How your choices change the ending in a StoryNova story
Inside the StoryNova engine: metrics, goals, beats, timers, memory and endings. Why the narrator cannot fake an ending and how the recap shows what mattered.
Last verified 2026-09-10
Most AI story apps let the model decide whether you won. StoryNova does not. This page explains, without jargon, what the engine tracks and how it decides.
Two roles: the narrator and the engine
The narrator is a language model. It writes the prose and the dialogue, and after each turn it proposes changes: trust went up, a flag was set, the chapter should end. The engine is ordinary code. It checks each proposal against the story's rules, applies what is allowed, and evaluates the story's goals and endings from the real state. The narrator writes; the engine decides.
Metrics
Every story declares metrics such as trust, suspicion, safety or, in a heist, heat. Each has a range and a cap on how far it can move in a single turn. A character-specific metric (affection with one person) is tracked per character. Some metrics are timers the engine counts down itself, so "hours until rescue" really do run out.
Goals with levels
Goals are ladders: document the evidence, identify the stalker, build a case. Each level has a condition written in the story's rules, for example evidence at 40 or a flag set when you found the key. The engine evaluates those conditions every turn; levels never skip and never go backwards.
Chapters and beats
Authors structure stories into beats with entry and exit conditions, required events and a turn budget. The engine advances the chapter when a beat's exit condition is met or its budget runs out, giving the narrator one last turn to land the scene. That is why a StoryNova story has a shape and a length.
Memory that survives
Recent turns are kept word for word. Older turns become short summaries. At each chapter break the engine writes a chapter summary. Separately, a fact ledger records what was revealed, promised, gained or lost. You can open the story panel at any time to read the facts, pin the ones that matter to you, and mark one wrong.
Endings and the recap
Endings are authored: a title, an epilogue and a condition, with a priority so a loss is checked before a win. When one fires you see the epilogue and a recap of the turns that decided it: the choice that unlocked a goal, the turn that crossed a threshold. Endings you have reached are counted, so you know there are others to find.
Typing your own action
Where an author allows it, you can write what you do instead of picking a choice. A validator checks the action against the story's scope and rating first, then the narrator plays it; the state changes it proposes go through the same engine as any other turn.
Frequently asked questions
Can the narrator end the story whenever it wants?
No. An ending fires only when its authored condition is true in the engine's state, or when the story runs out of chapters or days.
Why did a choice not move a metric as much as I expected?
Each metric has a per-turn cap set by the author. Proposals beyond it are trimmed; you can see what applied in the story panel.
Do stories have more than one ending?
Yes. Stories declare several win, lose and neutral endings; the ending screen shows how many you have unlocked.