Build your model-picker card.
One row per recurring task type from yesterday's triage list. Three levers — need, size, speed — land on a model class. Export the CSV before you close.
Goal: 6 rows minimum, stop at 8. One row per recurring task TYPE from yesterday's
triage list. The Why column is the part future-you re-reads when the model lineup changes.
0 / 6
Rows started
0
Fully filled
⚡ 0
Fast / small
⚖️ 0
Balanced
🧠 0
Strong / large
| Task type | Need | Doc size | Speed | Model class | Model name | Why (one line) | |
|---|---|---|---|---|---|---|---|
The three levers — what each one asks
- Need — drafting (you'll edit), extract/classify, or multi-step reasoning?
- Doc size — use the token rule. Tiny <1K · Small 1–10K · Medium 10–50K · Large 50K+.
- Speed — instant (customer-facing), minutes ok, or no rush?
Three model classes
- Fast / small — cheap, low-latency, shallower reasoning. Drafting, classification, instant flows.
- Balanced — mid-cost, good reasoning, decent window. Default for unclear cases.
- Strong / large-context — expensive, deep reasoning, big window. Decision-grade work on long docs.
Stretch: Mark which two rows would change model class if the document doubled in size.
Note the swap in the Why column. That's the test that the size lever is doing real work.