docs: propose active preference selection research

[PhysShell]

Summary

• add docs/proposals/active-preference-selection.md
• document CODA as prior art for active model selection
• translate the useful decision loop into a Griff-specific, contextual preference-learning design
• define deterministic replay, complete-panel evaluation, baselines, metrics, failure slices, and evidence-gated phases
• index the proposal in docs/proposals/README.md

Why

S8 now captures favorite / reject / history / provenance, while S9 plans the canonical feedback contract and a simple explainable reranking baseline. CODA suggests a useful later question: can Griff spend limited user attention on the comparisons that most reduce uncertainty?

CODA's classification machinery does not fit Griff directly. Musical preference is not an objective class label, candidates are different musical objects, and the preferred strategy is context-dependent. The proposal therefore adopts the posterior → hypothetical update → acquisition loop, but recommends pairwise/contextual preference models and offline replay rather than porting CODA's code or confusion matrices.

Guardrails

• proposal for discussion only
• no production reranking, sampling, generator, scorer, playback, or cockpit behaviour changes
• no Python/PyTorch/MLflow dependency
• no CODA code reuse
• no new roadmap stage or S9 contract amendment
• no active policy before stable S9 provenance and the strongest simple baseline exist

Decision requested

Approve only a future Phase 0 audit and benchmark contract after the S9 feedback contract and simple baseline are accepted. Implementation and product promotion require separate evidence and acceptance contracts.

Validation

Docs-only branch compared against main:

• 2 commits ahead, 0 behind
• 2 files changed
• proposal and index fetched back from the pushed branch
• no executable code changed; cargo checks are not applicable

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 42126320-cbad-444f-aff7-1792a578e12d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch agent/coda-active-preference-selection

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}