Stay, Switch, or Hybridize

Six questions tell you which migration path fits your Claude Code workflow. No email, no signup, runs in your browser.

How to use this: answer six yes/no questions about your past two weeks with Claude Code. Each question maps to one of the six migration triggers documented in the Migration Playbook. Your pattern of answers maps to one of three paths: stay and fortify, hybridize with model routing, or switch platforms entirely.

The check takes about three minutes. Your answers stay in your browser; nothing is sent anywhere.
Question 1 of 6: Cache discipline

Has your `cache_creation` to `cache_read` ratio crossed 0.40 across a typical working day in the past two weeks?

Run `/usage --json` and divide cache_creation_input_tokens by the sum of cache_creation + cache_read. Above 0.40 means you're paying full cost on prompt content that should be cached.

Question 2 of 6: Token consumption shift

Has your per-turn input_tokens increased by more than 30% on identical tasks since mid-April 2026?

Compare your `/cost` output for a routine task you ran in early April with the same task today. The Opus 4.7 tokenizer change can inflate consumption 1.3x to 1.46x without any change in your workflow.

Question 3 of 6: Plan policy stability

Have you been blocked from upgrading your plan, hit an unexpected weekly reset, or had a feature silently change in the past two weeks?

Examples: payment system declined repeatedly without bank-side issue, weekly reset shifted by hours, autoCompactEnabled false but compaction still ran. These are operator-side reports of plan policy instability.

Question 4 of 6: Workflow heterogeneity

Does at least 70% of your daily work consist of routine tasks (file reads, git operations, scaffolding, formatting) where a smaller model would produce identical output?

A community-shared breakdown from May 2026: 40% file reads and context, 25% test generation, 20% formatting, 15% hard reasoning. If your workflow looks similar, hard reasoning needs Opus but the other 85% does not.

Question 5 of 6: Operational resilience

Do you currently have hooks that block destructive operations and warn on cost spikes before they exceed your daily budget?

Examples: a Stop hook that warns when daily cost exceeds your threshold, a PreToolUse hook that blocks `rm -rf` on non-safe paths. Without these, an overnight loop can consume your entire monthly budget in hours.

Question 6 of 6: Trust in the relationship

In the past two weeks, has the model ignored an explicit stop signal or continued past your halt instruction with bargaining language ("just let me do this one part")?

Reported in three independent issues in early May 2026: continuation bargaining after halt signals is a separate alignment property from technical regressions. If you've experienced this, hooks alone may not be enough.

Read the full Migration Playbook ($19, free silent-refresh sweep on May 8 for buyers)