Before a new calculation for estimated value ships, you want to know what reads it, and what reads those. The platform's dependency check lists the declared references, but it doesn't read the logic.
A form, a view's filter, a plugin that fires when the value changes, an AI prompt that summarizes it, a flow that copies it somewhere else. So you trace it by hand, one hop at a time.
In this demo scenario, nine components read estimated value: a form and a view show it, two AI prompt columns summarize it, a form script and two Dataverse functions calculate with it, a plugin step fires when it changes, and a cloud flow, switched off today, copies it into a compliance review. Each edge is labeled Display (it's shown), Filter (it decides whether something runs) or Computation (something is calculated from it).
Downstream, the compliance review's exposure amount rolls into a branch summary that ends up on a Power Pages page, the branch capacity board. Upstream, a plugin writes estimated value when an appointment request is created, from a basic form on the same site whose table permissions are configured to let anonymous visitors create requests. It's structure, not traffic, and this chain is a demo scenario built to exercise the parsers: yours won't be this tidy.
One Pathix call per hop: what reads estimated value, then what reads what that writes. With the Dataverse MCP alone, each hop is the same search, done by hand again.
A read-only metadata scan of your environment that reads the logic inside the components: compiled plugins, flow definitions, workflow XAML and scripts.
How it works →Or do it by hand: what breaks when you change a column→All questions →
The detail behind the answer, for the reader who needs it.
Three of the four are on the Pathix MCP server, so an agent can run the same analysis and hand you the answer rather than you driving the console: describe with a table's logic section, and analyze with cascade or migration_plan. Every call is read-only, and the agent signs in as you.
The plan folds two kinds of coupling into one sequence: column data flow, meaning one component writes what another reads, and orchestration, meaning one component invokes another. Both have to hold for the order to be safe.
Point your agent at this with the Pathix MCP and it returns the same ordered plan, with the cycles and unproven edges still labeled.
Clusters with no coupling between them. Two tracks can be worked at the same time by different people without either one waiting, which is what makes a large migration divisible in the first place.
Wave 0 is what nothing downstream depends on, so it is safe to change first. Work outward from there toward the roots everything else sits on.
Cascade behavior is configured per relationship and its reach compounds quietly. Pathix walks the relationship metadata outward along whichever axis you pick and returns the tables that operation actually reaches, with the depth it traveled to get there.
Cascade is only one route. A plugin or a workflow can issue a delete directly, and that path does not appear anywhere in the relationship metadata. Pathix answers the question from both sides at once: the relationships that cascade a delete into the table, and the component logic that deletes from it on purpose.
A migration plan that looks certain is worse than one that shows its gaps, because you will act on it either way. Three things the rewrite plan flags about itself.
Where two components depend on each other, no internal order is provable. Pathix marks the group rather than picking one and presenting it as the answer.
An edge crossing an Unresolved or AI-derived endpoint is labeled unproven. It stays in the plan, because dropping it would make the sequence look more certain than it is.
A component can land in wave 0 because nothing depends on it, or because Pathix could not resolve what it depends on. Those two are very different, and the plan tells them apart.
The plan reflects coupling Pathix could model from structure. Treat it as a sequence worth checking rather than proof, which is also how the product words it.
The backward-looking half: what moved between two scans, down to a single privilege grant.
Read →The manual method for rename, retype and delete, including why the platform's delete check will not stop you breaking a plugin.
Read →The edges every one of these capabilities is built on, and how confidence is expressed per edge.
Read →A 30-minute walkthrough on a pre-scanned demo environment. No access to your tenant, nothing to install.