Phase 1.2: design diagnostics + 1.1.5 surfaced

[Gimpleberry]

• diagnose_cbc_design() with MNL D-efficiency under uniform priors
• QUALITY_GATE_MAX_LEVEL_IMBALANCE promoted to shared.py per Tenet 1
• Renames n_dominated -> n_duplicate alternatives (honest naming)
• Production D-eff ~0.38 below 0.85 gate; sentinel test asserts the failure with docstring explaining the 1.1.5 fix path
• 25+ new tests

Summary

Changes

• path/to/file.py — what changed and why

Linked work

• Closes: BACKLOG item 1.2 (design diagnostics)
• Backlog: Phase 1 — CBC core, item 1.2; surfaces new 1.1.5
• ADR: None — no architectural change; conforms to existing ADR-005

Validation checklist (Tenet 5)

• 1. Lint — ruff clean on every modified file
• 2. shared.py uniqueness — no symbol duplicated outside shared.py
• 3. Encoding sweep — no non-ASCII in Python source
• 4. Plugin registry — every entry has start()/stop()
• 5. Import chain — no circular imports
• 6. Lifecycle — main.py --check passes
• 7. Docs present — README/CHANGELOG/DECISIONS/BACKLOG/PROJECT_KNOWLEDGE
• 8. Config valid — kai validate-taxonomy passes
• 9. Unit tests — all green
• 10. Integration boot — main.py --boot-order works

scripts/validate.sh summary:

Passed: __ / 10

Documentation updated

• CHANGELOG.md entry added under appropriate version
• PROJECT_KNOWLEDGE.txt updated if architecture or folder map changed
• DECISIONS.md ADR added if this is a non-trivial architectural choice
• BACKLOG.md updated (phase items moved/completed/added)
• In-app Help section added/updated if user-facing behavior changed

Screenshots / log excerpts

N/A — no UI or user-facing output changes.

Risk + rollback plan

Diagnostic correctness change. The production sentinel test
intentionally asserts a gate failure (D-eff ~0.38 vs 0.85 threshold)
with a docstring explaining the path forward via BACKLOG 1.1.5.
No downstream consumers depend on the diagnostic output yet.
Revert is safe: restores the NotImplementedError stub.