Story: Entity classification and drift baseline across all components

Coordinate the drift-elimination program to its end state: every component regenerates byte-identical output from its org models, the drift check enforces it, and no hand-edited generated file survives. This story records the program's operating decisions, hosts the canonical per-component method as a reusable skill, and retires the three manual-tracking docs the program superseded. Each component is remediated by its own story; the ordering and the method come from here.

• The 14 ad hoc per-component drift tasks are re-scoped (or closed/replaced) using real data instead of guesses. Done 2026-08-04.
• entity_coverage_matrix.org, entity_evaluation_checklist.org, and entity_catalogue.org are deleted once whatever supersedes them is in place. Done 2026-09-04 (PR #2002): the checklist's per-layer criteria were merged into the commissioning reference, the three files reduced to retirement stubs that keep their :ID:.
• The canonical per-component method exists as a reusable skill (task below), and the program's ordering and operating rules are recorded in Decisions. Done 2026-09-04.
• The dormant product-backlog entries that this method supersedes are closed or re-pointed at the skill. Done 2026-09-04.
• (Resolved 2026-09-03) The classification tool is run across every remaining component producing a combined quantified report. Decision: not produced; the 12 stories' per-component surveys supersede it.

• 2026-08-04: re-scoped the 14 ad hoc "Apply safe drift to X-cpp" tasks directly against the codegen component catalogue (projects/ores.codegen/library/component_catalogue.org) and the ores.refdata pilot's proven acceptance bar, rather than waiting on the read-only classification/drift report this story originally scoped. Result: 12 new per-component stories (composite components – ores.dq, ores.iam, ores.trading – get one story with a task per SQL/C++ subcomponent; the other 9 get one story each), each requiring profile binding, zero-diff regen, and extending the check_component_drift.py CI check to that component – so the bar is "zero drift, CI-enforced," strictly stronger than this story's original "produce a report" scope. ores.controller was excluded: the component has been decommissioned (projects/ores.controller* no longer exists), so its ad hoc task was marked ABANDONED rather than re-scoped. See the sprint's Stories table for the 12 new story links.
• 2026-09-03: approved an execution program for the 12 remediation stories, replacing the ad hoc, sprint-by-sprint drift work of the previous sprints. Components run foundations-first: fix-codegen-template-drift and close-systemic-codegen-gaps land before any component story starts, so a codegen bug found mid-sweep does not waste the sweep. Each component story runs one documented algorithm per component: baseline regen before any model edit, survey every model file, bind entities to profiles, record each no-match with its real feature values, fix codegen gaps at their source, regenerate, require byte-identical output, review the generated code for C++ idiomaticity and smells, and only then extend the check_component_drift.py gate list. Partial codegen coverage is not a valid end state; a type too complex to code-generate is an exception that needs a recorded justification per entity. Automation starts granular (level 0: every fix verified by build, test, and byte-identical diff before its PR) and rises only on evidence. PRs stay small and human-reviewable. The algorithm is encoded as a reusable skill (task above) so every story runs the same hardened recipe. The quantified classification report is dropped: each component story produces its own survey, and a read-only cross-component report would be stale before it landed.
• 2026-09-07: delegation tidy-ups, so any environment can claim a remaining bind cold. The eight remaining bind stories and tasks name the compass-codegen-fix-drift skill as the canonical method (their August text pointed at the ores.refdata pilot), and their gate wording names the registry check: the refdata-codegen-drift CI job the old text cites is deleted, and the --components mode is gone. Done shape, decided once: a bind is done at the drift gate; the story is done at the live system test. The bind task closes at the gate (byte-identical regen, build, tests, registry entry, merged PR). The story closes only after the post-merge live system test on a fresh database: recreate the database, provision a tenant, and exercise the component's flows. Defects the test surfaces become follow-up tasks on the story (the compute/iam precedent: the iam service-grant fix rode the iam story). The skill gained the composition check as its step 7: verify every materialized registrar family is actually called from the service registrar/application. The trigger was trading: its messaging stacks were materialized but never registered (the PR #1950 audit findings), and the 2026-09-06 bind landed four registrar families with no caller on main (lifecycle_event, trade_id_type, trade_identifier, trade_party_role). The skill's Reference now also cites the iam and trading worked records.

• Fixing/regenerating entities was originally out of scope for this story; superseded – fixing is now the explicit acceptance bar of the 12 new stories.