Project Explorer

Files

Story files

Browse the notes, research, decisions, and workshops used to develop the series.

Browse the files used to develop the story.

Search 577 documents about the world, characters, plot, research, decisions, workshops, and earlier ideas.

Current document 08 Story Loop/DEVELOPMENT-ORCHESTRATOR.md

Development Orchestrator

Purpose

Provide one executable path through the development environment. This document decides which module runs, in what order, what state passes forward, and where the process must stop for the author.

Default pipeline

scope -> retrieve -> classify authority -> Gap Analyzer -> choose mode/scale -> diverge cheaply -> shortlist -> optional research -> expand one branch -> Character Factory as needed -> structure packet -> optional development prototype -> relevant critics -> author gate -> approved integration -> next-state record

When recent research, brainstorming, and canon have accumulated but no single candidate has been selected, insert Research → Creative Possibilities / Inspiration Pass after status-safe extraction and gap analysis. It creates a portfolio for author interest selection; it does not replace the later author gate or critics.

Finished-novel drafting is not part of the default pipeline.

When work comes from the weekly completion set, Story Completion Workflow controls horizontal progression. The orchestrator runs one task at the current sweep depth; it does not take that task through every scale before the rest of the registry catches up.

Step 1 — Scope the run

Declare:

  • one target;
  • current scale: era, arc, sequence, chapter, scene, or prototype;
  • classification: close-gap, strengthen-structure, or new-expansion;
  • explicit exclusions;
  • stop condition;
  • token mode: Micro, Standard, Deep, or Cascade.

Default to close-gap. Use Cascade only when the author asks to move from broad structure into a representative story sample in one run.

For a registered completion task, also declare task ID, current sweep, current depth, loop phase, prerequisites, and downstream review reach from 07 Coordination/Story Completion Workflow/TASK-REGISTRY.

If the target comes from Weekly Story Synthesis, copy its source paths, authority labels, queue classification, and exact stop condition into the run. The synthesis ranking recommends where to work; it does not bypass this orchestrator or the author gate.

Step 2 — Retrieve and classify

Build one compact context packet from direct sources. Separate:

  • established facts;
  • working structure;
  • proposed candidates;
  • unresolved decisions;
  • rejected or superseded material;
  • non-canon exploration.

Do not let later prose or a persuasive sample outrank an explicit author decision.

Step 3 — Analyze gaps

Run 08 Story Loop/GAP-ANALYZER. Select one highest-value gap or one tightly connected cluster. If an author decision blocks the next branch, prepare that decision and stop.

Step 4 — Choose the route

Routes may combine, but every module must answer a specific uncertainty.

Step 5 — Diverge breadth-first

Generate compact alternatives before detailed development.

  • Micro: 3 narrow options.
  • Standard: 4–6 options; expand the best 2.
  • Deep: 5–10 broad options; shortlist 3.
  • Cascade: 3–5 broad options; follow only 1 working branch downward.

An AI shortlist is a recommendation, not author approval or canon.

For a weekly or cross-note inspiration request, use the Creative Possibilities Pass here. Keep the source window bounded, generate candidates across different families, and save the unreviewed packet in Brainstorms/ with every candidate marked NON-CANON EXPLORATION.

Step 6 — Research selectively

Research only when the result could change plausibility, mechanism, constraints, or dramatic possibilities. Translate findings through:

mechanism -> human pressure -> Seeds behavior -> concrete situation -> choice -> consequence

Label supported fact, extrapolation, premise, and unsupported material. Research cannot resolve a story choice.

Step 7 — Build downward

Move through only the scales needed:

era -> arc -> sequence -> chapter -> scene -> prototype

At each scale record the starting state, objective, opposition, attempts, change, downstream consequence, assumptions, risks, and next-scale target.

Use 08 Story Loop/Templates/development-chapter-packet before a chapter or compressed-chapter prototype.

Step 8 — Create only the required cast

Generate role candidates first, then Tier C/D packets. Promote only characters who survive testing or receive author approval. Names created inside Story Exploration remain placeholders unless promoted.

Step 9 — Prototype only when useful

A 500–1,500 word development prototype is appropriate when prose-in-motion can test dialogue, explanation, chemistry, objective clarity, or emotional effect better than an outline.

Every prototype package must state:

  • what it tests;
  • provisional inventions;
  • what worked;
  • what failed;
  • decisions still required.

Readable does not mean manuscript-ready. Prototype prose never establishes canon.

Step 10 — Run only relevant critics

Choose from:

  • causality;
  • character;
  • continuity/status;
  • systems/permissions;
  • Creative Interest;
  • Story Fit;
  • problem-solving/listener clarity;
  • cost/token efficiency.

Critics diagnose their dimensions. They may repair only already-authorized issues.

Step 11 — Author gate

Return a compact decision packet:

  • surviving alternatives;
  • consequences and risks;
  • new assumptions;
  • critic disagreements;
  • exact decisions needed;
  • recommended next scale.

The author may accept, reject, park, combine, rerun, research, develop further, or promote. Silence is not approval.

Step 12 — Integrate and record

Only after approval:

  • update the relevant Story Unit or compiled note with explicit status;
  • update dependencies and open questions;
  • link any weekly-synthesis queue item or journal/evidence record that initiated the run;
  • record consequential decisions or contradictions in 07 QA/;
  • preserve rejected alternatives when useful;
  • record a compact next state rather than the entire reasoning transcript.

Then run decision propagation when anything approved changed:

Use 08 Story Loop/Templates/development-run for the durable run record.

Sequence, scene, and draft boundary

Use 08 Story Loop/Templates/sequence-packet once a task reaches sequence integration and 08 Story Loop/Templates/scene-packet once the whole sequence layer is coherent. Treat every packet as a state-transition contract, not canon.

Finished prose is admitted only through 08 Story Loop/Templates/draft-pipeline-run. The scene must already be structurally stable, the author must explicitly request drafting, and only author-approved prose may enter ordered manuscript material.

Hard stop conditions

Stop before further expansion when:

  • a load-bearing author choice is unresolved;
  • sources conflict and authority is unclear;
  • research is needed to distinguish viable mechanisms;
  • a candidate would require canonizing a new system rule;
  • the current scale does not support moving downward;
  • the token mode's depth limit has been reached;
  • the candidate reaches the dedicated Samuel–Konrad hierarchy boundary.

Current containment boundary

The following are established at the current level: Konrad genuinely controlled his own city; he never considered failure; the later loss occurs in the largest empire's containment environment; Samuel remains an outsider; Samuel's earlier first-primary placement and bounded jurisdiction matter; Konrad later reactivates or migrates machinery there; Samuel becomes decision-making alpha and Konrad eventually beta inside an ongoing power struggle; Samuel offers Konrad an apparently separate half but maintains the shared environment; and the false separation helps him take over initiatives and much of the group.

The exact lock trigger, technical transfer, supporting proof, automatic versus later powers, bounded resistance, evidence, reset limits, and destruction chain remain unresolved. Route those questions to the dedicated piece-by-piece brainstorm. Do not use a prototype or exploration run to decide them silently.

Success condition

The orchestrator succeeds when a broad request such as “develop this part of the Great War” produces a small context packet, prioritized gap, several cheap alternatives, one selectively deepened branch, an optional readable test, focused critic findings, and a real author decision—without context flooding or premature canon.