Rules
Part of April birthdays in plain terms: the 2027 view
Musicians with April birthdays, organised by date, updated for 2027
Musicians with April birthdays: how a name is corrupted in transit, why the damage is stable once saved, and why transliteration is a decision, not a step.
A performer's name is a string of characters before it is anything else, and strings break in ways that have nothing to do with the person. Half the duplicate entries in a music database are one name that survived a trip through the wrong encoding and one that did not.
What to take away
- Text is stored as numbers, and reading those numbers with the wrong table produces a name that is corrupted but still printable.
- Corruption of that kind is contagious: once saved, the damaged form is indistinguishable from a real spelling.
- Moving a name between writing systems is a separate problem with several defensible answers, none of which is reversible.
The name that came back wrong
Every stored character is a number, and turning numbers back into letters requires knowing which character encoding they were written in. Get it wrong and the text still renders, just as the wrong letters. Accented characters and non-Latin scripts turn into runs of unexpected symbols, and the result is the phenomenon known as mojibake: readable output that is not the text anyone wrote.
What makes it dangerous for a person index is that the damage is stable. Save the mangled form and it becomes the record. A later reader cannot tell whether an odd sequence is a corruption or an unusual but genuine spelling, and a search for the correct name will not find it. The entry has not disappeared; it has become a second, separate person.
How it produces duplicates
| The step | What happens | What the database ends up with |
|---|---|---|
| Import declares the wrong encoding | Accented letters become symbol runs | Two spellings of one performer |
| A name is stripped of its accents to be safe | A different, simpler string | A third spelling |
| A double conversion is applied | Symbols get converted again | A fourth, worse spelling |
| A search index normalizes but the display does not | Search finds one, the page shows another | An entry nobody can link to |
| A merge tool compares raw strings | The variants look like different people | Duplicates preserved as fact |
None of these steps is malicious and none is rare. They are the ordinary result of data assembled from several sources over a long period, which describes almost every public music database.
Transliteration is a decision, not a conversion
A separate problem starts where the name is not written in the Latin alphabet at all. Romanization turns a name from one script into another, and there is normally more than one accepted way to do it. Different systems produce different spellings of the same name, all correct under their own rules.
The consequences are worth stating plainly:
- A romanized name is a derived form. The original is the record, and it should be kept.
- Two romanizations of one name are not two people, and a matcher that compares only the Latin forms will treat them as such.
- The performer may prefer a spelling that no system produces, and that preference outranks the systems.
- A back-conversion from a romanized form to the original script is not reliable, which is why the original has to be stored rather than reconstructed.
- Sorting the romanized form does not sort the name. The March by-day pages work through why a sort key belongs in its own field.
What to do about it in practice
Store the name in one declared encoding, keep the original script, keep a separate display form and a separate sort key, and treat any transformation as a new field rather than an edit. When two entries look like duplicates, compare the normalized forms and ask a person before merging, because a merge is much harder to undo than a duplicate is to tolerate.
For a page organized by month, the entries are pointers to a description of a body of work, not to a spelling. The musicians on the January page sets out why the identifier is the durable field; a name that has been through three pipelines shows why. The April index covers what a month grouping does and does not assert, and the January index covers how to report an entry that is wrong.
Common questions
Can a corrupted name be repaired automatically?
Often, if you can work out which pair of encodings was involved and the damage happened only once. Repeated conversions lose information permanently, and a repair attempt on a name that was never corrupted will damage it.
Is stripping accents a reasonable simplification?
Only as an extra search field. As the stored name it is a decision that the person's own spelling is optional, and it merges genuinely different names that differ only by a mark.
Which spelling should a page show?
The one the person uses, where that is known and published, with the others held as variants so that a reader arriving by any of them finds the entry. This is the same preferred-form-plus-variants pattern catalogers settled on long ago.
Does this affect anything other than names?
Yes, but names are where it hurts most, because a name is the key people search by and there is no way to check a name against arithmetic or a dictionary. A corrupted title looks wrong. A corrupted name looks foreign.
