Discovery and Story Charters

Discovery is the interval before NovelGraph is allowed to treat a book idea as a production premise. It records what the author says, what an agent has inferred, what remains unresolved, and the alternatives considered. Its output is not a manuscript or an outline. Its decisive output is an approved Story Charter.

The lifecycle is shown in the canonical discovery-state diagram and charter-lifecycle diagram.

Textual equivalent:

Book shell → active discovery → charter proposed → author decision
                                     │                 ├─ approve → approved charter → production
                                     │                 ├─ revise → active discovery
                                     │                 └─ reject → keep exploring
                                     └─ material change after approval → impact review → replacement or keep current

What a discovery session records

Starting discovery for a book first ensures that its literary, series, and book knowledge bases exist. If there is already an active or charter-proposed session for that book, NovelGraph reuses the most recently updated one. Otherwise it creates a new active session with the opening question: “What experience should remain with the reader after the final page?”

Turns are deliberately small records: an author or sol role, an ordinal, content, and timestamp. A turn may carry observations. An observation has a key and a JSON value, but it also carries the distinction that matters during discovery:

Field	Meaning
Provenance	author-stated, agent-inferred, literary-guidance, or agent-proposed
Confidence	Optional number from 0 through 1
Status	working, unresolved, proposed, rejected, approved, or superseded

That metadata prevents a useful interpretation from being presented as something the author actually decided. “The reader should suspect the sister” can be author-stated; “the sister is the cleanest suspect because she carries the family obligation” is an inference; “a cited craft source recommends a competing suspect” is guidance. All three can inform discussion, but they have different authority.

Candidate thrusts are alternatives, not hidden commitments

A discovery session can contain three kinds of story-thrust candidate: core, stretch, and wild. These labels do not rank a candidate as correct. They provide a disciplined way to compare the most direct reading of the conversation, a version that intensifies an existing pressure, and a neighboring possibility that may reveal a preference through contrast.

Each candidate is a typed proposal with a title, main thrust, reader promise, character engine, central conflict, thematic pressure, genre obligations, ending shape, risks, and foreclosures. The last two fields are particularly useful in a charter discussion. A candidate should name what it makes harder to write, what it may rule out, and what promise it asks the author to keep. Proposing a candidate changes the session status to charter-proposed; it does not approve a premise and does not create canon.

The Charter is the production contract

The Story Charter is a versioned record attached to one book. Its required core is a mainThrust, at least one readerPromise, a structured protagonist value, and a centralConflict. It can also record:

• genreContract
• thematicQuestions
• narrativeForm
• endingHorizon
• constraints
• productiveUnknowns
• rejectedDirections

This is a charter, not a full synopsis. productiveUnknowns make room for discovery that should happen during planning and drafting. rejectedDirections preserve a deliberate “no,” so a later suggestion can be evaluated against a decision rather than rediscovered as an apparently fresh idea. The charter’s narrativeForm is a record rather than a fixed enumerated schema, so it can hold the form decisions a project actually needs without inventing one template for every book.

Charter field	The question it holds still
Main thrust	What is this book fundamentally doing?
Reader promise	What experience, pressure, or payoff does the book owe?
Protagonist and central conflict	Whose choices drive the work, and against what force?
Genre contract	Which genre obligations are accepted for this book?
Ending horizon	What kind of resolution must the ending move toward?
Constraints and rejected directions	What is out of bounds, even if it later seems expedient?

Approval changes the system state

When a charter is proposed, NovelGraph assigns the next version number and creates a pending approval record. Resolving that decision requires a non-empty author rationale. On approval, any previously approved charter for the same book becomes superseded, the proposed version becomes approved, and non-superseded discovery sessions for the book are marked approved. The decision is also recorded as an attributable event with the charter content and the author’s rationale.

If the book has no chapters, approval creates chapter 1, titled “Opening,” with empty content. That is a practical transition into writing; it is not evidence that an opening scene has been designed or drafted. Rejection marks only the proposed charter as rejected and leaves the author able to continue the conversation.

Production has a hard gate here. WorkflowHarness.start checks for an approved charter before it creates a job or inserts workflow nodes. An outline, drafting run, or review pass therefore cannot begin merely because a candidate is persuasive or an agent has produced a polished description.

Revision after approval

A charter is not immutable, but changing it should be explicit. The lifecycle diagram represents the path from an approved version through impact review to either a superseding replacement or retention of the current version. In the current core implementation, approving a later charter version supersedes the earlier approved version; the system does not yet calculate a complete dependency impact set for charter edits. Treat a proposed replacement as a premise-level change and inspect its consequences for existing scenes, obligations, mystery records, and audits before approving it.

The useful question is not whether a charter makes the book rigid. It is whether a later change is visible enough to revise the work that depended on the earlier choice. Discovery gives the author a record of why the governing premise was adopted. The charter gives production a stable premise until the author decides that premise has changed.