This reference helps developers and operators inspect a local CPG or investigate an architecture audit discrepancy. It explains how to check that architecture run tables exist and how to distinguish missing data from a run with no findings.
GoCPG owns writes to the project DuckDB. Application consumers should read through
gRPC. The gocpg query command is for local diagnostics and integration checks;
it does not replace the API contract.
This reference covers architecture result tables. It does not enumerate core graph nodes and edges; application consumers should use the gRPC contracts.
Architecture result tables
| Table | Stored data |
|---|---|
cpg_architecture_components |
Components and their stable UIDs for one run. |
cpg_architecture_memberships |
File and symbol classification with selector provenance. |
cpg_architecture_edges |
Direct and transitive component dependencies. |
cpg_architecture_runs |
Completed run identity: commit, model, rules, completeness and mode. |
cpg_architecture_result_heads |
Pointer to a project’s latest completed canonical run. |
cpg_architecture_findings |
Findings bound to a run ID and stable fingerprint. |
An empty cpg_architecture_findings result means there are no findings for the
selected run. A missing table means the database has an older or incomplete schema.
It cannot establish that the architecture passed a check.
Inspect a local database
Open PowerShell at the root of a CodeGraph source checkout. You need a GoCPG binary
with DuckDB support and an existing CPG. Replace <path-to-cpg.duckdb> with your
database path. This command reads the schema and does not change source files:
.\gocpg\gocpg.exe query --db "<path-to-cpg.duckdb>" --sql "SELECT table_name FROM information_schema.tables WHERE table_name IN ('cpg_architecture_runs', 'cpg_architecture_result_heads', 'cpg_architecture_findings') ORDER BY table_name"
A successful check returns three table names. If the database does not exist,
create a CPG using gocpg parse or a managed update. If fewer than three rows are
returned, update GoCPG and recreate or migrate the database through the supported
runtime process.
Implementation check
Run this command from the gocpg directory in the source checkout. It requires Go
and the DuckDB build dependencies described in the
environment setup.
The repository test creates temporary storage, persists a canonical result and
checks deterministic export. It verifies the implementation; it is not a procedure
for inspecting a user’s database:
go test ./pkg/architecture -run '^TestCanonicalResultsPersistAndExportDeterministicEvidence$' -count=1
Expect ok gocpg/pkg/architecture. Build failures and missing dependencies indicate
an environment problem rather than a result about the quality of the analyzed code.
For operational result access and export, use the interfaces and results reference.