| name | narrative-epistemic-targeting |
| description | Use when planning, drafting, or revising fiction to prevent lore dumps and control what the reader learns now: chapter openings, reveals, mystery clues, worldbuilding, exposition, progression mechanics, fan fiction context, autonomous story generation, and AI writing IDE reasoning loops. Defines a single Epistemic_Goal and filters entities, backstory, lore, and scene evidence through the reader's current mental-model update. Based on Judy Fan's audience-sensitive data visualization work. |
Narrative Epistemic Targeting
Use this skill to choose the right story evidence for the reader's current question. The core rule: a scene should update the reader's mental model on purpose.
Source Protocol
- Read
references/source-notes.md when the user asks for source grounding or wants the cognitive-science analogy.
- Use Judy Fan's audience-sensitive visualization work as craft guidance: choose the representation that helps this audience answer this question.
- Do not quote long transcript passages.
Operating Loop
Start every planning or drafting pass with:
<epistemic_target>
Reader_Current_Model:
Epistemic_Goal:
Target_Update:
Required_Evidence:
Allowed_New_Info:
Withheld_Info:
Misread_Risk:
</epistemic_target>
Field meanings:
Reader_Current_Model: what the reader likely believes before the passage.
Epistemic_Goal: the single question this passage should help answer.
Target_Update: how the reader's belief should change by the end.
Required_Evidence: concrete scene evidence needed for that update.
Allowed_New_Info: new entities, lore, backstory, or mechanics that directly serve the goal.
Withheld_Info: interesting material that belongs later.
Misread_Risk: the most likely wrong inference and how to prevent it.
Filtering Rules
Include a detail only if it does at least one of these:
- Changes danger, cost, status, relationship pressure, available choices, or reader suspicion.
- Lets the reader infer the target insight without being told.
- Prevents a likely misread that would damage the current scene.
- Pays off a prior promise or plants a necessary future one.
Withhold a detail if it is merely:
- Cool lore.
- Authorial explanation.
- Backstory without present pressure.
- A faction, technique, artifact, system rule, or family history that does not affect this scene's target update.
- A reveal that answers a later question too early.
Representation Choice
Choose the passage form that best serves the Epistemic_Goal:
direct scene: readers need to witness a choice, conflict, clue, or irreversible turn.
compressed summary: readers need continuity but not dramatized discovery.
dialogue: readers need conflicting knowledge states or social pressure.
object/image: readers need to infer history or emotion through evidence.
flashback: readers need a past cause that changes the present decision now.
exposition: readers need a rule to understand immediate stakes; keep it short and consequence-bearing.
Autonomous Writing Loop
For agentic generation:
- Define one
Epistemic_Goal.
- Select only the active story graph nodes needed for that goal.
- Draft the smallest passage that causes the target update.
- Check whether any introduced lore creates new unanswered questions.
- Defer any new question that does not sharpen the current conflict.
Self Check
<epistemic_check>
Single goal:
Reader update visible:
Required evidence present:
Unneeded lore removed:
Misread risk handled:
Future tension preserved:
Revision needed:
</epistemic_check>
If Single goal is no, split the passage or choose the highest-leverage update. If Unneeded lore removed is no, cut or defer the material.
Guardrails
- Do not starve the reader of orientation. Mystery is not the same as confusion.
- Do not answer every question the context window can answer.
- Do not introduce names, systems, or histories because they exist in canon; introduce them when the reader's current inference needs them.
- Fan fiction may rely on shared canon knowledge, but still define what this passage teaches, reframes, or withholds.