Task: Roll out the classification tool to all remaining components and delete stale docs

Narrowed 2026-08-04. Originally: extend the trusted, refdata-proven tool to every remaining component, produce the full quantified drift baseline, and retire everything the tool now supersedes. The re-scoping half is done — each of the 14 ad hoc tasks was re-scoped directly against the codegen component catalogue rather than a report (see the parent story's Decisions). What remains after 2026-09-03: retire entity_coverage_matrix.org, entity_evaluation_checklist.org, and entity_catalogue.org, capturing anything still useful first. The report question is settled: not produced. Each component story's survey supersedes it, and a read-only cross-component report would be stale before landing (see the parent story's Decisions, 2026-09-03).

• Each of the 14 ad hoc "Apply safe drift to X-cpp" tasks is closed as covered: 12 re-scoped into their own entity classification and drift baseline stories (one per component, composite components split into subcomponent tasks), 1 (ores.controller) ABANDONED as decommissioned. Done.
• entity_coverage_matrix.org, entity_evaluation_checklist.org, and entity_catalogue.org are deleted; anything from them still useful is captured first (in the 12 stories' notes, or elsewhere).
• (Optional) The classification tool is run across every remaining component and produces a combined drift report with per-component breakdowns and totals — only if judged still worth producing once the 12 stories close; not required to close this task if the per-component stories already supersede its value. Resolved 2026-09-03: not produced (see Goal and the parent story's Decisions).

• 2026-09-03: recorded the 2026-09-03 program decisions in the parent story; re-scoped this task; scaffolded the method-skill task.
• Raise a board PR (this commit): story/task text, sprint board rows for this story and for refactor_codegen_cpp (its row still described the pre-2026-08-04 sweep scope).
• Retire the 3 stale docs in one doc PR: merge the evaluation checklist's per-layer criteria into the commissioning process reference, reduce the three files to retirement stubs that keep their :ID: (the site export aborts on any unresolved id: link, and archived sprint records still cite these pages), and re-point or trim every live reference (analysis index, commissioning reference, model-unification analysis, system model domain, product backlog captures).
• Verify the doc PR with a site export check, then raise it for review.
• Close with a Result recording the retirement and the report decision.

Manual QA scenarios (scaffolded via compass add test_scenario, run through the QA Validation Runner panel) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

Closed 2026-09-04. Both closeout PRs landed:

• #2001 (board): task and story re-scoped, the 2026-09-03 program decisions recorded in the story, sprint 25 board rows synced.
• #2002 (docs): the three stale docs retired to stubs that keep their :ID:, so every archived id: link still resolves; the evaluation checklist's per-layer criteria merged verbatim into the commissioning reference (entity_commissioning_reference.org, section "Commissioning criteria by layer"); every live reference re-pointed to the commissioning reference or the component catalogue; the dormant entity_evaluation_skill capture moved next → discarded with its supersession note; backlog indexes regenerated.

Three bot findings on #2002 handled in the review round: fabricated method-skill task ID corrected (6509b26a8f), stale backlog indexes regenerated (992021d7a2), completed-plan link in doc/plans noted with no change (read-only history; the stub keeps the link resolving). The re-review confirmed both fixes and found no further issues.

Acceptance: the 14 ad hoc tasks are re-scoped (12 stories, 1 ABANDONED); the three docs retired with their useful content captured first; the quantified report question settled (not produced, recorded 2026-09-03 in the story's Decisions). The method-skill acceptance item is delivered by its own task Encode the per-component codegen drift-elimination method as a skill, still BACKLOG in this story.