Pipeline stages overview
The eight stages your Brief moves through, end to end.
A run proceeds through eight stages, in order. Each stage reads from the running graph + chunked Brief and adds new nodes.
The stages
| # | Stage | Input | Output |
|---|---|---|---|
| 1 | Ingest | Your Brief + support docs | Chunked, embedded text in the retrieval index |
| 2 | Brief Parser | Ingest output | Project metadata, seed constraints, initial stakeholders |
| 3 | Stakeholder Analysis | Brief Parser + chunks | NGOs, MOEs, scenarios, HSI, enabling products |
| 4 | Requirements | Stakeholders + chunks | Functional/performance/interface/constraint requirements, MOPs, TPMs |
| 5 | Decomposition | Requirements + chunks | Functions, components, interfaces, DTRs |
| 6 | Design | Decomposition + chunks | Design specs, interface specs, verification & validation activities |
| 7 | Critic | Everything generated so far | Quality gate; may trigger revisions |
| 8 | Render | Final graph | Artifact bundle (JSON + Markdown) |
Trade studies and risk assessment run alongside the Design stage, operating on graph synthesis (no new chunks). Phase-gate readiness is the final synthesis pass before render.
Why stages matter
Each stage is a checkpoint:
- Outputs are persisted between stages, so a failure in stage 5 doesn't lose the work from stages 1–4
- Revisions can target a single stage — you don't have to rerun everything to fix a stakeholder node (see Revisions)
- Citations flow forward — a requirement cites the stakeholder need it traces to, which cites the Brief passage that defined the need
Cost per stage
Approximate CU consumption for a Phase-1.4 run on a 1,500-word Brief:
| Stage | CU (approx) | % of total |
|---|---|---|
| Ingest | 1 | 2% |
| Brief Parser | 4 | 9% |
| Stakeholder | 6 | 14% |
| Requirements | 8 | 19% |
| Decomposition | 7 | 17% |
| Design | 10 | 24% |
| Critic | 4 | 9% |
| Render | 2 | 5% |
| Total | ~42 |
Actual costs vary with Brief length and complexity. The pre-run estimate uses the median for your Brief size.