liquidware-ninjaSecond brain & toolkit
Personal project. Research and learning only. Not for production. No warranty.Not for production
  • Built on my own time for research and learning. Not affiliated with, endorsed by or supported by my employer, Liquidware or Microsoft.
  • Bring your own Liquidware licenses. Don't run any of it in production or against tenants and subscriptions that matter. Test in a dedicated lab first.
  • Provided "as is", without warranty of any kind, under each repository's license. You use it at your own risk and own the Azure charges it incurs.
  • Contributions and forks are welcome and encouraged on public repos. See Contribute.

0013. Follow the family's model-selection convention

proposedReviewed 2026-10-19Source: decisions/0013-family-model-selection-convention.md

Context and Problem Statement

Work on this repo happens in Claude sessions: threads that write Bicep and PowerShell, research Liquidware's products and keep this knowledge base. The family has one convention for which model and effort each kind of Claude work uses, and model choices are the AI knowledge base's topic (family.yaml). Does this repo follow it, and what does it change here?

Considered Options

  • Follow the convention, with any exception recorded here as a "Model override" decision
  • Keep this repo's own model choices

Decision Outcome

Chosen option: "Follow the convention", because this repo's work is the convention's thread case (code, research, specs: Claude Opus 5.5 at medium effort), and nothing here needs an exception.

Also recorded in: 2ndbrain, the source of truth for the defaults, when to raise effort, fallbacks and the pilot dates. The facts behind it are on the AI KB's Claude API page.

For this repo:

  • Threads start on claude-opus-5-5 at medium effort. The repo pins the same pair for Claude Code in .claude/settings.json (model, effortLevel), with a full model ID, never an alias.
  • Effort goes up before the model changes: high for changes that touch secrets, identity or spend, which much of this repo's Bicep does.
  • No code in this repo calls a model, so the convention's layers for apps and for Microsoft Foundry don't apply here.

Consequences

  • Good, because model choices for this repo follow evidence and the family's rules.
  • Neutral, because no default changes today.

More Information

Revisit at the family's pilot review on 2026-12-01.

Linked from one page