documents the match. It does not describe the record fields or matching procedure in enough detail to reconstruct an algorithm.
More mapping decisions
Two language paths, one vacancy page.
/es/internships/paid-summer-camp-leisure-child-care-assistant-castellon-e300-400-month-accommodation-included//fr/internships/paid-summer-camp-leisure-child-care-assistant-castellon-e300-400-month-accommodation-included//internships/paid-summer-camp-leisure-child-care-assistant-castellon-e300-400-month-accommodation-includedSpanish copy of a paid vacancy.
French copy of the same vacancy. Same destination: two old URLs, one live page.
This is a documented many-to-one mapping. The example does not establish a site-wide rule for every language variant.
Change the route; keep the descriptive slug.
/es/internships/paid-inside-sales-internship-health-tech-startup-in-madrid-e500-month//internships/paid-inside-sales-internship-health-tech-startup-in-madrid-e500-monthHigh-intent paid listing (Madrid, €500/month). Its exact slug is kept.
The retained slug is visible in the mapping. A ranking or traffic effect is not established by that decision.
Retain the Dutch slug as written.
/nl/internships/strategische-administratieve-assistentstage-op-mallorca-balearen//internships/strategische-administratieve-assistentstage-op-mallorca-balearenThe only Dutch URL in the set. The Dutch slug is kept as written, not re-translated.
The path shows which text was retained. It does not establish the language of the destination page's content.
Show the specification (5 of 198 rows)
| Old URL | to | New URL | Lang | WHY |
|---|---|---|---|---|
| Old/?post_type=job_listing&p=24857 | New/internships/paid-media-internship-in-a-technology-company-in-barcelona-spain | LangEN | WHYOld WordPress ID link. Matched to the live vacancy by its record, even though its title and slug had changed. | |
| Old/es/internships/paid-summer-camp-leisure-child-care-assistant-castellon-e300-400-month-accommodation-included/ | New/internships/paid-summer-camp-leisure-child-care-assistant-castellon-e300-400-month-accommodation-included | LangES | WHYsame pageSpanish copy of a paid vacancy. | |
| Old/fr/internships/paid-summer-camp-leisure-child-care-assistant-castellon-e300-400-month-accommodation-included/ | New/internships/paid-summer-camp-leisure-child-care-assistant-castellon-e300-400-month-accommodation-included | LangFR | WHYFrench copy of the same vacancy. Same destination: two old URLs, one live page. | |
| Old/es/internships/paid-inside-sales-internship-health-tech-startup-in-madrid-e500-month/ | New/internships/paid-inside-sales-internship-health-tech-startup-in-madrid-e500-month | LangES | WHYHigh-intent paid listing (Madrid, €500/month). Its exact slug is kept. | |
| Old/nl/internships/strategische-administratieve-assistentstage-op-mallorca-balearen/ | New/internships/strategische-administratieve-assistentstage-op-mallorca-balearen | LangNL | WHYThe only Dutch URL in the set. The Dutch slug is kept as written, not re-translated. |
198 of 198 in sourceSheet 2 of 2
Same city, two languages, opposite decisions.
I decided case by case, for example restoring the German Barcelona address and leaving the Spanish one as it was.
- Retain
- Spanish Barcelona address
/es/internships-barcelona/Leave it as it was. - Restore
- German Barcelona address
/de/internships-barcelonaSpecified 301 →/de/praktikumsmoeglichkeiten-in-barcelona/ - Consolidate
- Spanish + French summer-camp pathsTwo old addresses share one specified destination.Inspect the approved pair ↑
Specified treatments, not a production response trace. The consolidation example comes from the vacancy specification; it is separate from the Barcelona decision.
Barcelona: what the record does not say
The recorded treatment differs by language: restore the German address and leave the Spanish one as it was. The example does not document the underlying evidence used to choose those treatments. It supports a case-specific decision, without supplying a rationale that can be generalised to other routes.
Apply only the validated list
- Moving the Spain site to a new structure meant changing redirects, canonicals, hreflang, internal links and sitemaps together.
- Fix each file as problems turned up, or apply only the changes from one validated old-to-new list.
- The list was approved by my manager, and I applied only the changes on it.
- If those files disagree, search engines get contradictory signals, and errors are harder to trace.
- I reviewed the mapping URL by URL across dated versions (20, 21 and 24 August). URLs that stayed the same are kept as commented lines. The redirects are recorded in source; this case does not show deployment.
Redirect only where the destination exists
- Some old vacancy addresses had no live page to point to.
- Redirect everything to the nearest page, or redirect only where the destination was live and confirmed.
- I told the implementation to add only the 198 rows whose destination was active and confirmed in the current sitemap, and left five records unmapped.
- A redirect to a missing or irrelevant page swaps one problem for another.
- The 198 rules were implemented in source and match my specification row for row. Five records stayed unmapped on purpose. Production behaviour is not shown.
From URL decisions to source code.
The implementation route was Lovable, AI-assisted development. My responsibility covered the diagnosis, the redirect and URL specification, and implementation through that platform. I can't split individual code changes between generated output and manual edits, or claim a separate developer contribution.
Implementation limits
Source that matches the specification is an implementation artefact. It doesn't show that every rule reached production or that every old address was tested there.
Checked in the migration environment.
- 198 historical vacancy redirects
- The documented mapping decisions.A specification is not a runtime test.
- 198 of 198 in source
- All specified historical mappings appear in source.Source correspondence is not deployment evidence.
- All 152 changed paths
- 200 · self-canonicalDestination-page checks, not tests of every old redirect.
- Not established
- Production deployment of every rule
- Production behaviour of every historical redirect
- Traffic or ranking effects
Validation limits
These checks cover the changed paths in the migration environment. They don't show production behaviour for every historical vacancy redirect, and neither source correspondence nor destination checks measure a search outcome.
What I left out.
English and Spanish paths were preserved by design; only displaced localised paths became one-hop redirects. Of the 341 redirects in source, 17 are parent and hub routes. Five vacancy records stayed unmapped on purpose: unpublished WordPress leftovers with no valid, indexable destination. Thirteen further rules were approved after the 21 August checkpoint and are not part of the 198.
Where the evidence stops.
I implemented the work in August 2026 and handed it over in September. My record ends at handover: no verified production deployment of every rule, and no measured traffic effect.
I count before I decide, make each keep, restore or redirect call on evidence, refuse a redirect with no valid destination, and write the decisions precisely enough that an engineer, or an AI coding tool, can build them without guessing. The limit is real: I can show the specification and the source, not production.
Reflection
Translating an address can break it as easily as fixing it.