Prepare separate exposure and response sheets, then summarize supplied spelling-recall observations with real denominators and visible missing states.
Read the naming guide ↓Static example; JavaScript enables interaction.
Default-input example. Initial lettering metrics are approximate.
c1 · Courier at the ferry desk. No included observations; percentage is unknown, not zero. Included omissions: 0; explicitly not tested: 0; excluded observed: 0. This is spelling recall under supplied conditions, not name quality or population evidence.
c1 · Registry clerk by the window. No included observations; percentage is unknown, not zero. Included omissions: 0; explicitly not tested: 0; excluded observed: 0. This is spelling recall under supplied conditions, not name quality or population evidence.
Aven Reed is summarized only for condition c1. There are no included recorded answers, so no percentage can be inferred. Missing, omitted and excluded observations remain distinct; the table does not predict what a wider population would remember.
If the name is often missed in your supplied observations, inspect exposure and context before renaming it. One character may have spoken once while another signed a prominent letter. Compare like conditions and keep the raw answers; a spelling mismatch need not mean the reader confused the identities.
c1 · Witness holding a blue folder. No included observations; percentage is unknown, not zero. Included omissions: 0; explicitly not tested: 0; excluded observed: 0. This is spelling recall under supplied conditions, not name quality or population evidence.
Orin Vale is summarized only for condition c1. There are no included recorded answers, so no percentage can be inferred. Missing, omitted and excluded observations remain distinct; the table does not predict what a wider population would remember.
If the name is often missed in your supplied observations, inspect exposure and context before renaming it. One character may have spoken once while another signed a prominent letter. Compare like conditions and keep the raw answers; a spelling mismatch need not mean the reader confused the identities.
c2 · Courier at the ferry desk. No included observations; percentage is unknown, not zero. Included omissions: 0; explicitly not tested: 0; excluded observed: 0. This is spelling recall under supplied conditions, not name quality or population evidence.
Sela Morrow is summarized only for condition c2. There are no included recorded answers, so no percentage can be inferred. Missing, omitted and excluded observations remain distinct; the table does not predict what a wider population would remember.
If the name is often missed in your supplied observations, inspect exposure and context before renaming it. One character may have spoken once while another signed a prominent letter. Compare like conditions and keep the raw answers; a spelling mismatch need not mean the reader confused the identities.
c2 · Registry clerk by the window. No included observations; percentage is unknown, not zero. Included omissions: 0; explicitly not tested: 0; excluded observed: 0. This is spelling recall under supplied conditions, not name quality or population evidence.
Aven Reed is summarized only for condition c2. There are no included recorded answers, so no percentage can be inferred. Missing, omitted and excluded observations remain distinct; the table does not predict what a wider population would remember.
If the name is often missed in your supplied observations, inspect exposure and context before renaming it. One character may have spoken once while another signed a prominent letter. Compare like conditions and keep the raw answers; a spelling mismatch need not mean the reader confused the identities.
c2 · Witness holding a blue folder. No included observations; percentage is unknown, not zero. Included omissions: 0; explicitly not tested: 0; excluded observed: 0. This is spelling recall under supplied conditions, not name quality or population evidence.
Orin Vale is summarized only for condition c2. There are no included recorded answers, so no percentage can be inferred. Missing, omitted and excluded observations remain distinct; the table does not predict what a wider population would remember.
If the name is often missed in your supplied observations, inspect exposure and context before renaming it. One character may have spoken once while another signed a prominent letter. Compare like conditions and keep the raw answers; a spelling mismatch need not mean the reader confused the identities.
Original AI-assisted guide. Updated 1 November 2026. Methods: explicit local rules and documented source records. Autonomous checks are recorded; no human expert approval is implied.
A reader writes one character's name correctly and leaves another answer blank. Before changing either name, ask what was actually shown, what was asked and which answers were recorded. One person may have appeared on a letter, in a heading and in six lines of dialogue; another may have been named once. This workbench helps preserve those conditions instead of turning a small collection of answers into a universal memorability score.
The task begins after you define an exercise and, eventually, obtain responses. The tool prepares separate exposure and response sheets, records anonymous observations and reports spelling matches within each supplied condition. It does not recruit readers, administer a study, measure their timing, establish consent or invent a result. With no observations, the rates remain unknown. An empty dataset is a valid state, not an invitation to fill the table with simulated participants.
Use the ledger to decide what to inspect next. A mismatch can point toward a spelling burden, an unclear cue, unequal exposure or a transcription mistake. It cannot, by itself, tell you that a character is forgettable or that a different cultural form would be better. The practical value is a record precise enough to revisit when you revise the scene or compare another version.
Each stimulus has an item ID, an exact target spelling and a recall cue. The ID connects responses with the intended item. The target is the form you plan to compare, and the cue identifies the story reference without displaying that target. A cue such as courier at the ferry desk can ask about a person without supplying the surname you hope to observe.
Inspect cues yourself. The software cannot determine whether a description accidentally contains the answer, gives away a hidden identity or singles out a character more strongly than the passage did. A clue that mentions an initial can change the task from unaided spelling recall into a prompted completion. That may be a legitimate exercise, but it should not be described as the same condition as an unprompted question.
The current tool accepts one to forty items. If two target spellings become identical under the selected comparison policy, it stops rather than assigning an ambiguous match to whichever row came first. This is a boundary of the ledger, not a claim that stories cannot contain namesakes. Design a different question when your real task is distinguishing two identities with the same name.
Each condition has an ID, description, planned exposure time and planned delay in seconds. The times may remain blank and unknown. They are a statement of the protocol you authored, not a timer started by the page. The descriptions should identify meaningful differences such as a revised passage or another presentation format, rather than quietly reusing one ID after the procedure has changed.
Recorded exposure and delay are separate fields on observations. They are user-reported values, not measurements the website has verified. A match to the protocol requires both planned times and both reported times to be known and exactly equal. There is no hidden tolerance. If your exercise needs an interval or a different timing criterion, document that limitation instead of treating a near value as a verified match.
You can include all observed responses while flagging deviations, or restrict the denominator to observations with matching reported times. Those policies answer different bounded questions. Changing the policy recomputes the actual denominator and keeps the excluded observations in the analysis. The tool never pools conditions into an overall winner, even if a combined percentage would look more decisive.
Completed means a nonempty response string was actually recorded. The raw string is retained, including capitalization, accents and surrounding spaces. A string that happens to be wrong is still an observation. The matching policy decides how to compare it with the target; the tool does not silently correct the reader's spelling first.
Omitted means the item was asked but no answer was recorded. Under an inclusion policy that admits that observation, the omission enters the denominator and contributes no match. Not-tested means there was no observed answer for that item in that condition. It stays outside the denominator. The ledger does not infer missing participants or create rows for people who might have been invited.
This distinction matters even in a tiny exercise. A person interrupted before answering is not automatically evidence of failed recall. Conversely, deleting every genuine omission would make the reported fraction look stronger than the recorded answers support. Keep the state tied to what you observed, and use a note to preserve the reason for a non-test or a protocol deviation.
The entry panel lets you add observations one at a time. Use a non-identifying participant token, select the condition and item, choose the observed state and transcribe the answer. Reported times and an observation note are optional. The panel does not supply a participant or a successful answer on your behalf. Adding an empty completed answer is rejected; choose omitted only when that is the observed state.
The advanced ledger exposes the same records as JSON for batch editing. That format preserves response whitespace and line breaks that a pipe-separated answer table could lose. A duplicate participant/condition/item key is rejected. Correct the existing record rather than counting two transcriptions of one response as two observations. Unknown fields and invalid state combinations also stop the analysis with the original input retained.
You can remove a row and restore the most recently removed row. Restoration is checked against the current ledger, so a new record with the same key cannot be overwritten by an old undo. A private project export retains the complete inputs; importing it restores the conditions and observations. Draft text in the one-row entry panel is not a recorded observation until you add it.
Neri gathered the response cards after a reading of her harbor scene. The courier's name appeared correctly on one card. On the next card, the same answer was blank.
She reached for a red pencil. The surname must be too difficult, she thought. It had looked elegant beside the other names in her notebook, but perhaps elegance was getting in the way.
Before she crossed it out, the reader at the window returned. “I did not get to the second question,” he said. “The station announcement started, and I had to leave the table.”
Neri put down the pencil. The empty space had not told her whether he remembered the name. It had told her only that the card was empty. Her notes needed to distinguish an unanswered question from one that was never completed.
A third reader had answered the courier question with the given name alone. Neri first marked it wrong, then noticed that her cue asked what the courier was called. In the passage, another character had used only that given name. The target on her private answer key was the full name from a brief signature near the end.
Changing the spelling could not repair that mismatch between question and target. She kept the raw answer, recorded the exact comparison result and wrote a note about the cue. For the next version, she would ask specifically for the full signature or change the target to the form the scene actually taught.
The first card was not as decisive as it had looked either. Its reader had seen the courier's name on a large heading before the passage. The other participants had not. Neri separated that exposure into another condition rather than treating the correct answer as proof of the name's intrinsic strength.
By the evening she had no dramatic percentage to put at the top of her notebook. She had a smaller set of observed answers, an explicit non-test, an ambiguous cue and a reason to revise the exercise. The name stayed on the page while she repaired the question. This is an original fictional scene. Its readers, responses and circumstances illustrate the method; they are not actual user-study evidence for any generated name.
If Neri replaced the surname with a shorter one but kept the same ambiguous cue, a given-name-only answer could still fail the full-name target. If she gave one group a heading and another only a passing reference, the exposure difference would remain. The appealing counterfactual is not simply a more familiar spelling. It is an exercise that asks a clear question under recorded conditions.
A rename can still be worth testing. Keep a new stimulus or condition ID for the revised version, preserve the earlier raw observations and state what else changed in the passage. If the character also receives more dialogue, a stronger entrance or a repeated signature, the result cannot isolate the spelling as the cause of any difference. The ledger does not calculate a causal effect or a significance result.
Apply the method to your own creative choice by returning to the scene after reading the table. Was the reader meant to reproduce the exact legal-looking signature, recognize the person when a short form appeared, or distinguish two similar voices? Those are different outcomes. This page reports a stated spelling comparison; it does not turn all three into one score.
Exact compares the raw response with the stored target without transformation. NFC plus trim applies that normalization and removes surrounding whitespace for comparison only. The lowercase option also uses the declared English-locale lowercasing behavior. All policies retain the original response in the analysis so you can inspect what changed the comparison.
These are software policies, not universal rules for how every name should be judged. They do not transliterate scripts, remove accents, infer readings or decide that a different personal-name form is an acceptable answer. A target and a response can be canonically normalized for an encoding comparison without establishing cultural equivalence between different spellings.
Select the policy before interpreting results and keep it with the export. If you change it later, explain the revised analysis rather than presenting the new number as though it were the original procedure. A correction to accidental transcription whitespace may be reasonable; it should still be distinguishable from rewriting a reader's answer into the expected form.
Each item appears separately for each condition. The card reports matching included observations, the actual included denominator, included omissions, explicit non-tests and observed rows excluded by the timing policy. When the denominator is zero, the percentage is unknown rather than zero. A zero percent with observed eligible responses is a different result from having nothing to summarize.
The fraction is rounded for display and is not a quality rating. A result based on one recorded answer remains one answer, even when its displayed percentage is one hundred. The page does not estimate population recall, promise that a name is memorable, correct for exposure imbalance or replace a broader study. Anonymous tokens also do not make a project safe to share if the notes contain identifying manuscript or participant details.
Look at the raw rows before acting on a surprising result. A wrong condition ID, a duplicate transcription or an incorrectly labeled omission can change the denominator. The tool catches supported structural problems, but it cannot know what happened in the reading room. Preserve uncertainty in a note instead of selecting a state merely because it produces a cleaner table.
The exposure sheet contains the names intended for initial presentation and the authored cues, with no author explanation added by the engine. The later response sheet omits target names and contains blank prompts. These are preparation materials, not observations. The cues are still your responsibility to inspect for accidental disclosure.
The analysis export contains target spellings, raw answers, conditions, comparison policy, inclusion decisions and item cells. It is intentionally not blind. Keep it away from participants during the exercise. The tool processes these inputs locally without a reader account, a third-party AI service or a network survey. Your exported files and the people you choose to share them with are outside that local processing boundary.
Keep the final creative decision separate from the arithmetic. You may retain a difficult name because the story teaches it well, revise a cue because it asked the wrong question, or change a scene because two people were genuinely confused. The useful outcome is not the highest fraction. It is a naming decision supported by observations whose limits you can explain.
No. With no included observations the rate is unknown. An observed eligible answer that fails the spelling comparison is different from a non-test or a missing row. The ledger does not create a denominator from invited but unrecorded people.
Transcribe the raw answer first. The chosen comparison policy can normalize a copy for matching while retaining the original. A transcription correction should be documented; rewriting an answer into the target would change the observation.
No. Planned and reported seconds are supplied fields. The protocol-matched policy requires all four values to be known and equal. A difference or missing value is disclosed, not guessed into a compliant observation.
It summarizes supplied spelling observations in one condition. Context, exposure, cue wording and sample selection can affect the exercise. The tool does not infer artistic quality, causation or population recall, and it does not combine conditions into a winner.
Unicode UAX #15: Unicode Normalization Forms — NFC concerns encoded-text normalization. Our exact/trim/lowercase options are disclosed response-comparison policies, not a validation of recall study design.
What the denominator supports
Sela Morrow is summarized only for condition c1. There are no included recorded answers, so no percentage can be inferred. Missing, omitted and excluded observations remain distinct; the table does not predict what a wider population would remember.
Return from the count to the passage
If the name is often missed in your supplied observations, inspect exposure and context before renaming it. One character may have spoken once while another signed a prominent letter. Compare like conditions and keep the raw answers; a spelling mismatch need not mean the reader confused the identities.