Entity commissioning process reference

# Run a single profile for one entity (recommended for targeted sync)
./projects/ores.codegen/codegen.sh regenerate \
    --component refdata-cpp --profile qt \
    --entity rounding_type

# Run a single profile for all entities in a component (generates everything)
./projects/ores.codegen/codegen.sh regenerate \
    --component refdata-cpp --profile qt

When running --component without --entity (once that flag lands; see Codegen developer experience improvements):

• 13 spurious ERROR lines appear for table and junction models under the qt profile. These are expected skips, not real failures — see Downgrade profile/model-type incompatibility from error to skip.
• Files for all 25+ entities are written to disk. Scope the diff to the target entities and discard the rest.

After running --component mode, discard out-of-scope generated files:

# Revert tracked files modified for non-target entities
git checkout HEAD -- projects/ores.qt/refdata/

# Remove untracked new files generated for entities not in target set
git status --short | grep '^?' | awk '{print $2}' | xargs rm -f

Supported model types: domain_entity, junction, schema, table

⚠ Known issue (open): The sql profile also fires sql_schema_domain_entity_create for domain_entity models alongside the newer sql_schema_create for table models. Entities that have both a _table.json and a domain_entity model have two codegen paths to the same file. See Codegen model safety guardrails (C1 concern).

Supported model types: schema, domain_entity

⚠ Known path inconsistency (open): The class template uses projects/ores.{component}/api/ → projects/ores.refdata/api/ (correct, follows the actual directory layout). The six secondary templates use projects/ores.{component_include}/ → projects/ores.refdata.api/ (a non-existent top-level directory). Git sees those six files as untracked, giving a false-clean diff. Status: pre-existing bug; deferred to Refactor ores.codegen C++ generation.

Supported model types: schema, domain_entity

Supported model types: schema, domain_entity

Supported model types: schema, domain_entity

Supported model types: domain_entity, schema

Supported model types: domain_entity, schema

Supported model types: domain_entity, schema

Supported model types: domain_entity only

A codegen sync task for any profile is complete when:

1. Output location check (do this first, before diffing): every generated file lands in the correct split project directory, not in the monolith:

   Profile	Expected root	Wrong root (monolith — do NOT write here)
   domain (class template)	projects/ores.{component}/api/	(same path, check secondary templates)
   domain (secondary: json_io, table, table_io)	projects/ores.{component_include}/	projects/ores.{component}/api/
   generator	projects/ores.{component_include}/	projects/ores.{component}/api/
   repository	projects/ores.{component_core}/	projects/ores.{component}/core/
   service	projects/ores.{component_core}/	projects/ores.{component}/core/
   protocol	projects/ores.{component_include}/	projects/ores.{component}/api/
   nats-eventing	projects/ores.{component_include}/	projects/ores.{component}/api/
   nats-handler	projects/ores.{component_core}/	projects/ores.{component}/core/
   qt	projects/ores.qt/{component}/	(no split variant; single path)

   For refdata (component_include=refdata.api, component_core=refdata.core):

   • API headers → projects/ores.refdata.api/include/ores.refdata.api/
   • Core headers → projects/ores.refdata.core/include/ores.refdata.core/

   If a file appears as untracked under projects/ores.refdata/ after running codegen, the output went to the wrong location. Check component_include / component_core in the entity model (* C++ / ** Flags).

2. Every generated file either: a. Produces zero diff against the corresponding hand-written file in the monolith (projects/ores.{component}/{api,core}/), or b. Has its delta classified (category from the drift table above) and the decision recorded (fix template / fix code / accept).
3. All "fix template" decisions have a corresponding template commit.
4. All "fix code" decisions have a corresponding code commit with justification.
5. All "accept" decisions are recorded in the task's * Notes or * Review table.
6. No generated files for non-target entities have been left modified.
7. The build passes (no C++ regressions).

• One commit per logical fix (template fix, code fix, or model update).
• Never amend commits on a branch under review.
• Never use git add -A — add files individually to avoid committing generated files for unrelated entities.