How It Works
Logic Guard scans a project with a fleet of specialist checkers, one per artifact type, then runs a cross-artifact pass that reasons over the whole set together. Most checkers use an LLM to apply their rules to the artifacts they read. One checker, Entity Fields, is pure code and validates references against live Dataverse metadata. The result is a report that combines deterministic integrity checks with judgment-based quality checks.
The scan pipeline
- Select. You choose a project and which sections to scan.
- Read. Each checker reads its artifacts through the Flowon MCP server, on behalf of the signed-in user. Nothing is written back.
- Check in parallel. The eleven specialist checkers run at the same time, each applying its own rules with your per-project overrides.
- Cross-check. Once every specialist has finished, the cross-artifact checker runs. It needs the full set of results because its rules are about relationships between artifacts.
- Stream. Findings stream to the browser section by section as each checker completes, so a large project reports progressively.
- Assemble. The report is built with per-section Critical, Warning, and Info counts, and the worst sections are ranked.
Architecture
The orchestrator runs eleven specialist checkers in parallel, then runs the cross-artifact checker once they have all finished, because its rules are about relationships across the whole project.
The checker fleet
An orchestrator coordinates twelve checkers. Each owns one section and the rules for it.
| Checker | Section it scans | Rules | Count |
|---|---|---|---|
| Business Processes | Business processes | BP-001 to BP-027 | 27 |
| Workflows | Logic flows | WF-001 to WF-025 | 25 |
| Business Rules | Logic blocks | BR-001 to BR-022 | 22 |
| Integrations | Service connections | IN-001 to IN-011 | 11 |
| Configurations | Configuration values | CF-001 to CF-008 | 8 |
| Event Handlers | Event handlers | EH-001 to EH-008 | 8 |
| Schedules | Scheduled jobs | SJ-001 to SJ-008 | 8 |
| Localization | Localized resources | LR-001 to LR-007 | 7 |
| Recipes | Logic recipes | RC-001 to RC-007 | 7 |
| Events | Event definitions | EV-001 to EV-003 | 3 |
| Entity Fields | Field and entity references | EF-001 to EF-003 | 3 |
| Cross-Artifact | Multi-artifact relationships | CA-001 to CA-014 | 14 |
The full list of rules is in the Rule Catalog.
Two kinds of checking
Deterministic integrity checks
The Entity Fields checker is pure code, with no LLM. It cross-references every field and entity your artifacts touch against the actual Dataverse schema and flags references that no longer resolve: a field that was removed, an entity that was deleted, a virtual text field used where it should not be. This is the class of problem that silently breaks logic at runtime, and it is caught with certainty.
Judgment-based quality checks
Every other checker applies its rules with the help of an LLM, because deciding whether a condition is trivially constant, a query is unbounded, or an error is being silently swallowed requires reading the artifact the way a reviewer would. The rules define what to look for; the model applies them consistently across every artifact.
Cross-artifact analysis
Some of the most damaging problems do not exist inside any single artifact. They live in the relationships between them: a schedule that fires a flow that was deleted, a configuration key that is read but never defined, a localized string that is used but never translated, two flows that call each other in an infinite loop. The cross-artifact checker runs after every specialist has finished, so it can reason over the entire project at once. Its fourteen rules (CA-001 to CA-014) are dedicated to these relationships.