Cascade Comparator
Two token cascades, side by side. See which one compounds and which one burns — across yield, compression, cache hit rate, and leverage.
| Metric | A | B | Edge |
|---|---|---|---|
| Υ Yield | 2.50 | 0.02 | A |
| Compression ratio | 0.38 | 0.07 | A |
| Cache hit rate | 84% | 67% | A |
| Leverage | 6.7x | 0.3x | A |
Comparing raw cascades is a quick diagnostic. For the full operator-vs-operator experience (radars, history, signed snapshots), see /compare.
Reading the comparison
Each row flags the cascade with the edge. Υ Yield is the headline — it folds cache reuse and output density into one number, so a cascade can win on raw output yet lose on yield if it re-sends fresh context every turn. Cache hit rate and leverage reveal whether the edge comes from reusing context or from denser output. Read them together: a high-yield cascade with low leverage is output-dense but cache-poor; a high-yield cascade with high leverage is the compounding ideal.
Frequently asked questions
What does the Cascade Comparator do?
It takes two sets of four token pillars (input, output, cache-read, cache-write) — call them A and B — and computes yield, compression ratio, cache hit rate, and leverage for each, flagging which cascade has the edge on every metric.
How is this different from the /compare page?
The /compare page compares two ranked operators from the leaderboard (radars, history, signed snapshots). This comparator is a quick raw-cascade diagnostic: paste any two sets of token counts and see the metric deltas instantly, no account or leaderboard entry required.
Which metrics matter most when comparing cascades?
Υ Yield is the headline — it captures cache reuse and output density in one number. Cache hit rate shows how well context is reused, and leverage shows how much cached context amplifies fresh input. Compression ratio rounds out the picture of output-per-input.
Can I compare cascades from different AI platforms?
Yes. The four token pillars are platform-neutral — Claude, ChatGPT, Gemini, Copilot, Cursor and 15+ platforms all expose equivalent counts. Comparing cascades across platforms is exactly how cross-platform operator efficiency is measured.