The Sample Size Calculator

Help me figure out how big a sample I need for this.

What I'm measuring: [PROPORTION / MEAN / DIFFERENCE BETWEEN GROUPS]
Baseline/current value: [IF KNOWN]
The smallest effect worth detecting: [DESCRIBE]
Confidence and power I want: [E.G. 95% CONFIDENCE, 80% POWER]

Provide:
1. The sample size needed, with the calculation and inputs shown
2. How the number changes if I relax the effect size or confidence
3. What's realistic given my traffic/population, and the trade-off if I
   can't hit it
4. The run time implied at my volume
5. The consequence of running underpowered (and why "no significant
   difference" won't mean "no difference")

Give me a number I can plan around and the assumptions behind it.