top of page

How-to: A/B Your Lean With a Fixed Seed

7 days ago
1 min read

On Edge, a seed makes the result repeatable for a given configuration. That turns tuning from a guess into a fair test. Background on the matrix: part 01

1. Pick one real goal. Write it once and do not touch the wording between runs.

2. Set a seed. Any integer works. Keep it the same for both runs.

3. Run A in the default band. For most technical and strategic work the guide starts bias at 0.78–0.85. Edge defaults to 0.82.

4. Run B with only bias changed. Go higher if you want a more decisive, committed brief, or lower if you want broader exploration of trade-offs.

5. Hold everything else still: n, m, steps, grid_size, goal and extra context.

6. Read both briefs side by side. Ask which one needs less human repair before your host can use it.

Change one knob per test. If you change two, you will not know which one did the work. Once a lean band works for a type of goal, write it down and reuse it.

Soft door: run your two seeded Edge calls from Members

 
 
 

Recent Posts

See All
Mini-Checklist: Run Your Own Before/After Test

No win-rate charts here. The real benchmark is your own goal on your own host: one before run, one after run with an Edge brief, scored for coherence, specificity and repair work.

 
 
 
One Idea: How Cloud Picks a Champion

Edge runs one matrix. Cloud runs a small crowd of cubits, lets each evolve on its own, and hands back only the global champion: still one brief, not a list.

 
 
 
Myth vs Fact: Turn Every Knob Up

More is not always better. Bias, n, steps and cubits are controls, not quality dials. Five myths about tuning Infinity Blip, and what actually helps.

 
 
 

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page