Cast distinction studio

A name is read beside other names

An individual name can be attractive and still fail in a cast. The reader encounters nearby introductions, repeated dialogue forms and chapter headings rather than an isolated list. Compare what actually appears on the page. If Arden is usually called Ave and Aven is also called Ave, different full spellings do not solve the practical collision. Enter stable IDs so that revising a spelling does not accidentally create another character.

Read the components before the combined number

The tool compares normalized grapheme edit distance for full names, short forms and supplied readings. The disclosed heuristic is (0.5 × full-form resemblance + 0.3 × short-form resemblance + 0.2 × reading resemblance) × scene weight. It is an editorial sorting aid. It is not a measured probability of confusion, a phonetic model or a visual perception study. Missing readings are marked unavailable and contribute zero to this particular total, so do not compare a well-documented pair with an undocumented pair as though the evidence were equally complete.

Weight scenes instead of changing everything

A high scene weight prioritizes people who appear together. With the names held fixed, increasing the Aven–Arden edge from one to three triples that pair’s weighted risk. Nothing about either character becomes objectively worse. You have told the tool that this pair matters more in your manuscript. Pairs without an entered edge use one. Use explicit zero only when you deliberately want a pair deprioritized, and retain a separate note if the characters later share a scene.

Fictional example: the harbor roll call

A harbor clerk calls Ave to collect a repair order. Both Aven Reed and Arden Reed step forward, and the order reaches the wrong boat. In revision, the clerk uses Arden’s established surname-and-role reference while Aven keeps the informal short form. The legal names remain unchanged. The repair works because it changes the references used in the confusing scene, rather than decorating the names with extra letters that no one says aloud.

A local repair can beat a global rename

If the author renamed Arden everywhere, the scene might become clearer but letters, family records and earlier clues would also need revision. Changing only the spoken form has a smaller continuity cost. Conversely, if the same collision occurs throughout every viewpoint, a local patch may only postpone the problem. Review the actual passages after comparing the pairs. Keep deliberate lookalikes when the story needs them, but give the reader a reliable contextual cue and test the scene without your explanatory notes.

Use this screen as a review queue

No candidate is automatically renamed or forbidden. Save the pairs that matter, compare proposed alternatives and bring a short passage to a reader. Ask which people they believe are present and which reference they use to remember each one. That observation is stronger evidence than a decimal. Approved reading strings help screen exact or near textual similarity in the supplied pronunciation notation; they do not establish whether two performed names will sound alike to every listener.

Keep visual observation distinct from edit distance

Task-specific controls show each pair in a chosen local font and accept your own visual observation with a reason and conditions. Its optional weight defaults to zero. This is subjective recorded evidence, not machine vision or a reader study. Recheck the same pair at production size and in fallback fonts before treating the observation as a reason to change an approved identity.

Run a controlled rename comparison

Use the supplied three-person example first. Aven Reed and Arden Reed share Ave as a dialogue form, while Sela Morrow has a distinct short reference. Keep the seed out of this task: nothing needs to be randomized. Enter scene weights for the two people whose exchanges matter most. Then read the pair report before touching a name. It identifies a relationship between references, not an intrinsic defect in either character.

To simulate a repair, add b | Arden Reed | Arden | ar-den to the rename table. The full identity stays fixed while the short form changes. The simulation reevaluates the proposed character against every other cast member. It reports the old and new weighted total without rewriting the source roster. This distinction matters when several people resemble the proposed replacement: a repair that helps the first pair can create a worse collision elsewhere.

Add b to the locked-ID field and run the same proposal again. The simulation must reject it. A lock represents an editorial commitment, not an extra preference that can be overridden by a better number. If the published full identity is fixed but a dialogue reference is still editable, decide that scope explicitly in your project record before revising the roster. This version treats the identity-level lock conservatively and will not change any part of that person through a repair proposal.

Now lower the reading weight to zero while leaving the names and scene graph fixed. The resulting order may change because you changed the evaluation rule. Do not describe that as the characters becoming easier to hear. It is a sensitivity check on an editorial model. A missing pronunciation and an explicitly ignored pronunciation are different evidence states, even though neither contributes to this total. Retain the unavailable labels when reporting results to another editor.

An original fictional scene: the order for the wrong boat

“Ave, take this to the north berth,” the clerk called.

Aven left the coil of rope on the desk. Arden reached the window at the same moment. The clerk, already turning to the next order, pushed the folded page toward the nearest hand.

At noon, Sela found a replacement rudder waiting beside a grain barge. The barge’s pilot had signed for it, politely and without understanding why a vessel repaired that morning needed another one. Across the harbor, Aven was preparing to leave with the damaged rudder still fitted.

“You both answer,” Sela told them.

“Most people use my whole name,” Arden said.

“Most people do not shout across three repair crews.”

The clerk retrieved the order. Aven’s name was written correctly. So was the berth. Nothing on the paper explained why the wrong person had taken it. The error lived in the exchange, not in the spelling.

Arden suggested calling him Rook, the name of a vessel he had once served on. Sela shook her head. A stevedore called Rook would arrive with the next tide. They would fix one confusion by importing another.

“Arden at the window,” the clerk said. “Ave on the quay. Will that do until the orders are finished?”

Aven nodded. Arden tried the sentence aloud, with his full given name where the clipped call had been. He disliked the extra formality, but he liked receiving the right work.

That afternoon the clerk called Arden. Only one person moved. When the north-berth order finally reached Aven, the page had not changed at all.

This is a fictional operational mistake. It does not prove that every reader will confuse these particular spellings. Its function is to separate three questions: who appears on the record, what is said in the scene, and which nearby alternative references would introduce another collision.

The repair is a decision about scope

The proposed Arden short form changes a public workplace reference while leaving family identity intact. A wholesale rename would affect earlier letters, shared surnames and later introductions. Calling him Rook might sound more distinctive in an isolated pair, but the incoming stevedore makes that solution locally wrong. This is why the tool reevaluates a proposal against the whole supplied cast instead of celebrating one improved comparison.

If the author wanted the mistake to reveal sabotage, the same dialogue repair could destroy a clue. Someone might deliberately use Ave despite knowing the approved distinction. In that version, preserve the confusion and record the character’s intent rather than treating every high score as an error to remove. The intentional-similarity table keeps a reason attached to the pair while leaving its score visible. A note explains a risk; it does not make the risk disappear.

If Arden dislikes the formal reference, the name can also carry a smaller relationship cost. A colleague might use Arden only when angry, then switch to the accepted work form when the task is urgent. Those differences belong in the address graph. Cast distinction should establish which references readers can tell apart; relationship permissions establish who may use them. Combining the two questions into one popularity ranking would hide the useful tradeoff.

How to interpret a high pair value

The full-form component is based on normalized grapheme edit distance. It does not compare the exact shapes of rendered letters, and it is not a phonetic algorithm. A narrow typeface, a low-resolution balloon or an unfamiliar pronunciation notation can change perception without changing this calculation. Where actual font or audio evidence matters, take the shortlisted pair into the lettering or rehearsal tool. Label the result as an additional observation, not a validation of the entire cast.

The scene multiplier expresses the author’s priority. A pair seen in five chapters together may warrant more attention than a pair separated by centuries. The number of shared scenes is not automatically imported or verified; you supply a weight. Avoid pretending that three means “three times more likely to confuse a reader.” It means three times the influence in this chosen sorting formula.

Do not compare a thoroughly documented cast with a sparsely documented cast and declare the lower total superior. Missing short forms or readings contribute zero to this implementation, but they may conceal the most important collision. First inspect completeness, then inspect ranking. A useful review packet identifies the known forms, the unavailable evidence, the scene assumptions and the intentional exceptions alongside the pair list.

A reader check that can change the decision

After choosing two plausible repairs, prepare two versions of the same short scene. Keep the actions, introduction order and descriptive detail fixed. Let different readers encounter the versions without telling them which one you prefer. Ask whom they believe received the order, which name they remember and where they had to reread. Record the actual response rather than translating hesitation into a precision score.

A small convenience sample cannot establish a universal naming law. It can still show that your intended distinction did not reach those readers under those conditions. If responses disagree, preserve the disagreement. Perhaps the full-form repair helps visual readers while the spoken version remains difficult. Perhaps a contextual action identifies the person so effectively that a global rename is unnecessary. Those are creative findings worth keeping.

Finish by updating the accepted roster once, then replay the comparison and the affected scenes. The simulation is not the committed edit. Retain the earlier form in the decision history so collaborators can understand why a manuscript, a character card or a recording may still show it. A cast becomes usable through consistent references and earned context, not by minimizing a similarity number at any cost.

Why does a pair remain visible when I mark it intentional?

The annotation explains why you are keeping the resemblance; it does not erase the comparison. You may need lookalikes for a mistaken-identity scene and still need clearer references elsewhere. Keep the risk, the intended mechanism and the compensating context together. Removing the pair would make a later collaborator assume it had never been reviewed rather than understand the choice.

Is a lower total always a better cast?

No. The total depends on your selected weights, supplied forms and scene priorities. Missing readings can lower it without improving anything. A harmless recurring family marker may raise it while helping the story. Use the value to order a review queue, then assess actual scenes and reader responses. Do not report it as an objective quality grade or compare unrelated projects as though they had identical evidence.

Can I compare a replacement without accepting it?

Yes. The repair table evaluates a proposed form against all other supplied people and leaves the source roster unchanged. This is especially useful when a promising local repair causes another collision. An identity-level lock blocks the proposal. After choosing a revision, update the accepted roster deliberately and rerun the comparison so the committed state, not merely the simulation, receives a final check.

What if the written names differ but the performance does not?

Supply confirmed reading strings where available, but remember that their edit distance is not a phonetic model. Take the pair into a real rehearsal and record what listeners can identify in the excerpt. A different name, a fuller reference, a response or a changed line order may help. Preserve the actual observation instead of claiming that a low spelling similarity has certified audio clarity.