Invented name change through eras

A later name needs a route, not a historical-looking spelling

An old and a modern name can look related because an author has a persuasive intuition about their sounds. That intuition is a useful beginning, but it does not explain which parts changed, which names resisted the change or why two unrelated roots now look alike. This workbench makes an invented route explicit. You supply segmented roots, dated rule positions and checkpoints; the tool derives each form without inventing a real language behind it.

Its main task is transforming an existing small lexicon across authored eras. A naming grammar combines components to create candidates. Here, the roots stay fixed while you inspect contextual changes, exceptions and convergence. The result is a reproducible fictional construction, not an etymological discovery, a pronunciation recording or a decision that a person has adopted every form the community could derive.

Original fictional scene: two names on the choir sheet

Kavena woke from a long suspension in a fantasy city whose annual choir still sang the refrain they remembered. The tune had survived, but the printed name beside the older solo was Kabin. The director pointed out that a gardener in the present choir also appeared as Kabin. For an afternoon, the rehearsal notes treated the two entries as one returning role.

The gardener brought a seed ledger signed Kabena. Kavena wrote their own older spelling on the back of the choir sheet. The director wanted to decide which was the original person by finding the more ancient-looking name. Neither performer accepted that solution. The two records had different owners before anyone compared how their spellings had developed.

The choir’s language notebook declared a route: an authored v unit became b between a and e; then e became i between b and n; a later convention removed a final a. Under those rules, Kavena passed through Kabina to Kabin. Kabena also reached Kabina and Kabin. The later display could represent two distinct roots. It did not establish one biography or make the gardener a descendant of the returning singer.

Kavena chose to keep Kavena on their own performance credit and to sing the inherited refrain in the choir’s current form. The gardener kept Kabena on the seed ledger. The director restored two cast IDs and put a note beside the modern lyric spelling. The problem was not solved by making one person surrender a name. It was solved by separating a derived word form, a current credit and a person.

This is an original fictional scene. The suspension, community and rules are invented. The root histories describe an imagined lexical convention, not the lifespan of every person using a root. The units do not certify how an actor should pronounce the words. A supplied reading or recording would be another piece of production evidence.

If the director replaced every source record with Kabin instead, the choir would lose the distinction that explains the mix-up. If Kavena chose a completely unrelated new credit, the return could still work, but the tension between continuity and personal choice would change. In your own scene, decide who uses an inherited form and who deliberately keeps another; a derivation does not make that choice for them.

Declare the units before writing the changes

The units field is a finite inventory of exact strings. A unit might be one written character, a digraph or another symbol you define. The program does not infer phonetic values, split a name automatically or normalize different Unicode representations into an assumed equivalent. The default capital K is simply an authored token. It does not stand for a certified phonological analysis.

Each root has its own ID, a person ID, an original display and a segment array. The display is preserved independently of the joined segments. That allows an editorial spelling to remain visible while the transformation operates on a deliberately specified representation. It also makes your responsibility clear: if the representation is wrong, a clean trace does not validate it.

Two segmentations can join to the same written string. A sequence containing t and h is different from one containing a single th unit, even when both render as th. A rule targeting th affects the latter unit, not the adjacent t and h. A convergence group compares the joined display and retains the separate roots and person IDs; it does not pronounce them alike.

Order belongs to the rule, not the pasted row

Checkpoints have unique authored integer positions and labels. Rules have an era position and a same-era order. The program sorts those declarations, so reversing JSON rows cannot silently rewrite the chronology. A duplicate era/order position is a conflict. The model has no claim that era 1 means a particular historical year, and it does not interpret calendars or reconstruct missing dates.

The default working route puts shift-v and raise-e in era 1, in that order. At the middle checkpoint, Kavena becomes Kabina because the first rule creates the environment that the second needs. Kabena already contains b in that environment and reaches the same middle form. In era 2, drop-a produces Kabin from both.

An unapplied order proposal reverses the two era-1 rules without changing roots or dates. Raise-e first sees no b in Kavena, so it does nothing there. Shift-v then gives Kabena. The later deletion gives Kaben. The other root still reaches Kabina and Kabin. This comparison is another route under another declared assumption, not a correction automatically applied to the working rules.

Read the whole context and one finite pass

A rule has a nonempty from sequence, a possibly empty to sequence, adjacent left and right contexts, and explicit start and end boundaries. Empty contexts impose no adjacent-unit requirement. Boundaries apply to the outside of the complete context: with left a, source v and right e, both boundaries require the whole a-v-e environment to fill the word.

Matches are read from the pre-rule sequence. The program scans left to right and does not overlap source spans. For a-a-a transformed by aa to b, the first two units become b and the last a remains. It does not also transform the overlapping second and third units. That is a declared operational convention, not a universal account of how natural sound change works.

Replacement material is not recursively processed by the same rule. A rule changing a into aa performs one pass and returns two a units; it does not grow indefinitely. The next ordered rule can see the result. This distinction matters whenever a change creates a new environment, removes a boundary-adjacent unit or feeds a later rule.

The trace records pre-rule units, post-rule units and zero-based match sites. Inspect the segment boundaries before interpreting a site number. A position counts declared units, not bytes, written characters or syllables. The exact token array is the reliable record when an attractive joined spelling conceals an important difference.

An exception is a scoped declaration

An exception names a root and a rule, gives the exact expected pre-rule sequence, and explains the authored reason for skipping that rule. It is not a general lock on the person’s name. Later rules still apply. Potential match sites remain visible so that an exemption is distinguishable from a rule that had no matching environment.

For example, exempt older from raise-e with expected units K-a-b-e-n-a. The working route retains Kabena at the middle checkpoint, while the other root reaches Kabina. The final deletion still changes the retained form to Kaben. If the intended tradition also keeps the ending, that requires another explicit decision rather than an invisible extension of the first exception.

The expected sequence protects against stale assumptions. In the reversed order proposal, raise-e runs before shift-v and sees K-a-v-e-n-a. That no longer matches the exemption’s expected pre-rule form. The proposal becomes unavailable with a reason, while the valid working derivation remains readable. A stale exception in the working route stops that derivation instead of silently treating an outdated note as authority.

The tool does not decide whether a ceremonial explanation is convincing. Your reason is a creative declaration, not a citation to scholarship. Use the exception ledger to make a choice inspectable, then support the choice in the world and story if it matters to readers.

Convergence asks for a writing decision

The current route groups equal nonempty displays at each checkpoint. Both input roots appear in the middle Kabina group and the late Kabin group, with their separate people. The alternative route has no such display group for the default pair. Changing a rule can therefore change the ambiguity that a later scene must handle, even when the source identities stay fixed.

Convergence is not automatically a bad result. A shared later spelling can support a mistaken address, a deliberate comparison or a need for a contextual qualifier. The relevant decision is what the reader needs to distinguish in the actual scene. The tool reports the overlap rather than forcing a new name on one person or claiming that a shared spelling proves kinship.

A form can also lose every unit. The result then remains an unresolved empty derivation, displayed as the empty marker rather than a newly invented replacement. Inspect the rule that removed the last material. This model does not insert a convenient vowel, guess an irregular survivor or recover a real historical form from an empty output.

Inspect, compare and keep the right record

The main table compares the working and proposed forms for every root and checkpoint. Its unit-comparison column can detect a changed segmentation even if the joined displays are identical. Expand a route’s trace when you need the exact operation, environment effect or exception reason. The current and unapplied traces remain labelled separately.

Result cards explain the identity and scope of each form. Copy includes the root, person, checkpoint and original display, with the unapplied alternative when available. Two cards named Kabina can be saved independently because their root and person scope are preserved. Removing one saved result must not remove the other merely because their headings match.

The private handoff contains units, roots, checkpoints, rules, exceptions, routes and their differences. A project file preserves the actual input needed to reconstruct the work. Inspect either before sharing: an identity ledger may reveal story relationships or a future convention even without containing an entire manuscript. A result file is not automatically a public or anonymized document.

Keep the worksheet finite and the claims narrow

This edition accepts up to 60 units, 20 roots, four checkpoints, 50 rules and 100 exceptions. Root/checkpoint results are limited to 80. A root begins with at most 64 units, an intermediate form with at most 128. Rule arrays have small explicit bounds. An oversized intermediate stops the affected derivation rather than returning a truncated route as complete.

All rules must lie within the final checkpoint so their declared exceptions can be checked. Unsupported wildcards, automatic phonetic classes, stress rules, historical language identification and invented etymologies are outside this model. A symbol inventory is not a research corpus, and an exact successful run is not a claim of cultural authenticity.

Use the grammar workbench when the task is constructing new roots from components. Use this page when the task is understanding what declared changes do to existing roots. Use an approved reading and performance workflow when the task is how a voice actor should say the final form. Keeping those questions distinct makes the naming decision easier to explain and revise.

Does a complete trace prove that a name has a real etymology?

No. It proves the result of the supplied fictional unit rules within their bounds. Source spellings, historical attestation, pronunciation and cultural explanations require separate evidence. This tool does not invent that evidence.

Why can two different people end up with one display?

Different roots can converge under the same rules. The result retains both roots and person IDs. The overlap may matter to a scene, but it does not establish shared identity or kinship and does not force either person to adopt the derived form.

Does an exception freeze the entire name?

No. It skips one specified rule for one root when its expected pre-rule units match. Later rules still apply. If the expectation is stale, the affected route fails; a failed proposal does not erase the valid working route.

Is the order proposal already applied to my rules?

No. It computes another route with the same roots and eras. Compare its units and trace, then deliberately edit the working rule orders if that is the convention you want. An invalid proposal can be unavailable while the current route remains valid.