Names remembered across time loops

Ask what survived this reset

A time-loop character can know a name in one scene and lack it in the next version of the same evening. That is more demanding than putting a reveal in an earlier chapter. The writer must decide what crosses the reset, whose memory carries it, and whether the surviving information is enough for the next naming use. This workbench makes those decisions explicit rather than assuming that an author-known identity is known to everyone in every repeat.

Use it for an authored reset system with declared passes, observers and name facts. It audits a small continuity model, not an entire manuscript or a real person’s memory. The names and reset mechanism in the example are fictional. A stored fact is a token you have chosen to preserve in the story; the software is not measuring recall, predicting amnesia or making a scientific claim about identity.

Start with the first repeat that feels wrong. Write the naming use you want to justify: recognizing a call sign, or connecting that call sign to a full name. Then identify the parent pass and the exact information the observer retains. A character’s later discovery should explain later dialogue. It should not quietly make an earlier line possible after you have already established that the relevant words were lost.

Original fictional scene: the greenhouse that repeated dusk

The mechanical greenhouse returned to dusk whenever its invented clock-flower folded its last leaf. Its irrigation arm began the same slow turn, and everyone stood where they had been at the first bell. The apprentice and the curator carried different sealed instructions deciding what their fictional memories would keep. The flower and reset mechanism are invented for this scene, not descriptions of real botany, technology or a condition.

On the first evening, a technician introduced the call sign Moth and the full name Teren Sable. The apprentice heard both and learned that they referred to one person. The curator heard only Moth. In the clouded repeat, the apprentice retained the relationship but lost both labels. When asked to connect the technician’s two introductions, they remembered that there had been a connection, but could not name its ends.

The irrigation arm was about to water the same bed twice. The curator called for Moth to stop it. That remembered call sign was enough for the immediate request, without an unlearned full name. Hearing it again gave the apprentice one end of the retained relation. Teren later introduced the full name, making the later connection usable under the story’s rule. The characters solved the garden task with the information available; they did not pretend the apprentice had known the full connection earlier.

The writer kept a clear-sky alternative in which the apprentice retained everything. That branch needed no reintroduction for the connection, while the curator could independently lose the call sign. It was a comparison history rather than a memory the clouded apprentice could borrow. The naming requirement also mattered: calling for help needed a usable call sign; connecting two introductions needed the separate relation. The author should not demand hidden identity knowledge for a scene that only needs an ordinary address.

The naming decision creates a cooperative scene rather than a hidden-identity surprise. Remembering a relation alone is insufficient under this fictional convention, so the characters must teach each other the missing words. If the call sign were replaced with an unrelated label, the specific name fact would change; the reset rule would still need to explain its survival. If both forms were retained from the start, the scene would lose its reason for a second introduction. Choose the rule that serves the relationship you want to show, and establish it in the actual writing.

With a different name in place of Moth, decide whether you are revising the label throughout the story or adding a new form learned later. The first changes the existing form text consistently; the second needs a new form and name-fact ID with its own learn event. Similar spelling alone does not connect the two.

Separate a label from a connection

The private people ledger identifies the author-known person. Forms attach exact strings to that identity for bookkeeping. Neither ledger automatically teaches an observer anything. A name fact records the literal label the observer remembers. A connection fact records an explicit relationship between two different forms of the same author-known person. The observer’s knowledge is therefore a separate state from the author’s complete cast sheet.

In the initial example, heard-call refers to Moth, heard-full refers to Teren Sable, and same-person connects those two forms. The connection requires both name facts to be currently stored before it can support a use. Retaining same-person alone keeps a record of the relation, but leaves it incomplete. The engine never restores the missing labels just because restoring them would make the scene easier.

This is a deliberately narrow fictional convention. You might write a different story in which remembering the relation inherently preserves both words. Model that choice by retaining all three facts, rather than describing the tool’s prerequisite rule as a universal account of minds. The workbench also does not infer a hidden bearer from a remembered label or derive every possible relation transitively. Give a required connection its own explicit fact.

Use one name fact per form and one connection fact per exact unordered form pair. A reversed pair is the same proposition, so it is rejected as a duplicate. Identical visible spellings can still have separate form IDs. A coincidence in text is not a reason to merge people or transfer one observer’s memory to another person with the same name.

Draw the parent history before editing retention

Each pass has an ID, a display label and either a parent or an explicit null root. A child begins from its own parent’s final state. Two children of the same parent are siblings, even if they appear next to one another in the input. They do not inherit from each other. Reordering JSON rows does not change that relationship, and cycles are rejected before a partial history is produced.

Root seeds are optional explicit declarations of initial knowledge. They belong to a particular root and observer. A seed cannot be attached to a child as an invisible way around its reset rule. If someone learns a name after a reset, record that as an event in the child. The distinction keeps an inherited memory separate from a new lesson.

Retention is scoped to one child and one observer. All retains the parent’s stored facts; clear retains none. Only retains the listed facts that the parent actually has. Except retains the parent’s facts other than the listed exclusions. A missing policy means clear for that observer. It does not mean copy the apprentice’s policy or assume that an unspecified person remembers everything.

A requested fact absent from the parent remains absent and is listed as a gap. The rule cannot manufacture a memory that was never in its source. All also preserves stored connections that are already incomplete; it is not a command to repair them. Examine the entry checkpoint rather than assuming that a generous-looking policy guarantees effective naming knowledge.

Read the event immediately before the name is used

Events have unique IDs, an integer position, a pass, an observer, an action and a fact ID. Positions run from 0 to 10000 and cannot tie within a pass. The action learn stores the fact with a new source event. Forget removes only that fact from current memory. Use checks the state immediately before its event and does not teach the missing information.

The clouded repeat first uses same-person with both labels missing. After rehear-call, the next use still lacks heard-full. After rehear-full, the later use becomes effective because the stored relation has its two prerequisites again. None of those later events changes the first use. Read the use table in event order when deciding which line of dialogue needs a repair.

Forgetting a name can make an otherwise retained connection incomplete. The relation’s earlier source remains in the private history, while its present usability changes. Relearning a fact records a new origin for that fact; carrying it through another reset appends the relevant reset path. Those source paths explain what the model did, rather than certifying that a character’s emotional response is convincing.

If a pass contains no use events for the selected observer, the tool still shows entry and final checkpoints. It does not invent dialogue or claim a successful naming scene was tested. An absent requested retention remains visible even in that sparse case. Add the actual authored event when you want a statement about a specific naming use.

Compare a policy without adopting it

The first selected pass is also the target of the unapplied retention alternative. Choose all, clear, only or except to replace that one observer’s reset policy in a second calculation. The default alternative retains everything for the apprentice in the clouded repeat. Its first two uses can then have effective evidence, while the current history still records the missing prerequisites.

The trial can affect descendants because their parent’s final memory may change. It cannot affect an unrelated sibling by borrowing a neighboring calculation. For example, if the clouded repeat fails to relearn the words and has a later child, retaining all at the clouded reset can change that child’s starting state. The clear repeat remains a separate branch from the original evening.

Nothing in the trial control edits the policies worksheet. Compare the current and alternative event rows, decide which scene consequences you want, then edit the actual policy deliberately if you choose to adopt it. Turning the trial off removes the alternative rather than presenting an unchanged duplicate as a second recommendation. Select a child as the first pass when testing retention; roots have initial seeds, not a reset from a parent.

Changing the observer is also a real change in scope. The curator’s remembered call sign does not resolve the apprentice’s missing full-name prerequisite. The comparison contains only the two selected passes for that observer, while the private audit retains all entered histories. A successful view is not a claim that every other observer’s dialogue has been checked and cleared.

Repair the scene that needs information

There are several valid creative repairs, and they change different things. Teach the missing label before the use, keep the difficulty and write its consequence, or change the declared retention policy. A later lesson repairs later uses. Moving it earlier can repair an earlier use, but the event order and surrounding scene must actually change. The tool will not move the event for you.

Avoid using a borrowed observer state as an unspoken fix. If the curator tells the apprentice a remembered name, enter a learn event for the apprentice at that point. The curator merely having it is insufficient. Likewise, a name written on a prop needs an authored learning event if the observer reads and remembers it; possession of the prop is not an automatic connection to the true identity.

Do not force a complete connection when the story needs uncertainty. A remembered word without its bearer can be useful. A retained relation with missing labels can motivate a new introduction. Make the limits intelligible in the scene, and keep the author’s complete ledger distinct from the character’s available evidence. The software audits your convention; the story must make that convention meaningful.

Keep the working audit private and reproducible

Copying a checkpoint produces a labelled private summary with stored facts, missing prerequisites, final effective facts and the current and alternative use states. Comparing or saving checkpoints preserves their pass and observer scope. Equal display labels do not collapse different histories. Save a project to restore the editable parameters, and inspect the restored scope before using a result in a revision.

The dedicated private audit includes all inputs, observers, traces and both calculations. General result downloads may contain the same hidden naming information. No output here is automatically observer-redacted or spoiler-safe. Inspect the actual file before sharing it with a collaborator who should not know a true name, an alternative ending or another observer’s memory.

The bounded model accepts up to 30 people, 60 forms, 80 facts, 12 observers, 12 passes, 120 root seeds, 144 reset policies and 240 events. It needs two different selected passes. Conflicting IDs, unsupported actions and an oversized worksheet stop calculation instead of silently dropping rows. Keep a small reproducible history, expand it deliberately, and use the recorded source paths to understand why each naming use is or is not supported.

Does retaining a connection automatically restore its two names?

Not under this tool’s explicit fictional convention. The connection stays stored but incomplete until both name facts are present. If your story preserves the words with the relation, retain all three facts in the rule.

Can a later lesson justify an earlier use?

No. Each use checks its immediately preceding state. Move an actual learn event earlier, change the retention rule, or preserve the difficulty and its consequence in the scene.

Does the alternative policy overwrite my current history?

No. It is a separate calculation for one selected reset and observer. Descendants may change through their own parent sources; unrelated siblings do not inherit from the trial.

Is the copied checkpoint safe to give to an uninformed reader?

No automatic redaction is applied. Copy and exports are private author records and can reveal names, connections, alternative branches and other observers’ states. Inspect the actual material before sharing.