Making a Sandbox

My thoughts:

The introduction: How railroady is too railroady?

a. Starter dungeon - In classic Moldvay/Mentzer D&D the expectation was that the PCs would start at the 
entrance to the starter dungeon, but with a motivation for exploring it. But once they got 
back to town the sandbox would open up.
b. Rumours. Those rumour tables in Judges Guild stuff and old TSR modules were there 
for a reason. Sandbox PCs should never be lacking in rumours of things to investigate. 
This also explains why old school PCs hang out at inns - that's where the rumours are. Be sure to always have at least 2-3 rumours ready for the PCs to investigate, if they choose. Most rumours 
should relate to interesting adventure sites. Some can be time-critical 'missions' but those 
are un-sandboxy; the PCs will always feel they need to rescue that princess before she gets
sacrificed to Orcus. 
c. Overarching mission - sandbox PCs don't have to be footloose wanderers; you can have the PCs start as members of an organisation such as the Texas Rangers or Starfleet, exploring/patrolling/peacekeeping. Arthurian questing knights are ok. But they need a lot of freedom of action in where to go and what to do.

Player vs monster level: This is much less of a problem in 5E than in 4E, but how do you plan appropriately balanced encounters in a sandbox when you don't know when or how players will tackle them? 

"Environment" Design - you create a range of environments suited to a range of PC levels. 
Traditionally this is a megadungeon with stacked threat levels, the deeper you go the 
bigger the challenge and reward. This lets players choose their preferred threat level. 
It can also be done in wilderness sandboxing, with different terrains & areas having 
different threat levels. It should be widely known that the Mountains of Doom are more 
dangerous than the Foothills of Mild Adversity. Again this lets the players choose their own threat level. The Ars Ludi West Marches posts discuss this well. Typically more dangerous areas are further away from starter town, but you can have the Barrow Mounds right on the edge of the Shire, too.

The conclusion: When is a sandbox adventure over? My players are a big fan of free agency, but they are equally big fans of story, including the climax, the denouement, and the eventual
end. 

a. Traditionally the sandbox dungeon has the dragon or mad wizard or lich in a chamber at 
the very bottom, defeat the ultimate foe to 'win' the dungeon. Full sandbox campaigns do 
not have endings, they are more like episodic TV serials or picaresques than movies. 
But many stories should emerge in the course of play, and these will have beginnings and 
endings.
b. Have lots of dynamic NPCs in your sandbox with their own motivations. Have them plot 
and act independently of the PCs. Some of them will be antagonists. From this antagonism 
great stories will naturally emerge - bottom-up, not top-down. The 2e Villains book is
very good for creating interesting antagonists - http://rpg.drivethrustuff.com/produc...llains-2e?it=1

Player paralysis: With no big sign saying "go here, do this" my players are apt to scratch their heads and say, "I dunnow." 

Rumours again - but not too many! Unless the players are actively seeking more rumours, 
three is plenty. That gives choices without overwhelming them. However if they 
absolutely refuse to decide what to do then you may need to have the old man offer 
them gold to do X. This should work unless they're actively Turtling, in which case sandbox
won't work.

Setting material: In no event has a player ever read any material I have ever written for an adventure. So is it a waste of my time? 

You want enough material prepared in advance for the first 2-3 sessions of play. 
You typically want a wilderness map, a starter town, some NPCs, and several adventure 
locales on that map - detailed starter locales and probably some undetailed higher level ones 
to explore later. A good approach is to put several short published site-based adventures on a 
wilderness map of your own creation, with the 1st level ones nearest to the starter town, higher 
level further out. This starter map can be 2-6 miles per hex, at higher level you can switch to a larger scale or add adjacent area maps with more challenging stuff.
For a megadungeon you might sketch the general structure, and a map of 25-30 
rooms on the first 1 or 2 levels for your first session. I realise now that megadungeons should NOT be fully detailed ahead of play!
You need enough prepped that players can make choices in play - unless you do West Marches 
where the players have to make their choices what to do ahead of the session & the GM preps in 
response to that.
But don't create everything up front, create in response to player activity & interest. 
You need eg some gods for Clerics, so eg choose a pantheon or three, but you don't need 
a detailed calendar or historical timeline unless you want it. In fact too much detail can
stifle creativity in play; lightly-sketched histories tend to work best.

Read more: http://www.enworld.org/forum/showthread.php?426911-What-separates-a-sandbox-adventure-from-an-AP/page4#ixzz3TsAkGR00