Define knowledge persistence requirements
This commit is contained in:
@@ -536,3 +536,40 @@ Guardrails:
|
||||
|
||||
Anti-exploit rule: rewards should come from meaningful decisions, varied
|
||||
practice, and good recovery, not from low-cost repetition.
|
||||
|
||||
## Knowledge Persistence Requirements
|
||||
|
||||
Knowledge and skill state must survive save/load, server restarts, and future
|
||||
character handoff systems without resetting learning progress or tutorial
|
||||
fatigue.
|
||||
|
||||
Persisted state:
|
||||
|
||||
- Stable knowledge profile ID linked to the player character or NPC.
|
||||
- Learned concepts by stable concept ID.
|
||||
- Practical skill experience by taxonomy domain.
|
||||
- Mastery/confidence tier per topic, separate from raw experience.
|
||||
- Failed attempts that produced useful feedback, including cause category,
|
||||
recovery action, and recent cooldown state.
|
||||
- Tutorial and contextual prompt state: seen, dismissed, repeated, snoozed, and
|
||||
next eligible display time.
|
||||
- Optional knowledge checks answered, skipped, failed, or recently displayed.
|
||||
- Teaching, observation, and shared-work learning events that granted credit.
|
||||
- Schema version, migration marker, and source build/version for future cleanup.
|
||||
|
||||
Save/load rules:
|
||||
|
||||
- The server is authoritative for writes in multiplayer.
|
||||
- Client UI may cache prompt visibility, but save data comes from validated
|
||||
server state.
|
||||
- Save enough recent action and prompt history to prevent reset-based farming.
|
||||
- Keep learned concepts stable across wording changes in the question bank.
|
||||
- Preserve failed attempts long enough to support better feedback and avoid
|
||||
repeating the same hint every few seconds.
|
||||
- Store tutorial state per character/profile, not globally across all worlds.
|
||||
- Never store private credentials, personal account data, or real-world source
|
||||
notes in player save records.
|
||||
|
||||
Persistence rule: learning state should make a returning character feel
|
||||
continuous while giving designers enough history to tune teaching, feedback, and
|
||||
anti-exploit behavior.
|
||||
|
||||
Reference in New Issue
Block a user