# Story Intelligence Single-System Runtime

Story Intelligence uses one production architecture:

Core Import -> canonical Project/Book/Chapter records -> canonical Scenes -> SceneManuscriptSources -> Scene Intelligence analysis -> retained raw Scene Intelligence results -> deterministic candidate preparation -> persisted review candidates/evidence -> Review Centre -> persisted candidate page GETs -> persisted decisions -> canonical materialisation.

Raw Scene Intelligence JSON is source and audit data. Normal review GET requests must not parse it. Scene JSON may be read only by the deliberate preparation/materialisation path or by explicit reanalysis.

## Sources Of Truth

| Concern | Authoritative Persisted Source |
|---|---|
| Scene text | `SceneManuscriptSources` |
| Scene analysis audit | `StoryIntelligenceSceneResults` |
| Entity review preparation | `StoryIntelligenceReviewCandidateGenerations` |
| Entity review candidates | `StoryIntelligenceReviewCandidates` |
| Entity review evidence | `StoryIntelligenceReviewCandidateEvidence` |
| Entity review completion | `StoryIntelligenceBookPipelines.*ReviewCompletedUtc` |
| Plot/Thread review candidates | `StoryIntelligenceNarrativeCandidates` |
| Plot/Thread evidence/events | `StoryIntelligenceNarrativeCandidateEvidence`, `StoryIntelligenceNarrativeCandidateEvents` |
| Metrics | `SceneMetricValues` |
| Review Centre tiles | `StoryIntelligenceReviewCentre_GetSummary` |

## Candidate Preparation

Candidate preparation is deliberate server work. It reads retained Scene Intelligence results, applies the module generator, and stores the resulting candidate page payloads and compact evidence. Preparation creates a current generation per Book and module.

Characters are prepared after scene import/materialisation. After Character Review completes, peer candidates are prepared for Locations, Assets, Relationships, and Knowledge. Whole-book Plot synthesis remains the only final Plot/Thread candidate generator.

## Review GETs

Normal detailed review GETs perform:

authorization -> compact review frame -> `StoryIntelligenceReviewCandidate_ListPage` -> ViewModel -> Razor.

They do not call candidate generators, do not call `GetProgressAsync`, do not call `ResumeBookAsync`, do not call `EnsureCandidatesAsync`, and do not read `StoryIntelligenceSceneResults`.

## Failure

If a current generation is missing, preparing, or failed, review pages show no generated candidate rows from a fallback path. The fix is deliberate preparation or retry, not silent Scene JSON reconstruction during page render.

## Removed In Phase 27D

Normal review GETs no longer use the transient Scene-JSON-backed entity candidate rebuild path. Those builders remain reachable only as candidate preparation/materialisation generators.