Recorded in the Appendix

In 2004 Hans-Jörg Uther’s revision of the tale-type index folded AaTh 1587 into ATU 927D. Nearly a century of older numbers otherwise carried across into the revised system unchanged.

The merger went into the appendix rather than disappearing inside the new code.

That small piece of catalogue maintenance preserves a distinction that incident systems erase routinely. AaTh 1587 and ATU 927D name a pattern in a classification. Neither names any particular telling that had once been filed under the old number.

A code can move while the observation stays exactly where it was.

The Aarne-Thompson-Uther index exists to make patterns findable across archives that did not collect the same performance, in the same language, on the same date, from the same narrator. Its number is an address in a comparative catalogue. Treat it as the identity of an observation and a retrieval instrument has been turned into a claim of sameness.

Accession comes before classification #

The Israel Folktale Archives was founded in Haifa in 1955 by Dov Noy. A recording there receives an accession number in the archive’s own sequence after transcription, independent of any ATU classification, and it keeps that number if a later reader reclassifies the item.

The record also holds the narrator’s name, place of origin, language, and the date and place of recording. Those are not decoration around the number. They establish what was observed and where it came from.

The archive now holds tens of thousands of tales from thousands of narrators across some seventy ethnic groups. One ATU number recurs across records that are plainly not interchangeable with each other.

The order is the design. Accession comes first because it fixes custody of a particular transcript. Classification comes afterwards because it makes that transcript retrievable alongside related material. The Finnish Literature Society’s folklore archive runs the same arrangement at a far larger scale, across millions of index cards.

Neither archive would be improved by giving every record in one class the same identity.

An incident register needs precisely that separation. A report identifier fixes the observed occurrence and its evidence. A category says which recurring structure the occurrence resembles.

A taxonomy revision can change the second field. It cannot retrospectively make several reports into one occurrence, or make one occurrence into a different source document.

The type code describes the plot, not the detail #

An indexer compares a transcript against Uther’s The Types of International Folktales by its plot skeleton — the sequence of characters and actions — rather than by whichever detail is most memorable.

A youngest son, a supernatural helper, a magic object, imposed tasks and a marriage map to ATU 550, in the Supernatural Helpers band. That match is deliberately stronger than resemblance.

A magic object or a jealous sibling turns up in dozens of unrelated types. The stated structure and the cast both have to match before the type number does any work at all.

The item then picks up another kind of code entirely. Thompson’s Motif-Index assigns one code to the magic object obtained from a grateful supernatural being and another to the quest or task sequence. ATU 550 classifies the whole plot. The motif codes classify parts that travel through many plots.

Neither can replace the other without losing the account of what actually recurred.

That is the design worth taking for an incident taxonomy. A whole-pattern category describes the incident class. Component codes preserve the mechanism, the contributing condition, the observable feature. The category permits comparison; the components let a later analyst see whether two records arrived there by the same route.

Recurrence tests the type, never the event #

Uther’s rule for a regular type is direct. It must be documented among at least three distinct ethnic groups, or across a long span of time.

That threshold prevents a new number for every improvisation recorded in one village or one collector’s notebook. It is a rule about whether a pattern exists across observations. It says nothing whatever about whether any two observations are the same one.

The shape of a code is no shortcut around it either. ATU 510A, Cinderella, and 510B, Catskin, sit beside 510 by convention, and the letter does not make either an inferior dependent of the other. Each is independently regular, with its own documentation.

And some records contain more than one type. A quest followed by a marriage-trial plot is recorded as a combination of independently regular types, not melted into a hybrid code. Other material gets a miscellaneous designation where a loose theme will not support a reproducible plot skeleton.

The catalogue keeps those distinctions because a clean code is worth less than a false claim about what recurs.

Where a mechanical rule does not help #

Folktale classification has a boundary a production taxonomy may not share, and it runs the direction people expect it to run.

The indexer judges whether a plot skeleton matches a type, whether a divergence recurs, whether a combination is present. The result is partly human judgment. A production rule assigns its category mechanically, from fields, with no judgment involved anywhere.

That difference does not rescue the production record from its own history.

A rule change still changes which label appears in a series, and it changes it for every record the rule touches, silently, on the day it ships. So the record has to preserve the observation identifier, retain the old assignment, state the taxonomy version, and make the revised assignment visible as a revision.

The AaTh 1587 change stays interpretable a generation later for one reason. Somebody wrote down that it had happened.

The same discipline separates a duplicate report from a shared category. Two reports of one outage or one control breach may properly collapse into a single record, but only where their event lineage establishes one occurrence. Two unrelated reports carrying the same category remain two observations.

Their shared code is the reason to compare them. It is not permission to delete one.

An accession number tells an archive which telling it is holding. An ATU number tells a reader which pattern that telling joins.

An incident ID tells a system what happened once. A category code tells it what might happen again.