Five nodes, and real edges

The schedule roadmap Shipped

Decompose, estimate, map dependencies, sequence, present — declared graph nodes with declared edges, rather than one prompt asked for all of it at once.

The roadmap engine walks a declared node graph, so decomposition, estimation, dependency mapping and sequencing are separate steps with separate outputs rather than one dispatch producing a wall of text. And it is a graph rather than a checklist: estimation and dependency mapping both hang off the decomposition and neither one reads the other, which is the point — an estimate that has already seen the blocking edges is an estimate anchored on them. Sequencing is the node that waits on both. Three of the specialist's six declared specialties — todos, schedule and dependencies — are what those dispatches run through.

The five nodes

DecomposeThe feature becomes work items. Runs through the todos specialty.
EstimateEach work item gets a size. Reads the decomposition, nothing else.
DependenciesWhich item blocks which — mapped from the same decomposition, independently of the sizes.
SequenceThe one node that waits on both: sizes plus edges become an order.
PresentThe schedule is rendered. This node dispatches no specialist at all.

Separate nodes because each one has an output the next reads — and two of them deliberately do not read each other. A single dispatch asked for all of it produces a plausible schedule with no visible reasoning to check.

What comes out

Screenshot pendingThe schedule output: a numbered list of todos with a size beside each, then the same todos in execution order, then an explicit dependency list naming which todo blocks which.

Separate blocks rather than one narrative, because each node wrote its own. The dependency list is the part you can argue with — it names pairs, so a wrong edge is a specific thing to point at rather than a paragraph to re-read.