Problem
CodeWhale has whale identity, DeepSeek V4 Pro/Flash model routing, and thinking-effort controls, but those concepts are not unified into a user-readable ladder. Users should be able to understand depth/cost/speed at a glance without decoding raw model ids plus max/high/none.
The desired mental model is:
deepseek-v4-pro + max thinking: largest whale, deepest reasoning.deepseek-v4-pro + high thinking.deepseek-v4-pro + no thinking.deepseek-v4-flash + max thinking.deepseek-v4-flash + high thinking.deepseek-v4-flash + no thinking: smallest/fastest whale.
Largest whales appear at the top; smaller/faster whales appear lower. This is a naming and comprehension layer, not a replacement for exact model ids.
Desired Behavior
Create a whale-size route taxonomy for model/effort choices.
Requirements:
- Keep exact model id and reasoning effort visible in details.
- Add friendly whale-size labels for the route picker, footer/status, agent cards, and route receipts.
- Sort routes from largest/deepest to smallest/fastest.
- Use whale species/size metaphors consistently and tastefully.
- Do not use provider logos or imply any outside endorsement.
- Keep Whale Dark/Whale Light as the product identity; route names are a helpful layer, not a brand replacement.
Example shape, names to verify during implementation:
Blue Whale Pro + max
Fin Whale Pro + high
Sperm Whale Pro + no thinking
Humpback Whale Flash + max
Minke Whale Flash + high
Porpoise/Calf Flash + no thinking
The exact species list should be source-checked and adjusted for clarity. If dolphins/porpoises are excluded from the user-facing whale pool in #2016, use a small whale species instead for the final tier.
Acceptance Criteria
- A central route taxonomy maps
(provider, model, reasoning_effort) to display label, sort order, model id, effort, and descriptive hint.
- The picker displays routes in size/depth order: Pro max -> Pro high -> Pro none -> Flash max -> Flash high -> Flash none.
- Agent/scout route receipts show both friendly whale label and exact model/effort.
/model auto or Dual/scout routing can explain "why this whale was chosen" in user terms.- Unknown models fall back to exact model id without fake whale labeling.
- Tests cover ordering, label lookup, fallback, and no loss of exact model/effort data.
Related
Problem
CodeWhale has whale identity, DeepSeek V4 Pro/Flash model routing, and thinking-effort controls, but those concepts are not unified into a user-readable ladder. Users should be able to understand depth/cost/speed at a glance without decoding raw model ids plus
max/high/none.The desired mental model is:
deepseek-v4-pro+maxthinking: largest whale, deepest reasoning.deepseek-v4-pro+highthinking.deepseek-v4-pro+ no thinking.deepseek-v4-flash+maxthinking.deepseek-v4-flash+highthinking.deepseek-v4-flash+ no thinking: smallest/fastest whale.Largest whales appear at the top; smaller/faster whales appear lower. This is a naming and comprehension layer, not a replacement for exact model ids.
Desired Behavior
Create a whale-size route taxonomy for model/effort choices.
Requirements:
Example shape, names to verify during implementation:
The exact species list should be source-checked and adjusted for clarity. If dolphins/porpoises are excluded from the user-facing whale pool in #2016, use a small whale species instead for the final tier.
Acceptance Criteria
(provider, model, reasoning_effort)to display label, sort order, model id, effort, and descriptive hint./model autoor Dual/scout routing can explain "why this whale was chosen" in user terms.Related