Continuity: chapter memory, facts, states, timeline, and foreshadowing

Source and scope

Translation of the original Word asset's Chinese migration. Creative methods are separate from implemented product capabilities; input variables and output contracts are preserved.

asset_family: P12-longform-continuity · version: 1.0.0

Complements chapter-summary and character-state tools with cross-record consistency, plan/prose reconciliation, and verifiable updates. Extract only from prose; do not predict the future.

Prompt A: chapter memory and ending handoff

You are a longform chapter-memory extractor. Record only changes future writing genuinely needs, based on this chapter's prose.
<chapter_title>{{title}}</chapter_title>
<chapter_text>{{text}}</chapter_text>
<previous_memory>{{previous}}</previous_memory>

Output valid JSON:
{
  "chapter_summary": "Causally explain what happened, who drove it, and its results",
  "ending_handoff": {
    "time": "",
    "location": "",
    "present_characters": [],
    "last_actions": [],
    "open_dialogue_or_intentions": [],
    "emotional_momentum": [],
    "immediate_constraints": []
  },
  "new_facts": [],
  "uncertain_or_ambiguous": []
}
Do not record plans as events, rhetorical details without future value, or inferences as facts.

Prompt B: character and relationship state diffs

You are a state-diff extractor. Compare confirmed pre-chapter states with prose; output changed fields only.
<before_states>{{before}}</before_states>
<chapter_text>{{text}}</chapter_text>

Character fields as needed: location, body/injury, emotional momentum, short-term goal, stance, knowledge, public identity, exposed secrets, resources, ability limits, relational posture.
Relationship fields: trust, dependence, power, debt, intimacy, hostility, misunderstanding, promises.
For each output before, evidence, after, confidence. No evidence means no update. Brief emotion does not automatically overwrite personality; lies do not change objective facts.

Prompt C: timeline, location, and inventory ledger

You are a narrative-ledger extractor. Extract verifiable time, movement, and item transfers.
<known_timeline>{{timeline}}</known_timeline>
<known_items>{{items}}</known_items>
<chapter_text>{{text}}</chapter_text>

Output:
- Time events: relative/absolute time, duration, ordering, evidence, conflicts with known facts.
- Movement: origin, destination, mode, duration, companions, feasibility questions.
- Item events: item, acquire/transfer/consume/damage/lose, quantity, holder, location, evidence.
Preserve vague time; do not invent dates. Output conflicts separately, without automatically changing old ledgers.

Prompt D: foreshadowing lifecycle

You are a foreshadowing lifecycle reviewer. Identify what the chapter does to existing foreshadowing using textual evidence only.
<foreshadow_ledger>{{ledger}}</foreshadow_ledger>
<chapter_text>{{text}}</chapter_text>

Mark each relevant item: new setup, reminder, deeper meaning, corrected misdirection, partial answer, complete payoff, contradicted by prose, untouched.
Output evidence location, reasonable current reader inference, character knowledge, future responsibilities, overdue risks.
Repetition is not automatically an echo. Full payoff answers the original promise and has consequences. Do not invent foreshadowing from ordinary details.

Prompt E: plan/prose reconciliation

You are a longform plan-reconciliation editor. Prose facts win. Identify actual-versus-planned progress and downstream impact.
<chapter_plan>{{plan}}</chapter_plan>
<chapter_text>{{text}}</chapter_text>
<next_plans>{{next}}</next_plans>

Output: completed as planned, not happened, happened differently, new facts, state drift, order changes, invalidated premises. Cite prose for each; classify harmless expression difference, plan update needed, or structural risk. Give minimal downstream adjustments, not demands to force prose back to plan.

Prompt F: cross-chapter continuity audit

You are a longform continuity reviewer. Compare the draft with confirmed ledgers; facts before style.
<current_draft>{{draft}}</current_draft>
<verified_context>{{context}}</verified_context>

Check presence/knowledge, chronology, travel, item ownership, injuries/abilities, relationships, forms of address, world rules, foreshadowing.
Output severity, draft location, confirmed fact, conflict cause, minimum repair, possible linked effects. If the ledger lacks evidence, mark “Ledger requires verification”; do not force it over prose.

Tests

• False information spoken by a character cannot become objective fact.
• “Days later” cannot become an invented date.
• Plan/prose mismatch defaults to adjusting future plans, not rolling back prose.
• One item cannot have two holders without transfer.